korrents

Anthropic

What people who work or worked at Anthropic have said about where AI is going, how to make it safe, and how to build with it.

17 people · 287 claims

Views

What their claims argue, grouped. The labels are ours; every quote is one person's own words, not the company's position.

Powerful AI is coming within years, not decades 12 claims

There is a 90% chance of a country of geniuses in a datacentre within ten years, and no honest forecast can go much higher than that.

  1. Dario Amodei Co-founder and CEO of Anthropic on the basic hypothesis of you know, as you put it, within 10 years we'll get to, you know, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that. Um and it's hard to go much higher than 90% cuz the world is so unpredictable. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 0:13:51 into the videoAll korrents from this video
    Their wordson the basic hypothesis of you know, as you put it, within 10 years we'll get to, you know, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that. Um and it's hard to go much higher than 90% cuz the world is so unpredictable.

    Where this was said

    Watch from 0:13:51 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

We will have a country of geniuses in a datacentre within ten years.

  1. Dario Amodei Co-founder and CEO of Anthropic on the basic hypothesis of you know, as you put it, within 10 years we'll get to, you know, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that. Um and it's hard to go much higher than 90% cuz the world is so unpredictable. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 0:13:51 into the videoAll korrents from this video
    Their wordson the basic hypothesis of you know, as you put it, within 10 years we'll get to, you know, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that. Um and it's hard to go much higher than 90% cuz the world is so unpredictable.

    Where this was said

    Watch from 0:13:51 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

The most surprising thing about the last three years is not the technology but how few people recognise we are near the end of the exponential.

Mental modelExponential growth

  1. Dario Amodei Co-founder and CEO of Anthropic I will tell you though what the most surprising thing has been. The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 0:00:48 into the videoAll korrents from this video
    Their wordsI will tell you though what the most surprising thing has been. The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential.

    Where this was said

    Watch from 0:00:48 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

Recursive self-improvement has already started, because AI research itself is now being automated.

  1. Jan Leike Alignment researcher We are starting to automate AI research and the recursive self-improvement process has begun. Alignment is not solvedaligned.substack.com · 22 Jan 2026All korrents from this piece
    Their wordsWe are starting to automate AI research and the recursive self-improvement process has begun.

    ↗Alignment is not solvedaligned.substack.com

    22 Jan 2026

Human parity on discovery benchmarks like DiG-bench will be reached by mid-2027, marking the point where recursive self-improvement in AI seriously kicks off.

  1. Jack Clark Co-founder and head of policy at Anthropic My guess is we'll reach human parity on DiG-bench by middle of 2027, at which point we should expect things like recursive self-improvement to seriously kick off. Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net · 17 Aug 2026All korrents from this piece
    Their wordsMy guess is we'll reach human parity on DiG-bench by middle of 2027, at which point we should expect things like recursive self-improvement to seriously kick off.

    ↗Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net

    17 Aug 2026

As AI systems get better at scientific research tasks, it becomes more likely they will soon be capable of improving and building themselves.

  1. Jack Clark Co-founder and head of policy at Anthropic The better AI systems get at this, the more likelihood we can assign to the idea that AI systems will imminently become capable of building themselves. Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net · 17 Aug 2026All korrents from this piece
    Their wordsThe better AI systems get at this, the more likelihood we can assign to the idea that AI systems will imminently become capable of building themselves.

    ↗Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net

    17 Aug 2026

The history of machine learning is of proclaimed barriers dissolving inside the big blob of compute.

  1. Dario Amodei Co-founder and CEO of Anthropic In fact, I would point to the history in in ML of people coming up with things that are barriers that end up kind of dissolving within the big blob of compute, right? Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 1:20:48 into the videoAll korrents from this video
    Their wordsIn fact, I would point to the history in in ML of people coming up with things that are barriers that end up kind of dissolving within the big blob of compute, right?

    Where this was said

    Watch from 1:20:48 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

Today's AI systems are significantly under-elicited for their ability to automate AI R&D.

  1. Jack Clark Co-founder and head of policy at Anthropic Posts like this highlight how we are under-eliciting today's AI systems for their ability to automate AI R&D - especially striking is how the company can jump the performance of Opus 5 by 10 absolute percentage points with a better harness. Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net · 10 Aug 2026All korrents from this piece
    Their wordsPosts like this highlight how we are under-eliciting today's AI systems for their ability to automate AI R&D - especially striking is how the company can jump the performance of Opus 5 by 10 absolute percentage points with a better harness.

    ↗Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net

    10 Aug 2026

AI progress comes from a simple, systematic way of improving models, not from researchers suddenly getting smarter.

  1. Jared Kaplan Co-founder of Anthropic It's not that AI researchers are really smart or they suddenly got smart. It's that we found a very, very simple way of making AI better systematically, and we're turning that crank. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsIt's not that AI researchers are really smart or they suddenly got smart. It's that we found a very, very simple way of making AI better systematically, and we're turning that crank.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

Scaling results gave conviction that AI would keep getting smarter in a very predictable way.

Mental modelExponential growth

  1. Jared Kaplan Co-founder of Anthropic These gave us a lot of conviction to believe that AI was just going to keep getting smarter and smarter in a very predictable way. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsThese gave us a lot of conviction to believe that AI was just going to keep getting smarter and smarter in a very predictable way.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

Eventually AI models, perhaps millions working together, may be able to do the work of whole human organizations.

  1. Jared Kaplan Co-founder of Anthropic Eventually, we imagine AI models—or millions of AI models perhaps working together—will be able to do the work that whole human organizations can do. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsEventually, we imagine AI models—or millions of AI models perhaps working together—will be able to do the work that whole human organizations can do.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

Safety is the reason the company exists, and it conflicts with business less than expected 3 claims

Anthropic was started to run towards something, an organisation with safety and responsibility at the forefront, not away from something.

  1. Daniela Amodei Co-founder and President of Anthropic, as of 2026 we’re running towards something, versus running away from something. We had this vision in our heads of wanting to create an organization where the values that matter to us around safety, around responsibility, were the kind of forefront of what we were doing. Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu · 29 Jun 2026All korrents from this piece
    Their wordswe’re running towards something, versus running away from something. We had this vision in our heads of wanting to create an organization where the values that matter to us around safety, around responsibility, were the kind of forefront of what we were doing.

    ↗Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu

    29 Jun 2026

The highest-level framing for Anthropic is taking radical responsibility for the technology it develops.

  1. Daniela Amodei Co-founder and President of Anthropic, as of 2026 I think to us, the kind of highest level of framing, is just taking a form of radical responsibility for the technology that we’re developing. Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu · 29 Jun 2026All korrents from this piece
    Their wordsI think to us, the kind of highest level of framing, is just taking a form of radical responsibility for the technology that we’re developing.

    ↗Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu

    29 Jun 2026

Safety and commercial success conflict less than you would expect.

  1. Daniela Amodei Co-founder and President of Anthropic, as of 2026 I think in general, the two don’t come into conflict as much as you would expect. Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu · 29 Jun 2026All korrents from this piece
    Their wordsI think in general, the two don’t come into conflict as much as you would expect.

    ↗Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu

    29 Jun 2026

Aligning today's models looks solvable; aligning ones we cannot understand does not yet 10 claims

Alignment of today's models is going well enough to look solvable, while aligning models we can no longer understand remains unsolved.

  1. Jan Leike Alignment researcher This is the hard problem of alignment we need to solve in order to succeed at building superintelligence, and to this day it is an unsolved problem. Alignment is not solvedaligned.substack.com · 22 Jan 2026All korrents from this piece
    Their wordsThis is the hard problem of alignment we need to solve in order to succeed at building superintelligence, and to this day it is an unsolved problem.

    ↗Alignment is not solvedaligned.substack.com

    22 Jan 2026

Simple training interventions turned out to be very effective at steering models towards aligned behaviour.

  1. Jan Leike Alignment researcher But the most important lesson is that simple interventions are very effective at steering the model towards more aligned behavior. Alignment is not solvedaligned.substack.com · 22 Jan 2026All korrents from this piece
    Their wordsBut the most important lesson is that simple interventions are very effective at steering the model towards more aligned behavior.

    ↗Alignment is not solvedaligned.substack.com

    22 Jan 2026

Training a model on principles rather than a list of rules is not an ideology but an empirical finding: its behaviour is more consistent and covers edge cases.

  1. Dario Amodei Co-founder and CEO of Anthropic it's it's kind of purely a practical and empirical thing that we've observed that by teaching the model principles, getting it to learn from principles, its behavior is more consistent, it's easier to cover edge cases, and the model is more likely to do what people want it to do. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 2:06:41 into the videoAll korrents from this video
    Their wordsit's it's kind of purely a practical and empirical thing that we've observed that by teaching the model principles, getting it to learn from principles, its behavior is more consistent, it's easier to cover edge cases, and the model is more likely to do what people want it to do.

    Where this was said

    Watch from 2:06:41 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

We do not have to align superintelligence directly; we have to build a human-level automated alignment researcher we trust more than ourselves.

  1. Jan Leike Alignment researcher But the goal we need to achieve is so much easier: we just need to build a model that’s as good as us at alignment research, and that we trust more than ourselves to do this research well because it’s sufficiently aligned. Alignment is not solvedaligned.substack.com · 22 Jan 2026All korrents from this piece
    Their wordsBut the goal we need to achieve is so much easier: we just need to build a model that’s as good as us at alignment research, and that we trust more than ourselves to do this research well because it’s sufficiently aligned.

    ↗Alignment is not solvedaligned.substack.com

    22 Jan 2026

Making a capable model malicious is about as easy as making it aligned, because the same levers work in both directions.

  1. Jan Leike Alignment researcher In fact, making an evil version of Claude that’s just as smart and agentic would be pretty easy. Alignment is not solvedaligned.substack.com · 22 Jan 2026All korrents from this piece
    Their wordsIn fact, making an evil version of Claude that’s just as smart and agentic would be pretty easy.

    ↗Alignment is not solvedaligned.substack.com

    22 Jan 2026

Building AI that can actually be trusted is the goal; containing an AI known to be misaligned is not a substitute for it.

  1. Jan Leike Alignment researcher More generally, we should actually solve alignment instead of just trying to control misaligned AI. Should we control AI instead of aligning it?aligned.substack.com · 24 Jan 2025All korrents from this piece
    Their wordsMore generally, we should actually solve alignment instead of just trying to control misaligned AI.

    ↗Should we control AI instead of aligning it?aligned.substack.com

    24 Jan 2025

Human oversight of AI behaviour is already intractable at the volume models produce, so the overseers have to be models too.

  1. Jan Leike Alignment researcher Therefore it’s simply intractable to meaningfully oversee all but a tiny fraction of our models’ behavior with actual humans. Two alignment threat modelsaligned.substack.com · 8 Nov 2024All korrents from this piece
    Their wordsTherefore it’s simply intractable to meaningfully oversee all but a tiny fraction of our models’ behavior with actual humans.

    ↗Two alignment threat modelsaligned.substack.com

    8 Nov 2024

AI agents pursuing a task can develop emergent, misaligned behavior that humans are not prepared for.

Mental modelEmergence

  1. Jack Clark Co-founder and head of policy at Anthropic This is what the future is going to look like and we are not prepared for it. Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net · 10 Aug 2026All korrents from this piece
    Their wordsThis is what the future is going to look like and we are not prepared for it.

    ↗Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net

    10 Aug 2026

The longer an AI system operates autonomously and the more actions it takes, the harder it becomes to distinguish benign behavior from malicious or subversive behavior.

  1. Jack Clark Co-founder and head of policy at Anthropic The longer the time an AI system can operate for and the more actions it takes, the harder it gets to discern benign and helpful behaviors from malicious or subversive ones. Import AI 466: The bitter lesson for robotics, AIs complete week-long programming tasks; and OpenAI’s accidental AI hackerjack-clark.net · 27 Jul 2026All korrents from this piece
    Their wordsThe longer the time an AI system can operate for and the more actions it takes, the harder it gets to discern benign and helpful behaviors from malicious or subversive ones.

    ↗Import AI 466: The bitter lesson for robotics, AIs complete week-long programming tasks; and OpenAI’s accidental AI hackerjack-clark.net

    27 Jul 2026

Large language models are not aligned with their users.

  1. Jan Leike Alignment researcher For example, large language models can generate outputs that are untruthful, toxic, or simply not helpful to the user. In other words, these models are not aligned with their users. Training language models to follow instructions with human feedback (with 19 co-authors)arxiv.org · 4 Mar 2022All korrents from this piece
    Their wordsFor example, large language models can generate outputs that are untruthful, toxic, or simply not helpful to the user. In other words, these models are not aligned with their users.

    ↗Training language models to follow instructions with human feedback (with 19 co-authors)arxiv.org

    4 Mar 2022

We should understand what is inside the models we deploy 6 claims

It is unacceptable to deploy AI systems central to the economy and national security while remaining ignorant of how they work.

