Author: Mike Krieger
Type: podcast
Published: 2025-06-05
Status: unread
Tags: source, ai-pm, claude-added

Anthropic’s CPO on What Comes Next | Mike Krieger (Co-Founder of Instagram)

By: Mike Krieger Host: Lenny Rachitsky Source: Lenny’s Newsletter / Lenny’s Podcast Type: podcast

Summary

Lenny’s Podcast interview with Mike Krieger (CPO Anthropic, co-founder Instagram). 90-95% of code for some Anthropic products is now AI-written — the bottleneck has shifted from engineering to decision-making and merge queues, and Krieger expects most companies to reach this point within a year. Claude Opus 4 has crossed a threshold from “helpful but obvious” to “genuinely creative” as a strategy partner. Anthropic’s competitive strategy: don’t try to beat ChatGPT at consumer mindshare — lean into developers, builders, agentic behavior, and coding. Product teams embedded directly with AI researchers drive 10x more value than those just building UX on top of models. MCP is potentially the most important thing Anthropic has shipped — already the fastest-growing standard in tech history with Microsoft integrating into Windows. The vision: everything becomes an MCP endpoint, making the entire digital world scriptable by AI agents. Artifact (Krieger’s post-Instagram startup) failed despite being loved — mobile web is broken, news doesn’t spread virally, and remote work made pivots nearly impossible. Key lesson: know when to call it.

Key Ideas Extracted

  • Bottleneck shift from engineering to decision-making: When 90%+ of code is AI-written, the constraint moves to what-to-build decisions and merge queue management — this is happening faster than expected
  • Claude as genuine strategy partner: Opus 4 crossed the threshold from “helpful but obvious” to providing novel angles the CPO hadn’t considered — qualitative shift happened within a single month
  • Embed PMs with AI researchers, not just on UX: Product teams working directly with research on post-training and fine-tuning drive 10x value vs. building features anyone could build with public APIs
  • MCP as universal scriptability layer: Vision is everything becomes an MCP endpoint — the entire digital world becomes composable by AI agents, similar to how APIs made web services composable
  • Differentiate, don’t imitate: Anthropic leans into developers/builders rather than competing for ChatGPT’s consumer mindshare — “embrace who you are and what you could be”
  • AI startup moats: Deep domain expertise (Harvey in legal), differentiated GTM with specific customer knowledge, or completely new interaction paradigms — don’t underestimate startup urgency as an advantage
  • Value-delivered metrics over engagement: When one good Claude conversation might be 2 messages or 200, traditional engagement metrics mislead — measure actual value delivered (25 min vs. 6 hours for prototyping)
  • Work at the edge of model capabilities: Best API customers pushed limits with earlier models, hit walls, then were ready when new capabilities emerged — Cursor and Lovable took off when Claude 3.5 launched because they’d been testing boundaries
  • Artifact failure lessons: 10 units of input for 1 unit of output — mobile web is broken, news doesn’t spread virally, remote work made pivots impossible; sometimes shutting down is the right call
  • Skills for AI era: Curiosity, scientific thinking, maintaining independent thought — don’t outsource all cognition to AI

Notes

  • Published Jun 5, 2025 on Lenny’s Newsletter. Podcast episode (~66 min).
  • Sponsors: Productboard, Stripe, OneSchema
  • Mike Krieger background: co-founder Instagram, then Artifact (AI news app, shut down), then CPO Anthropic (joined 2024)
  • Referenced people: Dario Amodei, Joel Lewenstein, Daniela Amodei, Boris Cherny, Gunnar Gray, Jack Clark, Kevin Scott
  • Cross-references: Claire Vo / How I AI, Cursor (Michael Truell), Lovable (Anton Osika), Bolt (Eric Simons), OpenAI CPO Kevin Weil
  • Recommended books: The Goal, The Way of the Code: The Timeless Art of Vibe Coding, The Hard Thing About Hard Things
  • Highly relevant to ai-pm taxonomy: ai-adoption, MCP/agents, compound-engineering, product-metrics

Raw Content

Re-scraped from Lenny’s Newsletter 2026-02-15. Full article content captured in Summary and Key Ideas above.


This site uses Just the Docs, a documentation theme for Jekyll.