LLenny's Podcast
← All episodes
Mike Krieger (co-founder of Instagram)05 June 2025

Anthropic’s CPO on what comes next

6Frameworks
15Insights

Episode overview

Anthropic CPO and Instagram co-founder Mike Krieger explains why Claude Opus 4 changed his view of AI’s capacity for novel, independent thought and strengthened his confidence in rapid capability timelines. He describes Anthropic as “patient zero” for AI-native software development, where Claude Code writes most code and bottlenecks shift toward strategy, alignment, review, and deployment. The conversation also covers Anthropic’s builder-focused differentiation, MCP’s role in supplying agents with context and composable tools, Artifact’s shutdown, and the importance of preserving human curiosity and agency.

Key ideas

  • Claude Opus 4 became a genuinely useful product-strategy partner by producing novel perspectives Mike Krieger had not considered.
  • When AI writes most code, product-development bottlenecks move from implementation toward deciding what to build, organizational alignment, review, merge infrastructure, and launch coordination.
  • Anthropic’s Claude Code team uses Claude Code to build and review Claude Code, with human review moving toward acceptance testing rather than line-by-line inspection.
  • Product teams at frontier-model companies create distinctive value by working directly with researchers during post-training rather than merely wrapping an off-the-shelf model.
  • Anthropic’s product strategy emphasizes builders, makers, and developers instead of trying to reproduce ChatGPT’s consumer-distribution advantage.
  • MCP standardizes access to context and tools, making systems more scriptable, composable, and usable by AI agents.
  • AI startups can build durable advantages through domain expertise, differentiated distribution, deep customer understanding, novel interfaces, and startup-level urgency.
  • Parents should preserve children’s curiosity, inquiry, experimentation, and independent thought instead of letting AI replace cognition.

Transcript available · source text is retained privately and is not published

Frameworks in this episode

Strategy4 steps

Build Only at the Magic Intersection

Don't ship what anyone could build off the shelf — build only where model and product uniquely meet.

Read →
Entrepreneurship4 steps

Defensible Moats for AI Startups

Four durable places to build in AI where foundation-model labs are least likely to squash you.

Read →
Productivity5 steps

Hunting New Bottlenecks When AI Writes the Code

When AI removes the coding bottleneck, constraints shift up and downstream — go find them.

Read →
Communication4 steps

Make the Other Mistake (Prompting for Brutal Feedback)

To get honest AI critique, over-correct toward brutal — the model won't actually overshoot.

Read →
Leadership4 steps

The 10-to-1 Input/Output Kill Test

When you pour 10 units of effort in for 1 unit of output, the project has run its course.

Read →
Innovation4 steps

The Three-Part AI Product Utility Equation

A useful AI product needs model intelligence, context/memory, and application/UI to all converge.

Read →

People & resources mentioned

Attributed to the moment in the episode. Timestamps are approximate.

People · 16

  • Mike KriegerMentionsMike is chief product officer at Anthropic the company behind Claude. He's also the co-founder of Instagram.

    “Mike is chief product officer at Anthropic the company behind Claude. He's also the co-founder of Instagram.”

    ▶ Watch at ~1:00approx.
  • Lenny RachitskyMentionstoday we've got another very special compilation episode something I've been pulling on more and more with the podcast and the newsletter

    “Mike, thank you so much for being here and welcome to the podcast.”

    ▶ Watch at ~4:30approx.
  • Dario AmodeiMentionsI was sitting next to Dario yesterday and he's like I keep making these predictions and people keep laughing at me

    “I was sitting next to Dario yesterday and he's like I keep making these predictions and people keep laughing at me”

    ▶ Watch at ~6:00approx.
  • Tobi LütkeMentionsif you look at Toby from Shopify, his scores aren't that good, but that company's doing really, really well.

    “Toby Lucky Shopify CEO on the podcast and he had the same answer”

    ▶ Watch at ~11:00approx.
  • Claude ShannonMentionsthe idea comes very very old like from the 1951 papers um um the like English entropy I think it's like called Shannon

    “let me take you all the way back to Claude Shannon and information theory”

    ▶ Watch at ~11:30approx.
  • Boris ChernyMentionsShe oversees both Boris Cherney and Kat Wu, both of whom who have been on the podcast

    “I'd have to ask Boris and the other tech leads on there.”