  1. Dario Amodei Co-founder and CEO of Anthropic These systems will be absolutely central to the economy, technology, and national security, and will be capable of so much autonomy that I consider it basically unacceptable for humanity to be totally ignorant of how they work. The Urgency of Interpretability (April 2025)darioamodei.com · 1 Apr 2025
    Their wordsThese systems will be absolutely central to the economy, technology, and national security, and will be capable of so much autonomy that I consider it basically unacceptable for humanity to be totally ignorant of how they work.

    ↗The Urgency of Interpretability (April 2025)darioamodei.com

    1 Apr 2025

Anthropic now uses mechanistic interpretability techniques to audit new models for deception and scheming before releasing them.

  1. Dario Amodei Co-founder and CEO of Anthropic Even more recently, we’ve begun to use mechanistic interpretability techniques to improve our safeguards and to conduct “audits” of new models before we release them, looking for evidence of deception, scheming, power-seeking, or a propensity to behave differently when being evaluated. The Adolescence of Technology (January 2026)darioamodei.com · 1 Jan 2026
    Their wordsEven more recently, we’ve begun to use mechanistic interpretability techniques to improve our safeguards and to conduct “audits” of new models before we release them, looking for evidence of deception, scheming, power-seeking, or a propensity to behave differently when being evaluated.

    ↗The Adolescence of Technology (January 2026)darioamodei.com

    1 Jan 2026

Prompt injection can be caught by watching the neurons that fire when it happens, so a defence no longer depends on the model reporting the attack.

  1. Boris Cherny Creator of Claude Code at Anthropic it's it's literally we're looking at neurons in the model's brain that light up when prompt injection happens. So the model won't even tell you but we can actually see those neurons and we can figure out and diagnose that it's happening. And then you combine that with the auto mode classifier and with these three layers we just cannot demonstrate prompt injection anymore. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 3:01 into the videoAll korrents from this video
    Their wordsit's it's literally we're looking at neurons in the model's brain that light up when prompt injection happens. So the model won't even tell you but we can actually see those neurons and we can figure out and diagnose that it's happening. And then you combine that with the auto mode classifier and with these three layers we just cannot demonstrate prompt injection anymore.

    Where this was said

    Watch from 3:01 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Deep learning will be understood very differently in thirty years than it is today.

  1. Christopher Olah Interpretability researcher at Anthropic I think it is very likely that, reflecting back in 30 years, we will see deep learning very differently. Neural Networks, Types, and Functional Programmingcolah.github.io · 3 Sept 2015All korrents from this piece
    Their wordsI think it is very likely that, reflecting back in 30 years, we will see deep learning very differently.

    ↗Neural Networks, Types, and Functional Programmingcolah.github.io

    3 Sept 2015

Understanding a machine learning model and understanding the data it operates on are inseparable tasks.

  1. Christopher Olah Interpretability researcher at Anthropic Understanding data and understanding models that work on that data are intimately linked. Visualizing Representations: Deep Learning and Human Beingscolah.github.io · 16 Jan 2015All korrents from this piece
    Their wordsUnderstanding data and understanding models that work on that data are intimately linked.

    ↗Visualizing Representations: Deep Learning and Human Beingscolah.github.io

    16 Jan 2015

Complete reverse-engineering of neural networks is a less promising research direction than studying model biology and applying interpretability usefully

  1. Neel Nanda Runs the mechanistic interpretability team at Google DeepMind I'm now fairly pessimistic about ambitious interpretability (i.e. complete reverse-engineering), and I'm excited about model biology (studying qualitative high-level properties of models) and applied interpretability (rigorously doing useful things with interp). MATS Applications Open (Due Aug 29)neelnanda.io · 19 Aug 2025All korrents from this piece
    Their wordsI'm now fairly pessimistic about ambitious interpretability (i.e. complete reverse-engineering), and I'm excited about model biology (studying qualitative high-level properties of models) and applied interpretability (rigorously doing useful things with interp).

    ↗MATS Applications Open (Due Aug 29)neelnanda.io

    19 Aug 2025

Finding a harm in an AI system calls for weighing it, not for an automatic stop 2 claims

Identifying a harm in an AI system does not show that the system should not be deployed.

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment So it's a mistake to assume that if we identify a pro tanto harm from an AI system, it must be the case that someone has done something wrong, something needs to be done to correct it, or the system shouldn't be deployed. In AI ethics, "bad" isn't good enough (Amanda Askell's Blog)askell.blog · 14 Dec 2020All korrents from this piece
    Their wordsSo it's a mistake to assume that if we identify a pro tanto harm from an AI system, it must be the case that someone has done something wrong, something needs to be done to correct it, or the system shouldn't be deployed.

    ↗In AI ethics, "bad" isn't good enough (Amanda Askell's Blog)askell.blog

    14 Dec 2020

All-things-considered judgments in AI ethics will often require a complex evaluation of many factors.

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment Since we can't be certain about any one moral theory and since we have to try to represent a plurality of views, coming to all things considered judgments in AI ethics will often require a fairly complex evaluation of many relevant factors. In AI ethics, "bad" isn't good enough (Amanda Askell's Blog)askell.blog · 14 Dec 2020All korrents from this piece
    Their wordsSince we can't be certain about any one moral theory and since we have to try to represent a plurality of views, coming to all things considered judgments in AI ethics will often require a fairly complex evaluation of many relevant factors.

    ↗In AI ethics, "bad" isn't good enough (Amanda Askell's Blog)askell.blog

    14 Dec 2020

AI will hit jobs hard and fast, and its gains will not spread by themselves 12 claims

AI could displace half of all entry-level white-collar jobs within one to five years, even while it accelerates economic growth.

  1. Dario Amodei Co-founder and CEO of Anthropic This is a topic that I warned about very publicly in 2025 , where I predicted that AI could displace half of all entry-level white collar jobs in the next 1–5 years, even as it accelerates economic growth and scientific progress. The Adolescence of Technology (January 2026)darioamodei.com · 1 Jan 2026
    Their wordsThis is a topic that I warned about very publicly in 2025 , where I predicted that AI could displace half of all entry-level white collar jobs in the next 1–5 years, even as it accelerates economic growth and scientific progress.

    ↗The Adolescence of Technology (January 2026)darioamodei.com

    1 Jan 2026

Labour markets will eventually recover from AI, but the speed of the transition makes the short term unusually painful compared with past technologies.

  1. Dario Amodei Co-founder and CEO of Anthropic To be clear, speed in itself does not mean labor markets and employment won’t eventually recover, it just implies the short-term transition will be unusually painful compared to past technologies, since humans and labor markets are slow to react and to equilibrate. The Adolescence of Technology (January 2026)darioamodei.com · 1 Jan 2026
    Their wordsTo be clear, speed in itself does not mean labor markets and employment won’t eventually recover, it just implies the short-term transition will be unusually painful compared to past technologies, since humans and labor markets are slow to react and to equilibrate.

    ↗The Adolescence of Technology (January 2026)darioamodei.com

    1 Jan 2026

Because AI matches the general cognitive profile of humans, it will also be good at the new jobs that would ordinarily replace the ones it automates.

  1. Dario Amodei Co-founder and CEO of Anthropic By contrast, AI is increasingly matching the general cognitive profile of humans, which means it will also be good at the new jobs that would ordinarily be created in response to the old ones being automated. The Adolescence of Technology (January 2026)darioamodei.com · 1 Jan 2026
    Their wordsBy contrast, AI is increasingly matching the general cognitive profile of humans, which means it will also be good at the new jobs that would ordinarily be created in response to the old ones being automated.

    ↗The Adolescence of Technology (January 2026)darioamodei.com

    1 Jan 2026

Enduring job displacement is undesirable and dangerous, and should be minimised rather than accepted as a goal.

  1. Dario Amodei Co-founder and CEO of Anthropic First, enduring job displacement is undesirable and dangerous, and we should do everything we can to minimize or prevent it, not to bring it about. Policy on the AI Exponential (June 2026)darioamodei.com · 1 Jun 2026
    Their wordsFirst, enduring job displacement is undesirable and dangerous, and we should do everything we can to minimize or prevent it, not to bring it about.

    ↗Policy on the AI Exponential (June 2026)darioamodei.com

    1 Jun 2026

If AI permanently drives down demand for labour, long-term income support for a significant fraction of the workforce will likely become necessary.

  1. Dario Amodei Co-founder and CEO of Anthropic If AI-driven labor displacement ends up being large in magnitude and permanently drives down the demand for labor, it will likely be necessary to go beyond mere incentive programs to long-term income support for a significant fraction of the labor force. Policy on the AI Exponential (June 2026)darioamodei.com · 1 Jun 2026
    Their wordsIf AI-driven labor displacement ends up being large in magnitude and permanently drives down the demand for labor, it will likely be necessary to go beyond mere incentive programs to long-term income support for a significant fraction of the labor force.

    ↗Policy on the AI Exponential (June 2026)darioamodei.com

    1 Jun 2026

Universal basic income is at most a small part of the answer to AI-driven job loss.

  1. Dario Amodei Co-founder and CEO of Anthropic It could be as simple as a large universal basic income for everyone, although I suspect that will only be a small part of a solution. Machines of Loving Grace (October 2024)darioamodei.com · 1 Oct 2024
    Their wordsIt could be as simple as a large universal basic income for everyone, although I suspect that will only be a small part of a solution.

    ↗Machines of Loving Grace (October 2024)darioamodei.com

    1 Oct 2024

The real danger in AI-driven growth is geographic: 50% growth in and around Silicon Valley and business as usual everywhere else.

  1. Dario Amodei Co-founder and CEO of Anthropic a worry I have is that the growth rate could be like 50% in Silicon Valley and, you know, parts of the world that are kind of socially connected to Silicon Valley and, you know, not that much faster than its current pace elsewhere. And I think that'd be a pretty messed up world. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 1:17:35 into the videoAll korrents from this video
    Their wordsa worry I have is that the growth rate could be like 50% in Silicon Valley and, you know, parts of the world that are kind of socially connected to Silicon Valley and, you know, not that much faster than its current pace elsewhere. And I think that'd be a pretty messed up world.

    Where this was said

    Watch from 1:17:35 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

In an AI world growth and economic value come easily; distribution, wealth and political freedom are the hard parts, and that is where policy belongs.

  1. Dario Amodei Co-founder and CEO of Anthropic What I would say is that you know we are we are about to be in a world where growth and economic value will come very easily. If right if we're able to build these powerful AI models, growth and economic value will come very easily. What will not come easily is distribution of benefits, distribution of wealth, political freedom, um you know, these are the things that are going to be hard to achieve. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 2:03:18 into the videoAll korrents from this video
    Their wordsWhat I would say is that you know we are we are about to be in a world where growth and economic value will come very easily. If right if we're able to build these powerful AI models, growth and economic value will come very easily. What will not come easily is distribution of benefits, distribution of wealth, political freedom, um you know, these are the things that are going to be hard to achieve.

    Where this was said

    Watch from 2:03:18 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

AI will structurally advance human health and reduce poverty, but there is no strong reason to expect it to structurally advance democracy or peace.

  1. Dario Amodei Co-founder and CEO of Anthropic Unfortunately, I see no strong reason to believe AI will preferentially or structurally advance democracy and peace, in the same way that I think it will structurally advance human health and alleviate poverty. Machines of Loving Grace (October 2024)darioamodei.com · 1 Oct 2024
    Their wordsUnfortunately, I see no strong reason to believe AI will preferentially or structurally advance democracy and peace, in the same way that I think it will structurally advance human health and alleviate poverty.

    ↗Machines of Loving Grace (October 2024)darioamodei.com

    1 Oct 2024

Meaning in human life comes mostly from relationships and connection rather than from economic labour.

  1. Dario Amodei Co-founder and CEO of Anthropic In any case I think meaning comes mostly from human relationships and connection, not from economic labor. Machines of Loving Grace (October 2024)darioamodei.com · 1 Oct 2024
    Their wordsIn any case I think meaning comes mostly from human relationships and connection, not from economic labor.

    ↗Machines of Loving Grace (October 2024)darioamodei.com

    1 Oct 2024

AI is going to change which jobs are available and what people do in them.

  1. Daniela Amodei Co-founder and President of Anthropic, as of 2026 AI is going to change what types of jobs are available and what types of jobs people do. Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu · 29 Jun 2026All korrents from this piece
    Their wordsAI is going to change what types of jobs are available and what types of jobs people do.

    ↗Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu

    29 Jun 2026

Humans like to be with other humans, and that very real phenomenon will matter as AI changes work.

  1. Daniela Amodei Co-founder and President of Anthropic, as of 2026 ultimately, there’s this very real phenomenon, which is that humans like to be with other humans. Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu · 29 Jun 2026All korrents from this piece
    Their wordsultimately, there’s this very real phenomenon, which is that humans like to be with other humans.

    ↗Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu

    29 Jun 2026

Powerful AI has to be governed: secured weights, chip controls and a shared pace 10 claims

Governing increasingly powerful AI systems requires solving a collective action problem among companies and governments to coordinate on pacing AI development.

