The AI-Native Development Loop
by Mike Krieger • Chief Product Officer (CPO) at Anthropic at Anthropic
Co-founder and former CTO of Instagram. Currently leading product at Anthropic (makers of Claude), where he oversees the development of AI models and products like Claude and Artifacts.
🎙️ Episode Context
Mike Krieger discusses the radical shift in software development at Anthropic, where 90% of code is now written by AI. He explores how product management evolves when engineering barriers vanish, the strategic importance of MCP (Model Context Protocol), and how to compete as a 'challenger' brand against OpenAI. He also shares lessons from shutting down his news app, Artifact, and advice for AI founders on avoiding being crushed by foundational models.
Problem It Solves
Adapting product development processes when AI generates code faster than humans can review or deploy it.
Framework Overview
When AI writes 90% of the code, the traditional PM-Designer-Engineer handover breaks down. The bottleneck shifts from 'writing code' to 'decision making' (upstream) and 'merge queues' (downstream). Teams must re-architect their infrastructure and review processes to handle this velocity.
🧠 Framework Structure
Shift Prototyping Left: PMs and desig...
Re-architect Critical Paths: Optimize...
AI-Assisted Review: Use AI agents to ...
When to Use
When integrating AI coding agents (like Claude Code or Cursor) into a team's workflow and noticing traffic jams in deployment.
Common Mistakes
Treating AI-generated code with the same slow, manual review process used for human code, causing a massive backlog.
Real World Example
The Claude Code team uses Claude Code to build itself (95% of code). They realized human line-by-line review wasn't scaling, so they started using a separate Claude instance to review the PRs, shifting humans to high-level acceptance testing.
We really rapidly became bottlenecked on other things like our merge queue... Over half of our pull requests are Claude Code generated. Probably at this point it's probably over 70%... or 90%.
— Mike Krieger