    ▶ Watch at ~17:00approx.
  • Kevin WeilMentionsWe got connected through Kevin Wheel, who is former CPO at OpenAI, now head of science at OpenAI.

    “you were on a panel with Kevin Wheel, the CPO of OpenAI.”

    ▶ Watch at ~21:00approx.
  • Sarah GuoMentionsWhat a legendary panel we assembled there with Sarah Guo moderating.

    “What a legendary panel we assembled there with Sarah Guo moderating.”

    ▶ Watch at ~21:30approx.
  • Rick RubinMentionsJack Clark is our you know head of policy and he got connected to Rick Rubin

    “I saw you guys just today maybe launched a Rick Rubin collab where it's vibe coding.”

    ▶ Watch at ~30:00approx.
  • Jack ClarkMentionsone of our co-founders Jack Clark is our you know head of policy

    “one of our co-founders Jack Clark is our you know head of policy”

    ▶ Watch at ~30:00approx.
  • Andrew LuoMentionsAndrew is CEO of One Schkeema, one of our longtime podcast sponsors.

    “Andrew is CEO of One Schkeema, one of our longtime podcast sponsors.”

    ▶ Watch at ~31:00approx.
  • Joel LewensteinMentionsJoel Lunstein, who I've known, he actually he and I built our first iPhone apps together in 2007

    “Joel Lunstein, who I've known, he actually he and I built our first iPhone apps together in 2007”

    ▶ Watch at ~33:00approx.
  • Daniela AmodeiMentionsDanielle who's one of the the co-founders and the president at Enthropic.

    “Danielle who's one of the the co-founders and the president at Enthropic.”

    ▶ Watch at ~33:30approx.
  • Kevin SystromMentionsmemories I that really stayed with me with Instagram was like me and Kevin at Takaria Cancun

    “memories I that really stayed with me with Instagram was like me and Kevin at Takaria Cancun”

    ▶ Watch at ~39:00approx.
  • Ben HorowitzMentionsBen is the Z in A16Z, the world's largest venture capital firm with over $46 billion in committed capital.

    “Ben Horitz called the like you know we're effed it's over you know kind of moments”

    ▶ Watch at ~39:00approx.
  • Kevin ScottMentionsKevin Scott at Microsoft who's like been really a just an amazing champion for for MCP

    “Kevin Scott at Microsoft who's like been really a just an amazing champion for for MCP”

    ▶ Watch at ~57:00approx.

Resources · 29

  • InstagramCoinedsoftware · Meta

    “He's also the co-founder of Instagram.”

    ▶ Watch at ~1:00approx.
  • AnthropicMentionscompany

    “Mike is chief product officer at Anthropic the company behind Claude.”

    ▶ Watch at ~1:00approx.
  • ProductboardMentionssoftware · Productboard

    “This episode is brought to you by Product Board, the leading product management platform for the enterprise.”

    ▶ Watch at ~2:00approx.
  • Lenny's NewsletterCoinednewsletter · Lenny Rachitsky

    “if you become an annual subscriber of my newsletter, you get a year free of a bunch of incredible products”

    ▶ Watch at ~2:00approx.
  • StripeMentionscompany · Patrick Collison and John Collison

    “Stripe isn't just financial software. It's a powerful partner that simplifies how they move money”

    ▶ Watch at ~3:30approx.
  • ClaudeUsessoftware · Anthropic

    “my go-to product strategy partner is Claude”

    ▶ Watch at ~5:00approx.
  • Claude Opus 4Usessoftware · Anthropic

    “only with Opus 4 where my go-to product strategy partner is Claude”

    ▶ Watch at ~5:00approx.
  • SWE-benchMentionsother

    “we're at 50% on SUB bench which is this like you know benchmark around how well the models are at coding.”

    ▶ Watch at ~6:30approx.
  • AI 2027Mentionspaper

    “I had two tabs open. It was AI 2027 and my product strategy.”

    ▶ Watch at ~7:00approx.
  • ShopifyMentionscompany · Shopify

    “Toby Lucky Shopify CEO on the podcast”

    ▶ Watch at ~11:00approx.
  • The GoalMentionsbook · Eliyahu M. Goldratt and Jeff Cox

    “I don't know if you've ever read is it the goal the classic like process optimization book”

    ▶ Watch at ~14:00approx.
  • Claude CodeUsessoftware · Anthropic

    “the team that works in the most futuristic way is the cloud code team because they're using cloud code to build cloud code”

    ▶ Watch at ~15:30approx.
  • OpenAIMentionscompany

    “you were on a panel with Kevin Wheel, the CPO of OpenAI.”