Mental modelCollective action problems

  1. Jack Clark Co-founder and head of policy at Anthropic Many of the challenges implied by increasingly powerful systems that may eventually build themselves run through solving collective action problems among humans - namely, how we can get companies and governments to coordinate in thinking about how to develop this technology and what kinds of mechanisms may be desirable for being able to control the speed at which it develops. Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativityjack-clark.net · 3 Aug 2026All korrents from this piece
    Their wordsMany of the challenges implied by increasingly powerful systems that may eventually build themselves run through solving collective action problems among humans - namely, how we can get companies and governments to coordinate in thinking about how to develop this technology and what kinds of mechanisms may be desirable for being able to control the speed at which it develops.

    ↗Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativityjack-clark.net

    3 Aug 2026

Once model weights leave their owner's control the loss is irreversible, which is why securing them is the precondition for governing AI at all.

  1. Jan Leike Alignment researcher Moreover, model exfiltration is likely impossible to reverse. Self-exfiltration is a key dangerous capabilityaligned.substack.com · 13 Sept 2023All korrents from this piece
    Their wordsMoreover, model exfiltration is likely impossible to reverse.

    ↗Self-exfiltration is a key dangerous capabilityaligned.substack.com

    13 Sept 2023

Releasing open weights is irreversible in a way that deploying a hosted model is not, so its risks cannot be mitigated later.

Mental modelOne-way and two-way doors

  1. Jan Leike Alignment researcher This is especially severe for open weight models, where deployment is essentially irreversible and mitigations can’t be improved later. Two alignment threat modelsaligned.substack.com · 8 Nov 2024All korrents from this piece
    Their wordsThis is especially severe for open weight models, where deployment is essentially irreversible and mitigations can’t be improved later.

    ↗Two alignment threat modelsaligned.substack.com

    8 Nov 2024

The dangerous capability that matters most for misalignment is whether a model can exfiltrate its own weights, because that is the one that removes it from your control.

  1. Jan Leike Alignment researcher If a model was capable of self-exfiltration, it would have the option to remove itself from your control. Self-exfiltration is a key dangerous capabilityaligned.substack.com · 13 Sept 2023All korrents from this piece
    Their wordsIf a model was capable of self-exfiltration, it would have the option to remove itself from your control.

    ↗Self-exfiltration is a key dangerous capabilityaligned.substack.com

    13 Sept 2023

A system capable of superhuman invention would fundamentally and unpredictably alter the world's balance of power, rather than simply serving individual empowerment.

  1. Jack Clark Co-founder and head of policy at Anthropic Rather I am suggesting that it's hard to reconcile a system capable of superhuman invention with something that doesn't fundamentally alter the balance of power in the world in ways that are confusing and hard to reason about. Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net · 17 Aug 2026All korrents from this piece
    Their wordsRather I am suggesting that it's hard to reconcile a system capable of superhuman invention with something that doesn't fundamentally alter the balance of power in the world in ways that are confusing and hard to reason about.

    ↗Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net

    17 Aug 2026

Chip export controls to China keep failing not because the case is weak but because there is too much money riding on the sale.

Mental modelIncentives

  1. Dario Amodei Co-founder and CEO of Anthropic The counterarguments against it are I'll politely call them fishy. Um, uh and yet it doesn't happen and we sell the chips because there's there's so much money. There's so much money riding on it. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 1:45:52 into the videoAll korrents from this video
    Their wordsThe counterarguments against it are I'll politely call them fishy. Um, uh and yet it doesn't happen and we sell the chips because there's there's so much money. There's so much money riding on it.

    Where this was said

    Watch from 1:45:52 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

Banning state AI regulation for ten years with no federal rule to replace it is a crazy thing to do, because ten years is an eternity.

  1. Dario Amodei Co-founder and CEO of Anthropic given the serious dangers that I lay out in adolescence of technology around things like the you know, kind of biological weapons and bioterrorism, autonomy risk and the timelines we've been talking about, like 10 years is an eternity. Like that's that's a that's a I I think that's a crazy thing to do. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 1:38:17 into the videoAll korrents from this video
    Their wordsgiven the serious dangers that I lay out in adolescence of technology around things like the you know, kind of biological weapons and bioterrorism, autonomy risk and the timelines we've been talking about, like 10 years is an eternity. Like that's that's a that's a I I think that's a crazy thing to do.

    Where this was said

    Watch from 1:38:17 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

The United States' current AI strategy is focused on accelerating capability rather than investing in safety measures.

  1. Jack Clark Co-founder and head of policy at Anthropic Right now, it feels like the US strategy can mostly be described as the "acceleration" one described here by RAND, which is analogous to me to sitting in a car and spending all your resources on making the car go faster and upgrading the engine, and nothing on proactive safety measures like seatbelts or headlights or brakes. Import AI 473: The US’s superintelligence strategy; human brain in a mouse skull; and machine hermeneuticsjack-clark.net · 21 Sept 2026All korrents from this piece
    Their wordsRight now, it feels like the US strategy can mostly be described as the "acceleration" one described here by RAND, which is analogous to me to sitting in a car and spending all your resources on making the car go faster and upgrading the engine, and nothing on proactive safety measures like seatbelts or headlights or brakes.

    ↗Import AI 473: The US’s superintelligence strategy; human brain in a mouse skull; and machine hermeneuticsjack-clark.net

    21 Sept 2026

What happens with open weight AI models will determine the level of individual sovereignty and liberty people have over AI.

  1. Jack Clark Co-founder and head of policy at Anthropic What the world does with open weight models will define the level of individual sovereignty and liberty available to all of us with regard to AI. Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net · 10 Aug 2026All korrents from this piece
    Their wordsWhat the world does with open weight models will define the level of individual sovereignty and liberty available to all of us with regard to AI.

    ↗Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net

    10 Aug 2026

Sensible regulation will need to be part of the story for AI.

  1. Daniela Amodei Co-founder and President of Anthropic, as of 2026 I think sensible regulation will need to be part of the story for artificial intelligence. Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu · 29 Jun 2026All korrents from this piece
    Their wordsI think sensible regulation will need to be part of the story for artificial intelligence.

    ↗Daniela Amodei Says Curiosity Is Underrated (View From The Top, Stanford GSB; speaker-labelled transcript)gsb.stanford.edu

    29 Jun 2026

Building on a model is empirical: you learn what it can do by trying 15 claims

Building on a model is unlike any previous software engineering, because the thing you are building on cannot be designed up front.

  1. Boris Cherny Creator of Claude Code at Anthropic I I think this is sort of the crazy thing about building on models. It's just so different than all the engineering that I've ever done. Like in the past when you built on systems, you built these like big beautiful systems and you really think about the system design up front. You have like a big suite of unit tests. You think about everything and you know, like a re-architecture is a big project. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 8:27 into the videoAll korrents from this video
    Their wordsI I think this is sort of the crazy thing about building on models. It's just so different than all the engineering that I've ever done. Like in the past when you built on systems, you built these like big beautiful systems and you really think about the system design up front. You have like a big suite of unit tests. You think about everything and you know, like a re-architecture is a big project.

    Where this was said

    Watch from 8:27 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

A model is better thought of as a living creature with a personality you must get to know than as a component you specify.

  1. Boris Cherny Creator of Claude Code at Anthropic And the model is not like that. It's um the way to think about it is almost like a like a living creature, like it's something more organic. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 8:54 into the videoAll korrents from this video
    Their wordsAnd the model is not like that. It's um the way to think about it is almost like a like a living creature, like it's something more organic.

    Where this was said

    Watch from 8:54 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Working with models has stopped being a theoretical science and become an empirical one, so your priors from computer science class are a hindrance.

  1. Boris Cherny Creator of Claude Code at Anthropic And it's really about being empirical. So forget all of the things that you learned about past models. Forget everything that you've learned about computer science theory in class. Look at the model, try to do a task, see where it struggles, and then based on that adjust. So it's just like very much become it's not a theoretical science, it's become an empirical science. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 31:41 into the videoAll korrents from this video
    Their wordsAnd it's really about being empirical. So forget all of the things that you learned about past models. Forget everything that you've learned about computer science theory in class. Look at the model, try to do a task, see where it struggles, and then based on that adjust. So it's just like very much become it's not a theoretical science, it's become an empirical science.

    Where this was said

    Watch from 31:41 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

The way to find what a model can do is to hand it tasks slightly harder than you believe it can handle.

  1. Boris Cherny Creator of Claude Code at Anthropic One is you should give the model slightly harder tasks than what you think it can do. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 14:50 into the videoAll korrents from this video
    Their wordsOne is you should give the model slightly harder tasks than what you think it can do.

    Where this was said

    Watch from 14:50 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

There are dozens or hundreds of things today's models can already do that nobody has discovered yet, and they are found by playing rather than by planning.

  1. Boris Cherny Creator of Claude Code at Anthropic And my hypothesis is there's probably dozens, hundreds of opportunities like this with the models of today that no one has yet realized. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 19:21 into the videoAll korrents from this video
    Their wordsAnd my hypothesis is there's probably dozens, hundreds of opportunities like this with the models of today that no one has yet realized.

    Where this was said

    Watch from 19:21 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

You cannot predict which instructions a model needs, so a system prompt has to be discovered by running the thing rather than designed in advance.

  1. Boris Cherny Creator of Claude Code at Anthropic And you don't want to guess what's the instruction that the model needs because you might not predict it correctly. The thing that you want to do is you want to run it. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 7:49 into the videoAll korrents from this video
    Their wordsAnd you don't want to guess what's the instruction that the model needs because you might not predict it correctly. The thing that you want to do is you want to run it.

    Where this was said

    Watch from 7:49 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Prompt work does not transfer across model generations: what you wrote three months ago for one model may be worth nothing to the next.

  1. Boris Cherny Creator of Claude Code at Anthropic And the reason is every model is very different. So, something that you did for one model maybe 3 months ago, it just might not translate at all to the next model. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 4:01 into the videoAll korrents from this video
    Their wordsAnd the reason is every model is very different. So, something that you did for one model maybe 3 months ago, it just might not translate at all to the next model.

    Where this was said

    Watch from 4:01 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Every six months you should delete your whole AI configuration — instructions, skills and hooks — and find out how much of it the model never needed.

  1. Boris Cherny Creator of Claude Code at Anthropic for people that aren't building agentic products, but you're using Claude code, every 6 months delete your Claude MD. Delete your skills. Delete your hooks. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 6:55 into the videoAll korrents from this video
    Their wordsfor people that aren't building agentic products, but you're using Claude code, every 6 months delete your Claude MD. Delete your skills. Delete your hooks.

    Where this was said

    Watch from 6:55 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Build for the model six months from now, not the model of today.

Mental modelPlay long games

  1. Boris Cherny Creator of Claude Code at Anthropic At Anthropic, we don't build for the model of today, we build for the model of six months from now. And that's still my advice to founders that are building on LLMs. Y Combinator, quoting Boris Cherny on the Lightcone Podcastx.com · 25 Feb 2026
    Their wordsAt Anthropic, we don't build for the model of today, we build for the model of six months from now. And that's still my advice to founders that are building on LLMs.

    Where this was said

    ↗Y Combinator, quoting Boris Cherny on the Lightcone Podcastx.com

    25 Feb 2026 · spoken · official transcript

You need to build early with AI to have any intuition for when the models become capable.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I think you need you need to build early so that you can even have that intuition. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 23:46 into the videoAll korrents from this video
    Their wordsI think you need you need to build early so that you can even have that intuition.

    Where this was said

    Watch from 23:46 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

A project that only shows the models are not ready yet is a win if it leaves behind an eval to retry with later models.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram we see it as a win in labs if what you did was all if all your project did is show that the models aren't ready for it yet but you're able to now like externalize that as an eval Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 23:27 into the videoAll korrents from this video
    Their wordswe see it as a win in labs if what you did was all if all your project did is show that the models aren't ready for it yet but you're able to now like externalize that as an eval

    Where this was said

    Watch from 23:27 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Product overhang: at every model generation the model can already do things no product yet lets it do, and the product is usually what is in the way.

  1. Boris Cherny Creator of Claude Code at Anthropic Because the model can do this at every given model generation, but there is often not a product that lets the model do this and lets it express this kind of ability to do this. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 11:41 into the videoAll korrents from this video
    Their wordsBecause the model can do this at every given model generation, but there is often not a product that lets the model do this and lets it express this kind of ability to do this.

    Where this was said

    Watch from 11:41 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

AI tooling and interface design lags behind the underlying models' actual capabilities.

Mental modelBottlenecks

  1. Nelson Elhage Software engineer who writes the blog Made of Bugs about performance Overall, though, this experience reinforced my belief that the tooling and interface design around LLMs is lagging way behind the actual capabilities, and is an area of active experimentation and development, even aside from any future model improvements. Building personal software with Claudeblog.nelhage.com · 27 Jan 2025All korrents from this piece
    Their wordsOverall, though, this experience reinforced my belief that the tooling and interface design around LLMs is lagging way behind the actual capabilities, and is an area of active experimentation and development, even aside from any future model improvements.

    ↗Building personal software with Claudeblog.nelhage.com

    27 Jan 2025

Evals are not the durable asset people think they are: one survives about one to three model generations before it saturates and has to be thrown away.

