Anthropic’s CPO on what comes next | Mike Krieger (co-founder of Instagram)

Episode Summary

Executive Summary: Mike Krieger explains how AI is reshaping product development at Anthropic: most code is now AI-written, causing new bottlenecks in strategy, alignment, and release processes rather than engineering throughput. He argues Claude is becoming a true thought partner, that product teams must focus on comprehensibility, strategy, and exposing AI’s potential, and that Anthropic should lean into its strengths with builders and enterprises rather than chase consumer mindshare directly.

Main Topics: AI is changing how products are built (Priority: 5/5): Anthropic has shifted to a world where AI writes most code, accelerating prototyping and coding while moving bottlenecks upstream into decision-making, alignment, and merge/release infrastructure. Claude as a strategic thought partner (Priority: 5/5): Krieger says Claude/Opus 4 has started offering genuinely novel strategic perspectives, changing his view of AI from capable assistant to collaborator with independent-sounding insight. Product strategy in the AI era (Priority: 5/5): Traditional product teams still matter, but their highest leverage is in strategy, making AI capabilities comprehensible, and helping users see what’s possible rather than merely building features. Anthropic’s positioning versus ChatGPT/OpenAI (Priority: 4/5): He acknowledges ChatGPT’s consumer mindshare lead, but argues Anthropic should embrace its builder brand and focus on areas where it is uniquely strong instead of mimicking consumer-first competitors. MCP and the future of context (Priority: 5/5): Krieger frames MCP as the missing middle layer between model intelligence and UI—bringing the right context, memory, and integrations so agents can actually do useful work across tools. Artifact, founder lessons, and shutdowns (Priority: 3/5): He reflects on shutting down Artifact as a rational but difficult decision driven by weak growth, broken mobile-web economics, remote culture limits, and opportunity cost as AI accelerated. Skills and values for the next generation (Priority: 4/5): For kids and users, he emphasizes curiosity, scientific inquiry, independent thinking, and using AI as augmentation rather than outsourcing cognition.

Key Arguments: AI has moved product development bottlenecks away from engineering speed and toward strategy, alignment, review, and release orchestration. Claude is no longer just a generator; with Opus 4 it can produce genuinely new strategic angles and act like a collaborator. Product teams remain essential for making AI systems understandable, choosing where to play, and showing users what they can do. Anthropic should not over-optimize for consumer hype; it should lean into builders, developers, and enterprise use cases where it has a durable edge. The strongest startup opportunities are in vertical domains, differentiated workflows, or new AI form factors that incumbents cannot easily absorb. MCP matters because context and memory are as important as raw model intelligence; reusable integrations unlock much better agentic behavior. Successful AI builders push the model to its limits, build repeatable evaluation loops, and adapt as models improve. Shutting down Artifact was hard but justified because growth, distribution, and product-market fit were not compounding enough to warrant continuing. The most important human skills in an AI-heavy world are curiosity, inquiry, and independent thought rather than blind reliance on AI outputs.

Data Points: Code written by AI at Anthropic: roughly 90% - Krieger said nearly all code is now AI-generated, especially in teams like Cloud Code. Cloud Code codebase written by Cloud Code: 95%+ (estimate) - He estimated Cloud Code itself is almost entirely written by AI, though he deferred to the tech leads for exact figures. SWE-bench performance: about 72% - Krieger cited the current model performance level after Dario Amodei’s earlier prediction about rapid improvement. Earlier SWE-bench prediction: 90% by end of 2025 - Referenced as a forecast Dario made when the benchmark was around 50%. Starting benchmark level mentioned: 50% - The point at which the 90% end-of-2025 prediction was made. Global GDP through Stripe: 1.3% - Ad read citing Stripe’s scale last year. Global GDP through Stripe in dollars: over $1.4 trillion - Ad read citing the amount of money flowing through Stripe. Hertz payment authorization lift: 4% - Example from Stripe’s customer results after migration. Forbes revenue lift: 23% - Example from Stripe’s subscription management migration. Anthropic year on the job: just over 1 year - Krieger discussed what changed since joining Anthropic.

Pivotal Quotes: "90% of your code roughly is written by AI now." — Interviewer/intro: Sets up the central theme of the conversation about AI-native product development at Anthropic. "I think there’s room for several generationally important companies to be built in AI right now." — Mike Krieger: Explains why Anthropic should focus on its unique strengths rather than directly chasing consumer mindshare. "What are you trying to do with cloud today that cloud is failing at is the most useful input I could possibly have." — Mike Krieger: His closing request to listeners for feedback to improve Claude.

Implications: AI companies will need to redesign product orgs around strategy, context, and evaluation—not just shipping code faster. The winners will likely be those that pair strong models with workflow-specific context, durable distribution, and clear user value.