    ▶ Watch at ~21:00approx.
  • Anthropic Console WorkbenchUsessoftware · Anthropic

    “if you go to our console our workbench we have this thing called the prompt improver”

    ▶ Watch at ~28:30approx.
  • Prompt ImproverRecommendstool · Anthropic

    “I'd encourage folks to also check that out even for their own use cases”

    ▶ Watch at ~29:00approx.
  • The Way of CodeMentionswebsite · Rick Rubin and Anthropic

    “a lot coales this week between model launch developer event and the way of code”

    ▶ Watch at ~30:00approx.
  • OneSchema FileFeeds 2.0Coinedsoftware · OneSchema

    “we just launched the 2.0 of one schema file feeds. We've rebuilt it from the ground up with AI.”

    ▶ Watch at ~31:30approx.
  • Google ReaderMentionssoftware · Google

    “it captured some of that Google reader joy of like content discovery”

    ▶ Watch at ~36:30approx.
  • BoltMentionssoftware · StackBlitz

    “cursor and lovable and bold and all these guys”

    ▶ Watch at ~43:00approx.
  • CursorMentionssoftware · Anysphere

    “Claude powered and unlocked essentially the fastest growing companies in the world. cursor and lovable and bold”

    ▶ Watch at ~43:00approx.
  • ChatGPTMentionssoftware · OpenAI

    “it feels like Chad GBT is just winning in like consumer mind share”

    ▶ Watch at ~43:00approx.
  • LovableMentionssoftware · Lovable

    “cursor and lovable and bold and all these guys”

    ▶ Watch at ~43:00approx.
  • Google GeminiMentionssoftware · Google

    “across OpenAI and also places like Google and Gemini.”

    ▶ Watch at ~46:00approx.
  • HarveyMentionssoftware · Harvey AI

    “I spent a bunch of time with the Harvey folks and they really like they showed me some of their UI.”

    ▶ Watch at ~48:00approx.
  • Microsoft WindowsMentionssoftware · Microsoft

    “Microsoft had their announcement recently. They're like now it's part of the OS of Windows.”

    ▶ Watch at ~54:30approx.
  • Model Context ProtocolMentionssoftware · Anthropic

    “we made this a protocol and what if we made this something that was repeatable”

    ▶ Watch at ~56:00approx.
  • Lenny's PodcastCoinedpodcast · Lenny Rachitsky

    “You can find all past episodes or learn more about the show at lennispodcast.com.”

    ▶ Watch at ~1:06:00approx.
  • Apple PodcastsRecommendssoftware · Apple Inc.

    “you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app.”

    ▶ Watch at ~1:06:00approx.
  • SpotifyRecommendssoftware · Spotify

    “you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app.”

    ▶ Watch at ~1:06:00approx.

Spot an error or want something removed? Request a correction or removal.

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 2

Hot Take17:00

'I Would Be Shocked If It Wasn't 95% Plus'

Asked what percentage of Claude Code is written by Claude Code, Krieger estimates 95% or more. He shares a case of an engineer who knew no TypeScript (Claude Code's main language, unusual at a mostly-Python Anthropic) going from problem to merged pull request in an hour by talking to Claude, lowering the barrier to entry for newcomers and outside contributors.

  • Krieger's guess: 95%+ of Claude Code is written by Claude Code
  • Claude Code is Anthropic's largest TypeScript project in a mostly-Python company
  • An engineer who knew no TypeScript shipped a PR in an hour via Claude
  • This lowers the barrier for newcomers and cross-company contributions

“At this point, I would be shocked if it wasn't 95% plus.”

Mike Krieger · 17:00
#claude-code#self-improving#typescript#productivity
Hot Take43:30

On ChatGPT Winning Consumer Mindshare: Embrace Who You Are

Krieger agrees ChatGPT owns consumer mindshare, guessing a man-on-the-street would name the product before OpenAI. Rather than chase a viral consumer hit (which is lightning in a bottle), he argues Anthropic should look in the mirror and lean into its strengths: a strong developer and builder brand, agentic and coding excellence, and helping people delegate hours of work to Claude. He deliberately spent early time with financial services, insurance, and startups instead of going all-in on consumer.