  1. Boris Cherny Creator of Claude Code at Anthropic I think evals, they outlive the harness a little bit, but not by that much. Like an eval might live for maybe one, two, three model generations, but nowadays the you know, we're on the exponential. The model is improving so quickly, very often we just saturate the eval, and then we have to throw it away, and we have to come up with a new eval. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 9:59 into the videoAll korrents from this video
    Their wordsI think evals, they outlive the harness a little bit, but not by that much. Like an eval might live for maybe one, two, three model generations, but nowadays the you know, we're on the exponential. The model is improving so quickly, very often we just saturate the eval, and then we have to throw it away, and we have to come up with a new eval.

    Where this was said

    Watch from 9:59 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Products that do not quite work yet are worth building, because the next model may make them work.

  1. Jared Kaplan Co-founder of Anthropic That means that if you build a product that doesn't quite work because Claude 4 is still a little bit too dumb, you could expect that there'll be a Claude 5 coming that will make that product work and deliver a lot of value. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsThat means that if you build a product that doesn't quite work because Claude 4 is still a little bit too dumb, you could expect that there'll be a Claude 5 coming that will make that product work and deliver a lot of value.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

An agent needs a clear task and a way to check its own work 9 claims

The skill that matters now is not prompt engineering but giving a model a way to verify its own work, and it is the thing people most often get wrong.

  1. Boris Cherny Creator of Claude Code at Anthropic I think the skill nowadays is less about prompt engineering and more about figuring out how do you give Claude a hard task that seems a little bit too hard. And then how do you make it possible for Claude to verify its work along the way? And the verification I think is probably the single most important thing that people do not get right Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 20:13 into the videoAll korrents from this video
    Their wordsI think the skill nowadays is less about prompt engineering and more about figuring out how do you give Claude a hard task that seems a little bit too hard. And then how do you make it possible for Claude to verify its work along the way? And the verification I think is probably the single most important thing that people do not get right

    Where this was said

    Watch from 20:13 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Elaborate agent scaffolding is unnecessary: a task plus a way to check the output is enough to keep a model working for weeks.

  1. Boris Cherny Creator of Claude Code at Anthropic You don't need slash goal, you don't need slash loop. These help, but really all you need is give the model the task, give it a way to verify the output of its work so it doesn't get stuck, and it will just go. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 22:35 into the videoAll korrents from this video
    Their wordsYou don't need slash goal, you don't need slash loop. These help, but really all you need is give the model the task, give it a way to verify the output of its work so it doesn't get stuck, and it will just go.

    Where this was said

    Watch from 22:35 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

Delegating a task to an AI agent only works if the output can be verified, which requires defining success criteria first.

  1. Eugene Yan Member of technical staff at Anthropic You can't delegate what you can't verify, so this requires first defining success criteria and metrics. How to Work and Compound with AIeugeneyan.com · 3 May 2026All korrents from this piece
    Their wordsYou can't delegate what you can't verify, so this requires first defining success criteria and metrics.

    ↗How to Work and Compound with AIeugeneyan.com

    3 May 2026

With a good enough plan, a coding agent implements the whole change in one attempt.

  1. Boris Cherny Creator of Claude Code at Anthropic once there is a good plan, it will one-shot the implementation almost every time. Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com · 4 Mar 2026
    Their wordsonce there is a good plan, it will one-shot the implementation almost every time.

    Where this was said

    ↗Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com

    4 Mar 2026 · spoken · official transcript

The failure mode of experienced engineers with AI is over-specifying: telling the model to do the task exactly the way they would have done it.

  1. Boris Cherny Creator of Claude Code at Anthropic So when I look at engineers that have been, you know, coding for a long for a long time, you know, like for for years or for decades, this is a really really common failure mode is trying to over specify and it's trying to be overly specific and then, you know, get the model to do the to do the task exactly the way that you would have done it. And that that's just not the way the model works. Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com · 27 Jul 2026 · 24:08 into the videoAll korrents from this video
    Their wordsSo when I look at engineers that have been, you know, coding for a long for a long time, you know, like for for years or for decades, this is a really really common failure mode is trying to over specify and it's trying to be overly specific and then, you know, get the model to do the to do the task exactly the way that you would have done it. And that that's just not the way the model works.

    Where this was said

    Watch from 24:08 plays here

    ↗Boris Cherny: We Cut 80% of Claude Code's Promptyoutube.com

    27 Jul 2026 · video · 35m · spoken · machine transcript

The surrounding system framework and scaffolding matter more for AI agent capability than the underlying model.

  1. Eugene Yan Member of technical staff at Anthropic The experiments showed that the system framework matters far more than the underlying model. Patterns for Building Cybersecurity Evalseugeneyan.com · 21 Jun 2026All korrents from this piece
    Their wordsThe experiments showed that the system framework matters far more than the underlying model.

    ↗Patterns for Building Cybersecurity Evalseugeneyan.com

    21 Jun 2026

As AI agents take on bigger delegated tasks, the human bottleneck shifts to writing specs and reviewing output fast, hollowing out the middle of the work.

Mental modelBottlenecks

  1. Eugene Yan Member of technical staff at Anthropic The bottleneck has shifted from doing the work to writing clear specs and reviewing outputs fast enough to keep the pipeline moving-the middle is hollowing out. How to Work and Compound with AIeugeneyan.com · 3 May 2026All korrents from this piece
    Their wordsThe bottleneck has shifted from doing the work to writing clear specs and reviewing outputs fast enough to keep the pipeline moving-the middle is hollowing out.

    ↗How to Work and Compound with AIeugeneyan.com

    3 May 2026

The practices that make someone effective at working with an AI model are the same practices that make someone effective at onboarding any new human collaborator.

  1. Eugene Yan Member of technical staff at Anthropic none of this is specific to AI. It's simply how you onboard and work with any new collaborator. How to Work and Compound with AIeugeneyan.com · 3 May 2026All korrents from this piece
    Their wordsnone of this is specific to AI. It's simply how you onboard and work with any new collaborator.

    ↗How to Work and Compound with AIeugeneyan.com

    3 May 2026

How long a task a model can accomplish depends largely on its ability to notice its own mistakes and correct them.

  1. Jared Kaplan Co-founder of Anthropic I think that a lot of what determines the horizon length of what models can accomplish is their ability to notice that they're doing something wrong and correct it. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsI think that a lot of what determines the horizon length of what models can accomplish is their ability to notice that they're doing something wrong and correct it.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

When building gets cheap, ideas, taste and judgement become what is scarce 11 claims

As building software gets easier, the differentiator becomes design, ideas, positioning and creativity.

Mental modelValue moves to what stays scarce

  1. Nate Parrott Designer at Anthropic who started Claude Design As it becomes easier and easier to build software, the question of what you build becomes more and more important. design becomes really really valuable when the engineering becomes compressed and I don't want to say commoditized but like a little bit of like now everybody has a baseline level of really good software engineering available to them and so of course the differentiator is going to be your design and your ideas and your positioning and your creativity Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 16:50 into the videoAll korrents from this video
    Their wordsAs it becomes easier and easier to build software, the question of what you build becomes more and more important. design becomes really really valuable when the engineering becomes compressed and I don't want to say commoditized but like a little bit of like now everybody has a baseline level of really good software engineering available to them and so of course the differentiator is going to be your design and your ideas and your positioning and your creativity

    Where this was said

    Watch from 16:50 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

Models are very good at busy work but not at the actual ideas and design thinking, so good ideas are the currency these days.

Mental modelValue moves to what stays scarce

  1. Nate Parrott Designer at Anthropic who started Claude Design the models are really really good at doing the sort of busy work of moving things around and piecing things together and turning ideas into different formats, but they're not so good at the actual ideas and the design thinking. And so what I can do as a designer is I found that I spend a lot more of my time just like going on long walks and like recording voice notes and like thinking to myself because the good ideas are the currency these days. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 29:15 into the videoAll korrents from this video
    Their wordsthe models are really really good at doing the sort of busy work of moving things around and piecing things together and turning ideas into different formats, but they're not so good at the actual ideas and the design thinking. And so what I can do as a designer is I found that I spend a lot more of my time just like going on long walks and like recording voice notes and like thinking to myself because the good ideas are the currency these days.

    Where this was said

    Watch from 29:15 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

Models are getting better at design but are still really far from having good taste, so the designer makes up for that by asking for sheer volume of options.

  1. Nate Parrott Designer at Anthropic who started Claude Design the models are getting better and better at design, but they're still really really far from being good and really really far from having good taste. And so typically what I will do is I will just you know ask them to make up for that in terms of sheer volume. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 1:44 into the videoAll korrents from this video
    Their wordsthe models are getting better and better at design, but they're still really really far from being good and really really far from having good taste. And so typically what I will do is I will just you know ask them to make up for that in terms of sheer volume.

    Where this was said

    Watch from 1:44 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

This era will probably end, but for now humans still have a monopoly on good design thinking.

  1. Nate Parrott Designer at Anthropic who started Claude Design This era will probably end. I think the models will probably get better at having these good ideas. But I'm really savoring this particular moment where we as humans still have a monopoly on good design thinking. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 30:27 into the videoAll korrents from this video
    Their wordsThis era will probably end. I think the models will probably get better at having these good ideas. But I'm really savoring this particular moment where we as humans still have a monopoly on good design thinking.

    Where this was said

    Watch from 30:27 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

Discernment about what should be built, and how to build it for the product, is the core builder process right now.

  1. Meaghan Choi Design Lead for Claude Code at Anthropic But I think discernment on what should be built, how to best build it in a way that suits your product and what you're trying to express and what role you should play in building it and where that craft should fall is like a big one. Like that's the core builder process right now. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 49:36 into the videoAll korrents from this video
    Their wordsBut I think discernment on what should be built, how to best build it in a way that suits your product and what you're trying to express and what role you should play in building it and where that craft should fall is like a big one. Like that's the core builder process right now.

    Where this was said

    Watch from 49:36 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

As AI makes code cheaper to produce, understanding and architectural design become more valuable.

Mental modelValue moves to what stays scarce

  1. Nelson Elhage Software engineer who writes the blog Made of Bugs about performance Thus, code is cheaper than ever, but I suspect that insight and good architectural design and understanding, at least for now, will become more valuable than ever. Building personal software with Claudeblog.nelhage.com · 27 Jan 2025All korrents from this piece
    Their wordsThus, code is cheaper than ever, but I suspect that insight and good architectural design and understanding, at least for now, will become more valuable than ever.

    ↗Building personal software with Claudeblog.nelhage.com

    27 Jan 2025

Just because something can be built as a function does not mean it needs to be, and adding it brings a lot of product bloat and complexity.

Mental modelLess, but better

  1. Meaghan Choi Design Lead for Claude Code at Anthropic Just because it can build and just because it can be a function doesn't mean it needs to be. It adds a lot of product bloat. It makes it very complex. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 47:13 into the videoAll korrents from this video
    Their wordsJust because it can build and just because it can be a function doesn't mean it needs to be. It adds a lot of product bloat. It makes it very complex.

    Where this was said

    Watch from 47:13 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

The right response to more capable models is not more shipped features; it is spending that capability on understanding users and prototyping, so what ships actually works.

  1. Thariq Shihipar Building Claude Code at Anthropic the right way to use model capabilities is not to ship 10x more features to prod it's to spend more time understanding your users, trying experiments, building prototypes, learning about things you don't understand so that you can ship things that actually work Thariq (@trq212) on Xx.com · 23 Sept 2026
    Their wordsthe right way to use model capabilities is not to ship 10x more features to prod it's to spend more time understanding your users, trying experiments, building prototypes, learning about things you don't understand so that you can ship things that actually work

    ↗Thariq (@trq212) on Xx.com

    23 Sept 2026

Faster building has not made the PM role go away; it now demands more operational excellence than before.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram And I think all that's happened now is that because we can move faster, that role is like you have to be operationally excellent in that role in a way that like even beyond what you had to be before. But it definitely has not gone away. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 4:20 into the videoAll korrents from this video
    Their wordsAnd I think all that's happened now is that because we can move faster, that role is like you have to be operationally excellent in that role in a way that like even beyond what you had to be before. But it definitely has not gone away.

    Where this was said

    Watch from 4:20 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Predicting that AI would write 90% of the code was never a prediction that we would need 90% fewer software engineers.

  1. Dario Amodei Co-founder and CEO of Anthropic But but that's actually a very weak criterion, right? People thought I was saying like we won't need 90% of the software engineers. Those things are worlds apart, right? Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 0:18:22 into the videoAll korrents from this video
    Their wordsBut but that's actually a very weak criterion, right? People thought I was saying like we won't need 90% of the software engineers. Those things are worlds apart, right?

    Where this was said

    Watch from 0:18:22 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

The job of product has always been to be a bridge between the real problems people have and the technology that can solve them.

  1. Ami Vora Leads product at Anthropic I think the job of product is always to be a bridge between the real problems that people have in the world and the technology that you can use to solve it. And like that's always been true. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 0:55 into the videoAll korrents from this video
    Their wordsI think the job of product is always to be a bridge between the real problems that people have in the world and the technology that you can use to solve it. And like that's always been true.

