Mike Krieger

Claude code reached roughly 95% AI-generated Typescript, enabling people unfamiliar with the language to ship fixes quickly and encouraging company-wide contributions.

Mike stresses choosing clear where-to-play and how-to-win decisions to focus limited compute and avoid confusing positioning.

He explains that good product design resolves the tension between the model taking over and engaging users enough to keep them in control.

Mike shares that Dario's accurate track record on capability predictions has made him take near-term AI timelines far more seriously.

Instagram grew organically because users reposted their photos on other networks, a viral mechanic Artifact could not ethically replicate.

Mike warns against optimising for likability or time-spent and instead focuses on whether Claude actually helps people accomplish work.

The Claude code team lets one Claude generate pull requests and another Claude review them while humans do acceptance tests, greatly accelerating merges.

Mike argues product managers remain vital for empathy, comprehension, and strategic focus even as AI automates execution.

Mike joined Anthropic to maximise his ability to nudge AI development toward better futures for his kids and society.

Anthropic broadens early-access so customers push models to the limit, capture traces, and validate gains with every new release.

Teams need a shared framework for defining what “going well” with human-AI relationships looks like and to guide research, product, and monitoring along the way.

By turning integrations into the open MCP protocol that any model can use, Anthropic encourages developers to build once and scale across Claude, ChatGPT and Gemini.

Mike presents a mental model where model intelligence, context & memory, and applications & UI must converge to make AI products genuinely useful.