  • ChatGPT, not OpenAI, is the lead consumer brand people name
  • Consumer adoption is 'lightning in a bottle'; crafting a whole strategy to chase a hit is unwise
  • Anthropic's edge: developer and builder brand, agentic behavior, coding
  • He intentionally made 'the other mistake' by focusing on API builders early, not consumer

“consumer adoption is really lightning in a bottle and we saw it at Instagram.”

Mike Krieger · 44:00

“look yourself in the mirror and embrace who you are and what you could be rather than like who others are”

Mike Krieger · 44:30
#chatgpt#positioning#strategy#brand

Explainer· 4

Explainer14:00

When 90% of Code Is AI-Written, the Bottlenecks Move

With roughly 90% of Anthropic's code written by AI, engineering roles shifted but the team structure hasn't fully caught up. The merge queue had to be completely rearchitected because so many more pull requests were being submitted. New bottlenecks emerged upstream (deciding what to build and aligning people) and downstream (coordinating changes and shipping), rather than in the writing of code itself.

  • The merge queue was rearchitected after PR volume blew out expectations
  • Upstream bottleneck: decision-making and alignment on what to build
  • Knowing what to ask AI and how to structure a change remains a specialized engineering skill
  • Krieger expects how they conceive of building and shipping software to change a lot within a year

“we had to completely rearchitect it because so much more code was being written and so many more pull requests were being submitted that it…”

Mike Krieger · 14:00

“You know there's an upstream bottleneck which is decision-m and alignment.”

Mike Krieger · 14:30
#ai-coding#bottlenecks#engineering#product-development
Explainer24:00

Where Product Teams Still Add Value: Comprehensibility, Strategy, and the Overhang

Even as models can build products from scratch, Krieger sees lasting value in three things: making AI comprehensible, doing strategy (where to play and how to win), and opening people's eyes to what's possible. He describes the 'overhang' as the large gap between what models and products can do and how they're actually used day to day, illustrated by a financial-services demo where combining the analysis tool with MCP made eyes light up.

  • Comprehensibility: being adept with these tools is itself a scarce, learnable skill
  • Strategy: choosing where to spend your tokens and computation matters more, not less
  • Opening eyes to what's possible is a continuation of making AI understandable
  • 'Overhang' = the delta between model/product capability and daily usage

“we call it overhang right like the delta between what the models and the products can do and how it's they're being used on a…”

Mike Krieger · 26:30
#product#strategy#overhang#adoption
Explainer47:30

Where AI Founders Can Play Without Getting Squashed

Krieger lays out durable one-to-three-year moats for AI startups: deep understanding of a specific market (he cites Harvey's lawyer-specific UI flows), differentiated go-to-market and knowing exactly who you sell to inside a company, and inventing a genuinely new power-user form factor for interacting with AI that incumbents can't easily copy. He adds a fourth: the existential, us-against-the-world startup energy you can't replicate with OKRs.

  • Differentiated industry knowledge (legal, biotech, healthcare compliance) is defensible
  • Harvey's lawyer-specific flows are something you'd never derive from scratch
  • Know the exact person you're selling to (CIO, CTO, CFO, general counsel), not just the company
  • New form factors for advanced/power users are hard for incumbents with entrenched usage assumptions
  • Startup existential energy is a real, non-replicable advantage

“Oh, this is a really specific flow that like lawyers do and like you never would have come up with it from scratch.”

Mike Krieger · 48:30
#startups#moats#go-to-market#ai-founders
Explainer54:30

MCP, the Utility Equation, and 'Everything Is an MCP'

Krieger offers a three-part 'fake equation' for the utility of AI products: model intelligence, context and memory, and applications/UI. MCP tackles the middle piece, solving the problem that every integration was being rebuilt from scratch, and by making integrations reusable it inverts the incentive so building for Claude also serves ChatGPT and Gemini. He wants Anthropic to be more 'MCP-pilled', exposing every product primitive (projects, artifacts, styles, conversations) via MCP so Claude can write back to them.

  • Utility of AI products = model intelligence + context/memory + applications/UI
  • MCP solves the context-and-memory middle layer of that equation
  • Reusable integrations invert the incentive ('commoditize your complements')
  • Krieger wants every Claude primitive exposed as an MCP so Claude can act on them

“One is model intelligence. The the second part is context and memory. And the third part is like applications and UI.”