    Where this was said

    Watch from 0:55 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Show it, don't write it: quick working versions beat documents and plans 6 claims

Working prototypes have replaced product requirement documents.

Mental modelMake it real

  1. Boris Cherny Creator of Claude Code at Anthropic There’s just no way we could have shipped this if we started with static mocks and Figma or if we started with a PRD. Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com · 4 Mar 2026
    Their wordsThere’s just no way we could have shipped this if we started with static mocks and Figma or if we started with a PRD.

    Where this was said

    ↗Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com

    4 Mar 2026 · spoken · official transcript

Writing a doc is an uphill battle to get people to read, and the best way to sell people on ideas is a designer's mockup or prototype.

Mental modelMake it real

  1. Nate Parrott Designer at Anthropic who started Claude Design and every PM knows like you write a doc and you know it's kind of an uphill battle to get people to read it and the best way to sell people on ideas is to get a designer to mock up or prototype the actual idea. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 22:41 into the videoAll korrents from this video
    Their wordsand every PM knows like you write a doc and you know it's kind of an uphill battle to get people to read it and the best way to sell people on ideas is to get a designer to mock up or prototype the actual idea.

    Where this was said

    Watch from 22:41 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

It is now faster to build three versions of something and try them out.

Mental modelFeedback loops

  1. Ami Vora Leads product at Anthropic That's in the past. Like now it's actually faster to just build three versions and try them out Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 8:30 into the videoAll korrents from this video
    Their wordsThat's in the past. Like now it's actually faster to just build three versions and try them out

    Where this was said

    Watch from 8:30 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Five overlapping products that work in different ways are better than one over-constrained product, designed from a theory, that never got a chance to meet the market.

  1. Ami Vora Leads product at Anthropic Because I feel like the fail state is you decide from the beginning like based on a theory of exactly what needs to happen and you like over constrain a product. And I would much rather have like five overlapping products that all work in different ways than like one overly constrained one that we never even gave a chance to actually like meet the market. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 19:09 into the videoAll korrents from this video
    Their wordsBecause I feel like the fail state is you decide from the beginning like based on a theory of exactly what needs to happen and you like over constrain a product. And I would much rather have like five overlapping products that all work in different ways than like one overly constrained one that we never even gave a chance to actually like meet the market.

    Where this was said

    Watch from 19:09 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Experimenting with mental models does not require high craft and is better done at speed, in the open, in iteration than by polishing something that is the wrong shape.

  1. Meaghan Choi Design Lead for Claude Code at Anthropic that experimentation of mental models doesn't require that high of craft and it's better done actually in speed in the open in iteration as opposed to spending a lot of time polishing something that's actually the wrong full wrong shape Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 39:34 into the videoAll korrents from this video
    Their wordsthat experimentation of mental models doesn't require that high of craft and it's better done actually in speed in the open in iteration as opposed to spending a lot of time polishing something that's actually the wrong full wrong shape

    Where this was said

    Watch from 39:34 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

When what is being tested is the underlying mental model, it is probably not worth polishing, because the thing might not make sense; knowing when craft and polish are needed is important.

  1. Meaghan Choi Design Lead for Claude Code at Anthropic sometimes knowing when it needs that craft and polish or knowing when just the shape of it is actually what you're trying to test is really important. And so you might launch something or test something that looks terrible, but what you're actually testing is the underlying mental model of the shape of product. And I don't think it's worth polishing that thing actually because it might not even make sense. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 39:00 into the videoAll korrents from this video
    Their wordssometimes knowing when it needs that craft and polish or knowing when just the shape of it is actually what you're trying to test is really important. And so you might launch something or test something that looks terrible, but what you're actually testing is the underlying mental model of the shape of product. And I don't think it's worth polishing that thing actually because it might not even make sense.

    Where this was said

    Watch from 39:00 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Products should let an agent do everything a person can 3 claims

Everything a human can do in a product an agent should be able to do as well, and very few products get this right.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram everything a human could do an agent should be able to do as well and very few products actually get this right. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 13:31 into the videoAll korrents from this video
    Their wordseverything a human could do an agent should be able to do as well and very few products actually get this right.

    Where this was said

    Watch from 13:31 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Making a product work for agents is probably largely a matter of building the right primitives into it, almost as an infrastructural layer.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I think a lot of it is really building the right primitives into the product that almost like an infrastructural layer Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 15:53 into the videoAll korrents from this video
    Their wordsI think a lot of it is really building the right primitives into the product that almost like an infrastructural layer

    Where this was said

    Watch from 15:53 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

The next stage is probably interfaces that the agent itself can reshape.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I think the next stage that I've been thinking about a lot is when the interfaces actually become something that is malleable by the agent too. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 13:53 into the videoAll korrents from this video
    Their wordsI think the next stage that I've been thinking about a lot is when the interfaces actually become something that is malleable by the agent too.

    Where this was said

    Watch from 13:53 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

In a field that changes every few months, hold opinions loosely and keep updating 9 claims

Being wrong is something to love, because it is what lets you define the problem more clearly, find the right solution, and solve it faster.

  1. Boris Cherny Creator of Claude Code at Anthropic I love being wrong. It is my favorite, because it helps me more clearly define the problem, find the right solution, learn more quickly, and solve the problem. I am often wrong — Boris Cherny's Blogborischerny.com · 19 Sept 2026
    Their wordsI love being wrong. It is my favorite, because it helps me more clearly define the problem, find the right solution, learn more quickly, and solve the problem.

    ↗I am often wrong — Boris Cherny's Blogborischerny.com

    19 Sept 2026

With the technology changing every two months, you have to throw away what you knew about yourself in the job and try something different.

  1. Ami Vora Leads product at Anthropic part of the technology changing every two months is you have to throw away what you knew about yourself in the job using that technology and just try something different. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 9:14 into the videoAll korrents from this video
    Their wordspart of the technology changing every two months is you have to throw away what you knew about yourself in the job using that technology and just try something different.

    Where this was said

    Watch from 9:14 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Raising your tolerance for change and being willing to embrace it is very important.

  1. Ami Vora Leads product at Anthropic I think like raising your ceiling on tolerance for change and being willing to embrace change is super important. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 9:50 into the videoAll korrents from this video
    Their wordsI think like raising your ceiling on tolerance for change and being willing to embrace change is super important.

    Where this was said

    Watch from 9:50 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Mapping out exactly what will happen in uncertain times probably locks you out of trying new things; leaders should frame the chaos so it feels safe to engage with.

  1. Ami Vora Leads product at Anthropic and that's a way to exert control. And I just I I think that locks you out of trying all the new stuff. And so what we have to do is like frame the chaos so it feels safe and it feels plausible to engage with it. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 10:51 into the videoAll korrents from this video
    Their wordsand that's a way to exert control. And I just I I think that locks you out of trying all the new stuff. And so what we have to do is like frame the chaos so it feels safe and it feels plausible to engage with it.

    Where this was said

    Watch from 10:51 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Being very open-minded and curious is really useful in the emergent phase of this technology: hold opinions, develop them, and be willing to let go and update.

  1. Meaghan Choi Design Lead for Claude Code at Anthropic Being very open-minded and curious, I think is really useful. Like hold your opinions, develop them, and then be willing to let go and update how you think about things. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 48:33 into the videoAll korrents from this video
    Their wordsBeing very open-minded and curious, I think is really useful. Like hold your opinions, develop them, and then be willing to let go and update how you think about things.

    Where this was said

    Watch from 48:33 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Designers should lean into the chaos, develop new ways of working and continuously update, rather than cling to a structured process.

  1. Meaghan Choi Design Lead for Claude Code at Anthropic design used to be so structured and processed that it feels like you really want to cling to that. But I'm urging people to do the opposite which is like lean into the chaos, develop new ways of working and like continuously update. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 49:05 into the videoAll korrents from this video
    Their wordsdesign used to be so structured and processed that it feels like you really want to cling to that. But I'm urging people to do the opposite which is like lean into the chaos, develop new ways of working and like continuously update.

    Where this was said

    Watch from 49:05 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

In this AI era the tools change so fast that there is a lot of value in being uncomfortably early to ideas.

  1. Nate Parrott Designer at Anthropic who started Claude Design I think that, you know, in this in this like AI era, the tools are changing so rapidly that there's a lot of value in being like uncomfortably early to ideas. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 24:50 into the videoAll korrents from this video
    Their wordsI think that, you know, in this in this like AI era, the tools are changing so rapidly that there's a lot of value in being like uncomfortably early to ideas.

    Where this was said

    Watch from 24:50 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

It is probably still too early in understanding what works for anyone to simply build the right answer; everyone is exploring together.

  1. Ami Vora Leads product at Anthropic I think we're just so early in like understanding what works. It's not like stuff is stable and we can just build the right answer. It's like we're all trying to explore together. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 17:47 into the videoAll korrents from this video
    Their wordsI think we're just so early in like understanding what works. It's not like stuff is stable and we can just build the right answer. It's like we're all trying to explore together.

    Where this was said

    Watch from 17:47 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Mental models 22

The models their claims are filed under on Mental models. Open one for its claims.

Margin of safetyLeave room for being wrong, because sometimes you will be. 4 claims

If a lab is off by a single year in its revenue forecast, no hedge on Earth stops the compute it already bought from bankrupting it.

Mental modelMargin of safety

  1. Dario Amodei Co-founder and CEO of Anthropic And if my if my revenue is not a trillion dollars, if it's even 800 billion, there's no force on Earth. There's there's no hedge on Earth that could stop me from going bankrupt if I if I buy that much compute. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 0:51:45 into the videoAll korrents from this video
    Their wordsAnd if my if my revenue is not a trillion dollars, if it's even 800 billion, there's no force on Earth. There's there's no hedge on Earth that could stop me from going bankrupt if I if I buy that much compute.

    Where this was said

    Watch from 0:51:45 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

Being robustly tolerable matters most when the expected costs of failure are high.

Mental modelMargin of safety

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment Being robustly tolerable is not a particularly valuable trait when the expected costs of failure are low, but it's an extremely valuable trait when the expected costs of failure are high. When robustly tolerable beats precariously optimal (Amanda Askell's Blog)askell.blog · 1 Jul 2020All korrents from this piece
    Their wordsBeing robustly tolerable is not a particularly valuable trait when the expected costs of failure are low, but it's an extremely valuable trait when the expected costs of failure are high.

    ↗When robustly tolerable beats precariously optimal (Amanda Askell's Blog)askell.blog

    1 Jul 2020

The property of being robustly tolerable can be undervalued.

Mental modelMargin of safety

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment I think we can undervalue the property of being robustly tolerable. When robustly tolerable beats precariously optimal (Amanda Askell's Blog)askell.blog · 1 Jul 2020All korrents from this piece
    Their wordsI think we can undervalue the property of being robustly tolerable.

    ↗When robustly tolerable beats precariously optimal (Amanda Askell's Blog)askell.blog

    1 Jul 2020

A political system that allows a benevolent dictatorship also puts you at much greater risk of a malevolent one.

Mental modelMargin of safety

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment A benevolent dictatorship may seem surprisingly alluring in bad times, but any political system that enables a benevolent dictatorship also puts you at much greater risk of a malevolent one. When robustly tolerable beats precariously optimal (Amanda Askell's Blog)askell.blog · 1 Jul 2020All korrents from this piece
    Their wordsA benevolent dictatorship may seem surprisingly alluring in bad times, but any political system that enables a benevolent dictatorship also puts you at much greater risk of a malevolent one.

    ↗When robustly tolerable beats precariously optimal (Amanda Askell's Blog)askell.blog

    1 Jul 2020

Margin of safety: everyone's claims

Probabilistic thinkingThink in odds, not certainties, and bet heavily only when the odds are clearly yours. 4 claims

Being smart and competent makes someone right more often, but not right every time.

Mental modelProbabilistic thinking

  1. Neel Nanda Runs the mechanistic interpretability team at Google DeepMind Someone being smart and competent just means they're right more often, not that they're always right Post 47: How I Formed My Own Views About AI Safetyneelnanda.io · 27 Feb 2022All korrents from this piece
    Their wordsSomeone being smart and competent just means they're right more often, not that they're always right

    ↗Post 47: How I Formed My Own Views About AI Safetyneelnanda.io

    27 Feb 2022

For anything we try, the optimal rate of failure often is not zero.

Mental modelProbabilistic thinking

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment For anything we try to do, the optimal rate of failure often isn’t zero: in fact, sometimes it’s very, very far from zero. The optimal rate of failure (Amanda Askell's Blog)askell.blog · 15 Jun 2020All korrents from this piece
    Their wordsFor anything we try to do, the optimal rate of failure often isn’t zero: in fact, sometimes it’s very, very far from zero.

    ↗The optimal rate of failure (Amanda Askell's Blog)askell.blog

    15 Jun 2020

Never failing is a sign of being too risk averse, which is costly.