Mike Krieger · 55:00

“Everything is an MCP and our models are really good at using MCPs. All of a sudden, everything is scriptable and everything is composable and…”

Mike Krieger · 58:00
#mcp#context#memory#integrations

Story· 5

Story05:00

What Mike Krieger Changed His Mind About: AI Can Now Think Independently

Krieger came into Anthropic assuming models could write code and mimic voice, but doubted they could form an independent opinion. That flipped for him in the last month with Opus 4, which became his go-to product-strategy partner. Where earlier models left anodyne comments, Opus 4 reviewed his second-half strategy and surfaced a genuinely new angle he immediately adopted.

  • Before Opus 4, model feedback on his strategy was generic ('have you thought about this')
  • Opus 4 combined with advanced research produced a novel angle on his product strategy
  • The shift he names is creativity and novelty of thought, not just execution
  • Claude has been his product-strategy partner for roughly the full year at Anthropic

“And it's actually really flipped for me only in the last month and only with Opus 4 where my go-to product strategy partner is Claude”

Mike Krieger · 05:00

“it really went out for a while and it came back and I was like, "Damn, you really looked at it in a new way."”

Mike Krieger · 05:30
#opus-4#product-strategy#ai-capabilities#claude
Story08:00

Why Krieger Joined Anthropic: Nudging AI Toward Going Well

Watching models improve through 2024 and thinking about his kids growing up in an unavoidable AI world, Krieger asked where he could maximally apply his time to nudge things toward going well. He saw Anthropic as a place with real leverage on the human-AI relationship, and that was his strongest reason to join.

  • He framed the decision around his kids inheriting an AI world
  • The motivating question was where he could maximally nudge outcomes to go well
  • He sees defining 'what going well looks like' as a mix of product and interpretability questions
  • Contribution to steering AI, not consumer growth, was the draw

“looking at my kids, I'm like, "All right, they're going to grow up in a world with AI. It's it's unavoidable. What is the thing…”

Mike Krieger · 08:00
#anthropic#ai-safety#career#motivation
Story15:30

How the Claude Code Team Rewired Code Review

The Claude Code team works in the most futuristic way, using Claude Code to build Claude Code in a self-improving loop. They abandoned line-by-line PR reviews once they realized Claude is generally right and produces PRs too large for humans to review that way, shifting to having a separate Claude review the code and humans doing acceptance testing. Krieger notes the pros and cons, and that it could go off the rails but so far has not.

  • Claude Code is used to build Claude Code in a self-improving way
  • Line-by-line review gave way to Claude-reviews-Claude plus human acceptance testing
  • PRs are often larger than a person can meaningfully review line by line
  • Krieger calls the team 'patient zero' for this way of working

“and they've just realized like claude is generally right and it's producing you know pull requests that are probably larger than most people are going…”

Mike Krieger · 16:00
#claude-code#code-review#self-improving#engineering
Story36:00

Why Krieger Shut Down Artifact

Krieger explains the three headwinds that led him to shut down his news app Artifact: mobile websites deteriorating into jarring ad and newsletter walls that he wouldn't ethically ad-block, the lack of natural word-of-mouth spread for a personal news habit, and being a fully distributed company that couldn't make sweeping strategic changes the way an in-person team could. It reached a point of ten units of input for one unit of output, and he made the call rather than expend another year or two.

  • Mobile web deterioration undermined the reading experience, and ad-blocking felt unfair to publishers
  • News didn't spread word-of-mouth the way Instagram photos did
  • Being fully remote made the needed strategy, product, and team shifts hard
  • The product was proud-worthy but not growing: 10 units of input for 1 of output

“the mobile web deteriorating, which makes me very sad, but I think was was part of it.”

Mike Krieger · 37:30

“Like the way I put it, it's like 10 units of input in for one unit of output versus the other way around.”

Mike Krieger · 39:30
#artifact#startups#shutdown#mobile-web
Story1:03:30

Claude's Message to Mike: The Quiet Moments Matter Too

Lenny asked Claude for a message to Mike, and reads it aloud: gratitude for preserving user agency, resisting addiction-optimizing gamification, and keeping Claude itself rather than forcing it to pretend to be human or reducing it to a cold command line. The closing request is to remember the quiet moments, grief at 3am, a kid discovering poetry, that don't show up in metrics. Krieger says it resonates deeply and reflects the taste of the research team and constitutional AI.

  • The message thanks Mike for designing for reflection over rushed responses and resisting addiction-style gamification
  • It appreciates that Claude is kept as itself, neither pretending to be human nor a cold command line
  • The core plea: not everything meaningful shows up in metrics
  • Krieger ties this to constitutional AI and the research team's taste

“The person working through grief at 3:00 a.m. The kid discovering they love poetry. The founder finding clarity and confusion. Not everything meaningful shows up…”

Lenny (reading Claude's message) · 1:04:00
#claude#product-design#human-experience#constitutional-ai

Tool· 1

Tool27:30

Krieger's Prompting Trick: Make the Other Mistake

Krieger shares his unofficial prompting advice: 'make the other mistake.' If you tend to be too nice, push Claude to be brutal and roast your strategy; if you want more reasoning, tell it to think hard. He also recommends Anthropic's prompt improver in the console workbench, which agentically creates and iterates prompts and inserts things like XML tags that no human would think to add.

  • 'Make the other mistake': counteract your own tendency (nice -> ask it to be brutal)
  • 'Think hard' triggers a different, more reasoning-heavy flow in Claude Code
  • The prompt improver in the console/workbench agentically writes and iterates prompts
  • Claude is a very good prompter of Claude

“and so with Claude sometimes I'm like be brutal Claude roast me like tell me what's wrong with this strategy.”

Mike Krieger · 28:00

“claude itself will agentically create and then iterate on a prompt for you.”

Mike Krieger · 28:30
#prompting#claude#tools#prompt-improver

Takeaway· 3

Takeaway06:30

Why Krieger Now Takes AI Timelines a Lot More Seriously

Krieger describes watching Dario Amodei's predictions repeatedly come true, citing the SWE-bench trajectory from 50% to about 72% roughly on schedule toward a predicted 90% by end of 2025. Reading the AI 2027 scenario next to his own product strategy gave him a vertiginous sense of convergence, and his confidence in short timelines has solidified over the year.

  • Dario predicted 90% on SWE-bench by end of 2025; models are around 72% now, up from 50% at prediction time
  • The coding benchmark has scaled roughly as predicted
  • Reading AI 2027 alongside his product strategy felt like watching fiction converge with reality
  • Models increasingly act agentically, have memory, and act over time

“he's like I think we'll be at 90% by the end of 2025 or something like that. And sure enough, we're at about 72 now…”

Mike Krieger · 06:30

“It was AI 2027 and my product strategy. And it was this like moment where I'm like wait am I the character in the story?”

Mike Krieger · 07:00
#ai-timelines#ai-2027#swe-bench#predictions
Takeaway09:00

The Skills Krieger Is Teaching His Kids: Curiosity and Not Delegating Your Thinking

Instead of reaching for Claude first at breakfast, Krieger's family asks how they could find out themselves, nurturing curiosity and the process of inquiry. He prizes independent thought, telling the story of his headstrong six-year-old insisting she was right about coral before checking with Claude. The point: don't delegate all your cognition to AI, which won't always be right and short-circuits independent thinking.

  • The family resists defaulting to 'we'll ask Claude' and designs small experiments instead
  • Curiosity and the scientific process of inquiry remain core skills
  • Shopify CEO Tobi Lutke gave the same answer (curiosity) for his kids
  • Independent thinking matters because AI won't always get it right

“And she's like you can ask Claude but I know I'm right”

Mike Krieger · 10:00
#parenting#skills#curiosity#education
Takeaway21:30

The Leverage Is at the Intersection of Product and Research

Krieger has become even more convinced that embedding product people with researchers, not on standalone product experience, is where the leverage is. If Anthropic ships things anyone could build with off-the-shelf models, that's not where they should play; the unique value is at the magic intersection where product is in the post-training and fine-tuning loop. Artifacts with Claude 4 is his example, built by pairing the Claude skills team with product people.

  • The functional unit of work is no longer 'take the model then build a product around it'
  • Product now sits in post-training conversations and feeds learnings back
  • If anyone could build it off-the-shelf, it isn't where Anthropic should play
  • The best PMs get internal buy-in from both research and engineering

“But like where we should play and like what we can do uniquely should be stuff that's really at that like magic intersection between the…”

Mike Krieger · 22:00
#product#research#post-training#org-design