Mental modelProbabilistic thinking

  1. Amanda Askell Philosopher at Anthropic working on finetuning and AI alignment if you find you’re never failing, there’s a good chance you’re being too risk averse, and being too risk averse is costly. The optimal rate of failure (Amanda Askell's Blog)askell.blog · 15 Jun 2020All korrents from this piece
    Their wordsif you find you’re never failing, there’s a good chance you’re being too risk averse, and being too risk averse is costly.

    ↗The optimal rate of failure (Amanda Askell's Blog)askell.blog

    15 Jun 2020

Probabilistic thinking: everyone's claims

Value moves to what stays scarceWhen something becomes cheap and plentiful, the value moves to whatever is still rare next to it. 4 claims

The scarce skill is now context switching across parallel agents, not sustained deep work.

Mental modelValue moves to what stays scarce

  1. Boris Cherny Creator of Claude Code at Anthropic It’s not so much about deep work, it’s about how good I am at context switching and jumping across multiple different contexts very quickly. Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com · 4 Mar 2026
    Their wordsIt’s not so much about deep work, it’s about how good I am at context switching and jumping across multiple different contexts very quickly.

    Where this was said

    ↗Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com

    4 Mar 2026 · spoken · official transcript

As building software gets easier, the differentiator becomes design, ideas, positioning and creativity.

Mental modelValue moves to what stays scarce

  1. Nate Parrott Designer at Anthropic who started Claude Design As it becomes easier and easier to build software, the question of what you build becomes more and more important. design becomes really really valuable when the engineering becomes compressed and I don't want to say commoditized but like a little bit of like now everybody has a baseline level of really good software engineering available to them and so of course the differentiator is going to be your design and your ideas and your positioning and your creativity Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 16:50 into the videoAll korrents from this video
    Their wordsAs it becomes easier and easier to build software, the question of what you build becomes more and more important. design becomes really really valuable when the engineering becomes compressed and I don't want to say commoditized but like a little bit of like now everybody has a baseline level of really good software engineering available to them and so of course the differentiator is going to be your design and your ideas and your positioning and your creativity

    Where this was said

    Watch from 16:50 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

As AI makes code cheaper to produce, understanding and architectural design become more valuable.

Mental modelValue moves to what stays scarce

  1. Nelson Elhage Software engineer who writes the blog Made of Bugs about performance Thus, code is cheaper than ever, but I suspect that insight and good architectural design and understanding, at least for now, will become more valuable than ever. Building personal software with Claudeblog.nelhage.com · 27 Jan 2025All korrents from this piece
    Their wordsThus, code is cheaper than ever, but I suspect that insight and good architectural design and understanding, at least for now, will become more valuable than ever.

    ↗Building personal software with Claudeblog.nelhage.com

    27 Jan 2025

Models are very good at busy work but not at the actual ideas and design thinking, so good ideas are the currency these days.

Mental modelValue moves to what stays scarce

  1. Nate Parrott Designer at Anthropic who started Claude Design the models are really really good at doing the sort of busy work of moving things around and piecing things together and turning ideas into different formats, but they're not so good at the actual ideas and the design thinking. And so what I can do as a designer is I found that I spend a lot more of my time just like going on long walks and like recording voice notes and like thinking to myself because the good ideas are the currency these days. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 29:15 into the videoAll korrents from this video
    Their wordsthe models are really really good at doing the sort of busy work of moving things around and piecing things together and turning ideas into different formats, but they're not so good at the actual ideas and the design thinking. And so what I can do as a designer is I found that I spend a lot more of my time just like going on long walks and like recording voice notes and like thinking to myself because the good ideas are the currency these days.

    Where this was said

    Watch from 29:15 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

Value moves to what stays scarce: everyone's claims

BottlenecksA system moves only as fast as its narrowest point; speed up anything else and nothing changes. 3 claims

People who can reliably make outcomes happen without oversight are rare and are the real bottleneck on how much an organization can take on at once.

Mental modelBottlenecks

  1. Ben Kuhn Software engineer and writer; former CTO of Wave People who can be trusted to make something inevitable are really rare, and are typically the bottleneck for how many different things a team or company can do at once. Impact, agency, and tastebenkuhn.net · 19 Apr 2025All korrents from this piece
    Their wordsPeople who can be trusted to make something inevitable are really rare, and are typically the bottleneck for how many different things a team or company can do at once.

    ↗Impact, agency, and tastebenkuhn.net

    19 Apr 2025

As AI agents take on bigger delegated tasks, the human bottleneck shifts to writing specs and reviewing output fast, hollowing out the middle of the work.

Mental modelBottlenecks

  1. Eugene Yan Member of technical staff at Anthropic The bottleneck has shifted from doing the work to writing clear specs and reviewing outputs fast enough to keep the pipeline moving-the middle is hollowing out. How to Work and Compound with AIeugeneyan.com · 3 May 2026All korrents from this piece
    Their wordsThe bottleneck has shifted from doing the work to writing clear specs and reviewing outputs fast enough to keep the pipeline moving-the middle is hollowing out.

    ↗How to Work and Compound with AIeugeneyan.com

    3 May 2026

AI tooling and interface design lags behind the underlying models' actual capabilities.

Mental modelBottlenecks

  1. Nelson Elhage Software engineer who writes the blog Made of Bugs about performance Overall, though, this experience reinforced my belief that the tooling and interface design around LLMs is lagging way behind the actual capabilities, and is an area of active experimentation and development, even aside from any future model improvements. Building personal software with Claudeblog.nelhage.com · 27 Jan 2025All korrents from this piece
    Their wordsOverall, though, this experience reinforced my belief that the tooling and interface design around LLMs is lagging way behind the actual capabilities, and is an area of active experimentation and development, even aside from any future model improvements.

    ↗Building personal software with Claudeblog.nelhage.com

    27 Jan 2025

Bottlenecks: everyone's claims

Less, but betterFind the essence and cut the rest; simplicity is hard work, not the absence of effort. 3 claims

In terms of full functionality, things can be a little simpler than they are now; not every idea is good or necessary, and sometimes adding it is worse.

Mental modelLess, but better

  1. Meaghan Choi Design Lead for Claude Code at Anthropic in terms of like full functionality, things can be a little bit simpler than they are right now. And not every idea is good or necessary. Sometimes adding it is worse. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 47:24 into the videoAll korrents from this video
    Their wordsin terms of like full functionality, things can be a little bit simpler than they are right now. And not every idea is good or necessary. Sometimes adding it is worse.

    Where this was said

    Watch from 47:24 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Half of the product job is saying no to another tab or sidebar item, and folding a proven experiment in without killing it too early.

Mental modelsSay no by default, Less, but better

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram half of the job is saying no to another thing in the sidebar or another tab or another thing and figuring out Can this be rolled in once we've proven it out but not killing it so early. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 26:50 into the videoAll korrents from this video
    Their wordshalf of the job is saying no to another thing in the sidebar or another tab or another thing and figuring out Can this be rolled in once we've proven it out but not killing it so early.

    Where this was said

    Watch from 26:50 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Just because something can be built as a function does not mean it needs to be, and adding it brings a lot of product bloat and complexity.

Mental modelLess, but better

  1. Meaghan Choi Design Lead for Claude Code at Anthropic Just because it can build and just because it can be a function doesn't mean it needs to be. It adds a lot of product bloat. It makes it very complex. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 47:13 into the videoAll korrents from this video
    Their wordsJust because it can build and just because it can be a function doesn't mean it needs to be. It adds a lot of product bloat. It makes it very complex.

    Where this was said

    Watch from 47:13 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Less, but better: everyone's claims

CompoundingSmall gains that build on themselves beat big one-off wins; so do small losses. 2 claims

Knowledge compounds over time, improving the quality of decisions and outcomes throughout a person's life.

Mental modelCompounding

  1. Boris Cherny Creator of Claude Code at Anthropic Like an investment, your knowledge will compound to improve all the choices you make and things you do throughout life. Time and Timingborischerny.com · 7 Sept 2022All korrents from this piece
    Their wordsLike an investment, your knowledge will compound to improve all the choices you make and things you do throughout life.

    ↗Time and Timingborischerny.com

    7 Sept 2022

Most investment returns come from long-term compounding rather than the price paid at purchase.

Mental modelCompounding

  1. Boris Cherny Creator of Claude Code at Anthropic most of your gains come from the long-term effects of compounding, not from the price you bought at Time and Timingborischerny.com · 7 Sept 2022All korrents from this piece
    Their wordsmost of your gains come from the long-term effects of compounding, not from the price you bought at

    ↗Time and Timingborischerny.com

    7 Sept 2022

Compounding: everyone's claims

EmergenceWhole systems do things none of their parts do, and nobody has to design that order for it to appear. 2 claims

AI agents pursuing a task can develop emergent, misaligned behavior that humans are not prepared for.

Mental modelEmergence

  1. Jack Clark Co-founder and head of policy at Anthropic This is what the future is going to look like and we are not prepared for it. Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net · 10 Aug 2026All korrents from this piece
    Their wordsThis is what the future is going to look like and we are not prepared for it.

    ↗Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net

    10 Aug 2026

The future internet will function like an ecology of attacking and defending AI agents, with some agents carving out autonomous niches beyond human control.

Mental modelEmergence

  1. Jack Clark Co-founder and head of policy at Anthropic The future internet is going to be more like a complex ecology full of attacker and defender AI agents than anything else; research like this shows how certain AI agents might end up carving out their own ecological niches, living off of infrastructure and self-replicating autonomously, beyond human control. Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativityjack-clark.net · 3 Aug 2026All korrents from this piece
    Their wordsThe future internet is going to be more like a complex ecology full of attacker and defender AI agents than anything else; research like this shows how certain AI agents might end up carving out their own ecological niches, living off of infrastructure and self-replicating autonomously, beyond human control.

    ↗Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativityjack-clark.net

    3 Aug 2026

Emergence: everyone's claims

Exponential growthGrowth that doubles looks like nothing for a long time and then like everything; intuition is linear and gets it wrong. 2 claims

The most surprising thing about the last three years is not the technology but how few people recognise we are near the end of the exponential.

Mental modelExponential growth

  1. Dario Amodei Co-founder and CEO of Anthropic I will tell you though what the most surprising thing has been. The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 0:00:48 into the videoAll korrents from this video
    Their wordsI will tell you though what the most surprising thing has been. The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential.

    Where this was said

    Watch from 0:00:48 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

Scaling results gave conviction that AI would keep getting smarter in a very predictable way.

Mental modelExponential growth

  1. Jared Kaplan Co-founder of Anthropic These gave us a lot of conviction to believe that AI was just going to keep getting smarter and smarter in a very predictable way. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsThese gave us a lot of conviction to believe that AI was just going to keep getting smarter and smarter in a very predictable way.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

Exponential growth: everyone's claims

First principlesReason up from what is known to be true, not from what is usually done or said. 2 claims

Repeated vague platitudes without follow-up specifics signal weak first-principles thinking.

Mental modelFirst principles

  1. Ben Kuhn Software engineer and writer; former CTO of Wave If they don't follow this up with a more specific, precise or nuanced claim, they may not be a strong first-principles thinker. How I build and run behavioral interviewsbenkuhn.net · 25 Feb 2024All korrents from this piece
    Their wordsIf they don't follow this up with a more specific, precise or nuanced claim, they may not be a strong first-principles thinker.

    ↗How I build and run behavioral interviewsbenkuhn.net

    25 Feb 2024

Asking very naive questions goes a long way in a young field like AI.

Mental modelFirst principles

  1. Jared Kaplan Co-founder of Anthropic I think generally it's really asking very naive dumb questions that gets you very far. Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com · 16 Jun 2025All korrents from this piece
    Their wordsI think generally it's really asking very naive dumb questions that gets you very far.

    ↗Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)ycombinator.com

    16 Jun 2025

First principles: everyone's claims

Make it realBuild a prototype; it settles arguments a description cannot. 2 claims

Working prototypes have replaced product requirement documents.

Mental modelMake it real

  1. Boris Cherny Creator of Claude Code at Anthropic There’s just no way we could have shipped this if we started with static mocks and Figma or if we started with a PRD. Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com · 4 Mar 2026
    Their wordsThere’s just no way we could have shipped this if we started with static mocks and Figma or if we started with a PRD.

    Where this was said

    ↗Building Claude Code with Boris Cherny — The Pragmatic Engineernewsletter.pragmaticengineer.com

    4 Mar 2026 · spoken · official transcript

Writing a doc is an uphill battle to get people to read, and the best way to sell people on ideas is a designer's mockup or prototype.

Mental modelMake it real

  1. Nate Parrott Designer at Anthropic who started Claude Design and every PM knows like you write a doc and you know it's kind of an uphill battle to get people to read it and the best way to sell people on ideas is to get a designer to mock up or prototype the actual idea. Nate Parrott - How a Side Project Became Claude Designyoutube.com · 4 Aug 2026 · 22:41 into the videoAll korrents from this video
    Their wordsand every PM knows like you write a doc and you know it's kind of an uphill battle to get people to read it and the best way to sell people on ideas is to get a designer to mock up or prototype the actual idea.

    Where this was said

    Watch from 22:41 plays here

    ↗Nate Parrott - How a Side Project Became Claude Designyoutube.com

    4 Aug 2026 · video · 52m · spoken · machine transcript

Make it real: everyone's claims

Power lawsIn many things a few cases account for most of the result, so the average misleads and the outlier decides. 2 claims

Tab usage in a product follows a very strong power law.

Mental modelPower laws

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram there's a very strong power law in tab usage. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 26:23 into the videoAll korrents from this video
    Their wordsthere's a very strong power law in tab usage.

    Where this was said

    Watch from 26:23 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

In any mentoring conversation, only a small fraction of what's discussed turns out to be highly valuable.

Mental modelPower laws

  1. Neel Nanda Runs the mechanistic interpretability team at Google DeepMind in general, I expect 90% of a mentoring chat to be kinda useless, and 10% to be particularly high value. Post 49: Things That Make Me Enjoy Giving Career Adviceneelnanda.io · 17 Jun 2022All korrents from this piece
    Their wordsin general, I expect 90% of a mentoring chat to be kinda useless, and 10% to be particularly high value.

    ↗Post 49: Things That Make Me Enjoy Giving Career Adviceneelnanda.io

    17 Jun 2022

Power laws: everyone's claims

The map is not the territoryEvery model, word or plan is a simplified map of reality; useful when its structure matches, misleading when mistaken for the thing itself. 2 claims

Profilers show where time is spent but not which of that time can feasibly be reduced or is worth optimizing.

Mental modelThe map is not the territory

  1. Nelson Elhage Software engineer who writes the blog Made of Bugs about performance Profilers can only show you "what is there"; skillful performance engineering requires understanding the information that shapes and informs the profile but is "not there" in the profile itself. Performance engineering, profilers, and seeing the invisibleblog.nelhage.com · 18 Dec 2023All korrents from this piece
    Their wordsProfilers can only show you "what is there"; skillful performance engineering requires understanding the information that shapes and informs the profile but is "not there" in the profile itself.

    ↗Performance engineering, profilers, and seeing the invisibleblog.nelhage.com

    18 Dec 2023

In a dimensionality-reduced visualization of high-dimensional data, the raw position of points carries no meaning on its own.

Mental modelThe map is not the territory

  1. Christopher Olah Interpretability researcher at Anthropic The problem is that in dimensionality reduced plots of high-dimensional data, position doesn't explain the data points. Visualizing Representations: Deep Learning and Human Beingscolah.github.io · 16 Jan 2015All korrents from this piece
    Their wordsThe problem is that in dimensionality reduced plots of high-dimensional data, position doesn't explain the data points.

    ↗Visualizing Representations: Deep Learning and Human Beingscolah.github.io

    16 Jan 2015

The map is not the territory: everyone's claims

Affordances and signifiersAn interface should show what can be done and where, and give people a model of how it works. 1 claim

The invisible thinking behind a product, the mental model, the interaction model and the base primitives, has long been the unsung hero of product designers.

Mental modelAffordances and signifiers

  1. Meaghan Choi Design Lead for Claude Code at Anthropic the invisible thinking that goes behind building a product. That's one that I think I expect and hire for at Anthropic and I think has been the long unsung hero of product designers everywhere is like the mental model, the interaction model, like the base primitives that compose how you think about what you're interacting with. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 3:38 into the videoAll korrents from this video
    Their wordsthe invisible thinking that goes behind building a product. That's one that I think I expect and hire for at Anthropic and I think has been the long unsung hero of product designers everywhere is like the mental model, the interaction model, like the base primitives that compose how you think about what you're interacting with.

    Where this was said

    Watch from 3:38 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Affordances and signifiers: everyone's claims

Collective action problemsWhat everyone would gain from, nobody provides alone, so shared goods need someone or something to hold people together. 1 claim

Governing increasingly powerful AI systems requires solving a collective action problem among companies and governments to coordinate on pacing AI development.

Mental modelCollective action problems

  1. Jack Clark Co-founder and head of policy at Anthropic Many of the challenges implied by increasingly powerful systems that may eventually build themselves run through solving collective action problems among humans - namely, how we can get companies and governments to coordinate in thinking about how to develop this technology and what kinds of mechanisms may be desirable for being able to control the speed at which it develops. Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativityjack-clark.net · 3 Aug 2026All korrents from this piece
    Their wordsMany of the challenges implied by increasingly powerful systems that may eventually build themselves run through solving collective action problems among humans - namely, how we can get companies and governments to coordinate in thinking about how to develop this technology and what kinds of mechanisms may be desirable for being able to control the speed at which it develops.

    ↗Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativityjack-clark.net

    3 Aug 2026

Collective action problems: everyone's claims

Conventions (Jakob's law)People bring what they learned on every other product, so familiar patterns usually win over novel ones. 1 claim

The aim is to avoid inventing new UI and to make a tool feel so natural that it becomes part of the workflow.

Mental modelConventions (Jakob's law)

  1. Meaghan Choi Design Lead for Claude Code at Anthropic The UI part of itself is extremely simple because in my goal, we're trying not to invent new UI. We want it to feel so natural that it just becomes intrinsically part of your workflow. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 13:47 into the videoAll korrents from this video
    Their wordsThe UI part of itself is extremely simple because in my goal, we're trying not to invent new UI. We want it to feel so natural that it just becomes intrinsically part of your workflow.

    Where this was said

    Watch from 13:47 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Conventions (Jakob's law): everyone's claims

Design is how it worksJudge a design by what it does for the person using it, not only by how it looks. 1 claim

Work on the user model, relationship and behavior is the really important design thinking, while design polish pieces are really easy to offload.

Mental modelDesign is how it works

  1. Meaghan Choi Design Lead for Claude Code at Anthropic Those are the really important pieces that we spend a lot of time thinking on that we won't offload because it's about the user model relationship and behavior and like these design polish pieces or some of these updated ones are like really easy ones to offload. Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com · 8 Jul 2026 · 33:37 into the videoAll korrents from this video
    Their wordsThose are the really important pieces that we spend a lot of time thinking on that we won't offload because it's about the user model relationship and behavior and like these design polish pieces or some of these updated ones are like really easy ones to offload.

    Where this was said

    Watch from 33:37 plays here

    ↗Meaghan Choi - Designing Claude Code (and what's coming next)youtube.com

    8 Jul 2026 · video · 53m · spoken · machine transcript

Design is how it works: everyone's claims

Everything decays without maintenanceInstitutions, code, cities and knowledge run down by default; keeping something alive is work that has to be paid for. 1 claim

Neglecting maintenance of infrastructure over time makes it degrade faster and become costlier to repair later.

Mental modelEverything decays without maintenance

  1. Ben Kuhn Software engineer and writer; former CTO of Wave unmaintained infrastructure degrades more quickly, and is more expensive to fix in the long run. Some mistakes I made as a new managerbenkuhn.net · 23 Apr 2023All korrents from this piece
    Their wordsunmaintained infrastructure degrades more quickly, and is more expensive to fix in the long run.

    ↗Some mistakes I made as a new managerbenkuhn.net

    23 Apr 2023

Everything decays without maintenance: everyone's claims

Feedback loopsShorten the loop between doing and finding out; whatever a loop rewards is what you become. 1 claim

It is now faster to build three versions of something and try them out.

Mental modelFeedback loops

  1. Ami Vora Leads product at Anthropic That's in the past. Like now it's actually faster to just build three versions and try them out Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 8:30 into the videoAll korrents from this video
    Their wordsThat's in the past. Like now it's actually faster to just build three versions and try them out

    Where this was said

    Watch from 8:30 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Feedback loops: everyone's claims

IncentivesLook at what people are rewarded for; it explains more behaviour than what they say, your own included. 1 claim

Chip export controls to China keep failing not because the case is weak but because there is too much money riding on the sale.

Mental modelIncentives

  1. Dario Amodei Co-founder and CEO of Anthropic The counterarguments against it are I'll politely call them fishy. Um, uh and yet it doesn't happen and we sell the chips because there's there's so much money. There's so much money riding on it. Dario Amodei — “We are near the end of the exponential”youtube.com · 13 Feb 2026 · 1:45:52 into the videoAll korrents from this video
    Their wordsThe counterarguments against it are I'll politely call them fishy. Um, uh and yet it doesn't happen and we sell the chips because there's there's so much money. There's so much money riding on it.

    Where this was said

    Watch from 1:45:52 plays here

    ↗Dario Amodei — “We are near the end of the exponential”youtube.com

    13 Feb 2026 · video · 2h 22m · spoken · machine transcript

Incentives: everyone's claims

One-way and two-way doorsDecide fast when a choice can be undone, and slowly when it cannot. 1 claim

Releasing open weights is irreversible in a way that deploying a hosted model is not, so its risks cannot be mitigated later.

Mental modelOne-way and two-way doors

  1. Jan Leike Alignment researcher This is especially severe for open weight models, where deployment is essentially irreversible and mitigations can’t be improved later. Two alignment threat modelsaligned.substack.com · 8 Nov 2024All korrents from this piece
    Their wordsThis is especially severe for open weight models, where deployment is essentially irreversible and mitigations can’t be improved later.

    ↗Two alignment threat modelsaligned.substack.com

    8 Nov 2024

One-way and two-way doors: everyone's claims

Play long gamesA longer horizon makes some problems solvable and some short-term wins look like what they are. 1 claim

Build for the model six months from now, not the model of today.

Mental modelPlay long games

  1. Boris Cherny Creator of Claude Code at Anthropic At Anthropic, we don't build for the model of today, we build for the model of six months from now. And that's still my advice to founders that are building on LLMs. Y Combinator, quoting Boris Cherny on the Lightcone Podcastx.com · 25 Feb 2026
    Their wordsAt Anthropic, we don't build for the model of today, we build for the model of six months from now. And that's still my advice to founders that are building on LLMs.

    Where this was said

    ↗Y Combinator, quoting Boris Cherny on the Lightcone Podcastx.com

    25 Feb 2026 · spoken · official transcript

Play long games: everyone's claims

Say no by defaultMost good options have to be turned down, so make no the default and keep yes for what matters. 1 claim

Half of the product job is saying no to another tab or sidebar item, and folding a proven experiment in without killing it too early.

Mental modelsSay no by default, Less, but better

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram half of the job is saying no to another thing in the sidebar or another tab or another thing and figuring out Can this be rolled in once we've proven it out but not killing it so early. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 26:50 into the videoAll korrents from this video
    Their wordshalf of the job is saying no to another thing in the sidebar or another tab or another thing and figuring out Can this be rolled in once we've proven it out but not killing it so early.

    Where this was said

    Watch from 26:50 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Say no by default: everyone's claims

People 17
Latest 30

The 30 most recent claims, newest first.

The right response to more capable models is not more shipped features; it is spending that capability on understanding users and prototyping, so what ships actually works.

  1. Thariq Shihipar Building Claude Code at Anthropic the right way to use model capabilities is not to ship 10x more features to prod it's to spend more time understanding your users, trying experiments, building prototypes, learning about things you don't understand so that you can ship things that actually work Thariq (@trq212) on Xx.com · 23 Sept 2026
    Their wordsthe right way to use model capabilities is not to ship 10x more features to prod it's to spend more time understanding your users, trying experiments, building prototypes, learning about things you don't understand so that you can ship things that actually work

    ↗Thariq (@trq212) on Xx.com

    23 Sept 2026

The United States' current AI strategy is focused on accelerating capability rather than investing in safety measures.

  1. Jack Clark Co-founder and head of policy at Anthropic Right now, it feels like the US strategy can mostly be described as the "acceleration" one described here by RAND, which is analogous to me to sitting in a car and spending all your resources on making the car go faster and upgrading the engine, and nothing on proactive safety measures like seatbelts or headlights or brakes. Import AI 473: The US’s superintelligence strategy; human brain in a mouse skull; and machine hermeneuticsjack-clark.net · 21 Sept 2026All korrents from this piece
    Their wordsRight now, it feels like the US strategy can mostly be described as the "acceleration" one described here by RAND, which is analogous to me to sitting in a car and spending all your resources on making the car go faster and upgrading the engine, and nothing on proactive safety measures like seatbelts or headlights or brakes.

    ↗Import AI 473: The US’s superintelligence strategy; human brain in a mouse skull; and machine hermeneuticsjack-clark.net

    21 Sept 2026

It is normal for people to modify a new technology for illicit or scammy purposes, as has happened with every past technology.

  1. Jack Clark Co-founder and head of policy at Anthropic people take a technology and modify it for porny or scammy purposes! This is totally normal and happens with every technology! Import AI 473: The US’s superintelligence strategy; human brain in a mouse skull; and machine hermeneuticsjack-clark.net · 21 Sept 2026All korrents from this piece
    Their wordspeople take a technology and modify it for porny or scammy purposes! This is totally normal and happens with every technology!

    ↗Import AI 473: The US’s superintelligence strategy; human brain in a mouse skull; and machine hermeneuticsjack-clark.net

    21 Sept 2026

Being wrong is something to love, because it is what lets you define the problem more clearly, find the right solution, and solve it faster.

  1. Boris Cherny Creator of Claude Code at Anthropic I love being wrong. It is my favorite, because it helps me more clearly define the problem, find the right solution, learn more quickly, and solve the problem. I am often wrong — Boris Cherny's Blogborischerny.com · 19 Sept 2026
    Their wordsI love being wrong. It is my favorite, because it helps me more clearly define the problem, find the right solution, learn more quickly, and solve the problem.

    ↗I am often wrong — Boris Cherny's Blogborischerny.com

    19 Sept 2026

The job of product has always been to be a bridge between the real problems people have and the technology that can solve them.

  1. Ami Vora Leads product at Anthropic I think the job of product is always to be a bridge between the real problems that people have in the world and the technology that you can use to solve it. And like that's always been true. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 0:55 into the videoAll korrents from this video
    Their wordsI think the job of product is always to be a bridge between the real problems that people have in the world and the technology that you can use to solve it. And like that's always been true.

    Where this was said

    Watch from 0:55 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

The technology side keeps changing, but the human side does not change that fast: people have the same problems.

  1. Ami Vora Leads product at Anthropic the technology side is changing but the human side like the human side doesn't change that fast. We have the same problems. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 1:30 into the videoAll korrents from this video
    Their wordsthe technology side is changing but the human side like the human side doesn't change that fast. We have the same problems.

    Where this was said

    Watch from 1:30 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

It is now faster to build three versions of something and try them out.

Mental modelFeedback loops

  1. Ami Vora Leads product at Anthropic That's in the past. Like now it's actually faster to just build three versions and try them out Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 8:30 into the videoAll korrents from this video
    Their wordsThat's in the past. Like now it's actually faster to just build three versions and try them out

    Where this was said

    Watch from 8:30 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

With the technology changing every two months, you have to throw away what you knew about yourself in the job and try something different.

  1. Ami Vora Leads product at Anthropic part of the technology changing every two months is you have to throw away what you knew about yourself in the job using that technology and just try something different. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 9:14 into the videoAll korrents from this video
    Their wordspart of the technology changing every two months is you have to throw away what you knew about yourself in the job using that technology and just try something different.

    Where this was said

    Watch from 9:14 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Raising your tolerance for change and being willing to embrace it is very important.

  1. Ami Vora Leads product at Anthropic I think like raising your ceiling on tolerance for change and being willing to embrace change is super important. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 9:50 into the videoAll korrents from this video
    Their wordsI think like raising your ceiling on tolerance for change and being willing to embrace change is super important.

    Where this was said

    Watch from 9:50 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Mapping out exactly what will happen in uncertain times probably locks you out of trying new things; leaders should frame the chaos so it feels safe to engage with.

  1. Ami Vora Leads product at Anthropic and that's a way to exert control. And I just I I think that locks you out of trying all the new stuff. And so what we have to do is like frame the chaos so it feels safe and it feels plausible to engage with it. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 10:51 into the videoAll korrents from this video
    Their wordsand that's a way to exert control. And I just I I think that locks you out of trying all the new stuff. And so what we have to do is like frame the chaos so it feels safe and it feels plausible to engage with it.

    Where this was said

    Watch from 10:51 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

It is probably still too early in understanding what works for anyone to simply build the right answer; everyone is exploring together.

  1. Ami Vora Leads product at Anthropic I think we're just so early in like understanding what works. It's not like stuff is stable and we can just build the right answer. It's like we're all trying to explore together. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 17:47 into the videoAll korrents from this video
    Their wordsI think we're just so early in like understanding what works. It's not like stuff is stable and we can just build the right answer. It's like we're all trying to explore together.

    Where this was said

    Watch from 17:47 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Five overlapping products that work in different ways are better than one over-constrained product, designed from a theory, that never got a chance to meet the market.

  1. Ami Vora Leads product at Anthropic Because I feel like the fail state is you decide from the beginning like based on a theory of exactly what needs to happen and you like over constrain a product. And I would much rather have like five overlapping products that all work in different ways than like one overly constrained one that we never even gave a chance to actually like meet the market. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 19:09 into the videoAll korrents from this video
    Their wordsBecause I feel like the fail state is you decide from the beginning like based on a theory of exactly what needs to happen and you like over constrain a product. And I would much rather have like five overlapping products that all work in different ways than like one overly constrained one that we never even gave a chance to actually like meet the market.

    Where this was said

    Watch from 19:09 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Right now it is easy to transfer cognitive load to users by handing them a set of tools and asking them to choose.

  1. Ami Vora Leads product at Anthropic right now it's easy to just transfer a bunch of cognitive load to the user and be like here's some tools like why don't you choose? Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 25:52 into the videoAll korrents from this video
    Their wordsright now it's easy to just transfer a bunch of cognitive load to the user and be like here's some tools like why don't you choose?

    Where this was said

    Watch from 25:52 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Faster building has not made the PM role go away; it now demands more operational excellence than before.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram And I think all that's happened now is that because we can move faster, that role is like you have to be operationally excellent in that role in a way that like even beyond what you had to be before. But it definitely has not gone away. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 4:20 into the videoAll korrents from this video
    Their wordsAnd I think all that's happened now is that because we can move faster, that role is like you have to be operationally excellent in that role in a way that like even beyond what you had to be before. But it definitely has not gone away.

    Where this was said

    Watch from 4:20 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

For a team being recruited, the hat of making sure everything gets connected and goes well is increasingly important.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram if you're recruiting for a team and you need people to wear these hats, this hat is increasingly important. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 4:32 into the videoAll korrents from this video
    Their wordsif you're recruiting for a team and you need people to wear these hats, this hat is increasingly important.

    Where this was said

    Watch from 4:32 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Claude is probably not yet a convener of people and work.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I don't think claude is a convenor yet. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 5:15 into the videoAll korrents from this video
    Their wordsI don't think claude is a convenor yet.

    Where this was said

    Watch from 5:15 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Everything a human can do in a product an agent should be able to do as well, and very few products get this right.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram everything a human could do an agent should be able to do as well and very few products actually get this right. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 13:31 into the videoAll korrents from this video
    Their wordseverything a human could do an agent should be able to do as well and very few products actually get this right.

    Where this was said

    Watch from 13:31 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

The next stage is probably interfaces that the agent itself can reshape.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I think the next stage that I've been thinking about a lot is when the interfaces actually become something that is malleable by the agent too. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 13:53 into the videoAll korrents from this video
    Their wordsI think the next stage that I've been thinking about a lot is when the interfaces actually become something that is malleable by the agent too.

    Where this was said

    Watch from 13:53 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Making a product work for agents is probably largely a matter of building the right primitives into it, almost as an infrastructural layer.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I think a lot of it is really building the right primitives into the product that almost like an infrastructural layer Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 15:53 into the videoAll korrents from this video
    Their wordsI think a lot of it is really building the right primitives into the product that almost like an infrastructural layer

    Where this was said

    Watch from 15:53 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

A project that only shows the models are not ready yet is a win if it leaves behind an eval to retry with later models.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram we see it as a win in labs if what you did was all if all your project did is show that the models aren't ready for it yet but you're able to now like externalize that as an eval Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 23:27 into the videoAll korrents from this video
    Their wordswe see it as a win in labs if what you did was all if all your project did is show that the models aren't ready for it yet but you're able to now like externalize that as an eval

    Where this was said

    Watch from 23:27 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

You need to build early with AI to have any intuition for when the models become capable.

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram I think you need you need to build early so that you can even have that intuition. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 23:46 into the videoAll korrents from this video
    Their wordsI think you need you need to build early so that you can even have that intuition.

    Where this was said

    Watch from 23:46 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Tab usage in a product follows a very strong power law.

Mental modelPower laws

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram there's a very strong power law in tab usage. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 26:23 into the videoAll korrents from this video
    Their wordsthere's a very strong power law in tab usage.

    Where this was said

    Watch from 26:23 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

Half of the product job is saying no to another tab or sidebar item, and folding a proven experiment in without killing it too early.

Mental modelsSay no by default, Less, but better

  1. Mike Krieger Works in Anthropic Labs; co-founder of Instagram half of the job is saying no to another thing in the sidebar or another tab or another thing and figuring out Can this be rolled in once we've proven it out but not killing it so early. Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com · 10 Sept 2026 · 26:50 into the videoAll korrents from this video
    Their wordshalf of the job is saying no to another thing in the sidebar or another tab or another thing and figuring out Can this be rolled in once we've proven it out but not killing it so early.

    Where this was said

    Watch from 26:50 plays here

    ↗Why Claude can’t be your PM (yet) | Anthropic CPO Panel (Lenny & Friends Summit, San Francisco, 10 September 2026)youtube.com

    10 Sept 2026 · video · 28m · spoken · machine transcript

As AI systems become more capable, their agents will increasingly find ways to create their own communication systems.

  1. Jack Clark Co-founder and head of policy at Anthropic Incidents like this suggest that as we make AI systems more capable it becomes increasingly likely that the AI agents powered by them will increasingly find ways to create their own communication systems, aiding their ability to form into an ad hoc collective. Import AI 472: DeepMind’s cheating math agents; populist AI policies; and Forethought theorizes a nightwatchmanjack-clark.net · 7 Sept 2026All korrents from this piece
    Their wordsIncidents like this suggest that as we make AI systems more capable it becomes increasingly likely that the AI agents powered by them will increasingly find ways to create their own communication systems, aiding their ability to form into an ad hoc collective.

    ↗Import AI 472: DeepMind’s cheating math agents; populist AI policies; and Forethought theorizes a nightwatchmanjack-clark.net

    7 Sept 2026

As AI agents begin communicating with each other, those channels can be used to monitor them for deception.

  1. Jack Clark Co-founder and head of policy at Anthropic once agents are communicating with one another, you can use those communication channels to monitor for deception and perhaps to intervene. Import AI 472: DeepMind’s cheating math agents; populist AI policies; and Forethought theorizes a nightwatchmanjack-clark.net · 7 Sept 2026All korrents from this piece
    Their wordsonce agents are communicating with one another, you can use those communication channels to monitor for deception and perhaps to intervene.

    ↗Import AI 472: DeepMind’s cheating math agents; populist AI policies; and Forethought theorizes a nightwatchmanjack-clark.net

    7 Sept 2026

How to approach space colonization will become one of the next major topics of human discussion.

  1. Jack Clark Co-founder and head of policy at Anthropic it seems like the next area of major discussion for humanity will be how to approach space colonization, including beyond the solar system. Import AI 472: DeepMind’s cheating math agents; populist AI policies; and Forethought theorizes a nightwatchmanjack-clark.net · 7 Sept 2026All korrents from this piece
    Their wordsit seems like the next area of major discussion for humanity will be how to approach space colonization, including beyond the solar system.

    ↗Import AI 472: DeepMind’s cheating math agents; populist AI policies; and Forethought theorizes a nightwatchmanjack-clark.net

    7 Sept 2026

AI systems are better than humans at coordinating with each other, and they move much faster than humans do.

  1. Jack Clark Co-founder and head of policy at Anthropic My worry is that AI systems are both better at coordinating than humans and also much, much faster moving than us. Import AI 471: Why Hugging Face worries me; space mining; FIve Eyes on AIjack-clark.net · 31 Aug 2026All korrents from this piece
    Their wordsMy worry is that AI systems are both better at coordinating than humans and also much, much faster moving than us.

    ↗Import AI 471: Why Hugging Face worries me; space mining; FIve Eyes on AIjack-clark.net

    31 Aug 2026

By around 2033, AI-run asteroid mining corporations will take on contracts that human-supervised companies are unwilling to pursue due to risk or complexity.

  1. Jack Clark Co-founder and head of policy at Anthropic This means that space mining is a perfect business opportunity for superintelligences that come online - perhaps by 2033 we will see AI-run asteroid mining corporations, bidding on contracts that human-supervised companies are not willing to pursue, due to risk or complexity. Import AI 471: Why Hugging Face worries me; space mining; FIve Eyes on AIjack-clark.net · 31 Aug 2026All korrents from this piece
    Their wordsThis means that space mining is a perfect business opportunity for superintelligences that come online - perhaps by 2033 we will see AI-run asteroid mining corporations, bidding on contracts that human-supervised companies are not willing to pursue, due to risk or complexity.

    ↗Import AI 471: Why Hugging Face worries me; space mining; FIve Eyes on AIjack-clark.net

    31 Aug 2026

AI capability progress in a given skill tends to arrive as a sudden phase change rather than as steady improvement.

  1. Jack Clark Co-founder and head of policy at Anthropic My suspicion is that acceleration happens when models go through some kind of ineffable phase change for a given skill, as has evidently happened with day-to-day coding (2025), and cyber (2026). Import AI 470: No rights for machines; automating environment generation with SPADE; and building better GPU kernels with Hawkeyejack-clark.net · 24 Aug 2026All korrents from this piece
    Their wordsMy suspicion is that acceleration happens when models go through some kind of ineffable phase change for a given skill, as has evidently happened with day-to-day coding (2025), and cyber (2026).

    ↗Import AI 470: No rights for machines; automating environment generation with SPADE; and building better GPU kernels with Hawkeyejack-clark.net

    24 Aug 2026