AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Inside Anthropic Labs: Mike Krieger on Claude Fable 5

· AI & I

Mike Krieger, head of Anthropic Labs and Instagram co-founder, joins host Dan to discuss hands-on experience using Anthropic's new Mythos-class model (referred to as "Fable 5") in daily work at Anthropic Labs. Krieger describes how the model fundamentally changes workflows: he now delegates complex multi-hour or overnight tasks, trusts the model to handle obstacles like failed remote services on its own, and runs multiple concurrent Claude Code sessions in parallel. He emphasizes that software engineering is being transformed rather than eliminated — the craft has shifted toward architectural planning, ownership/DRI-ship, and verification rather than line-by-line coding. Krieger shares concrete demonstrations, including a personal weekend-built media tracker app with self-modifying software (where the agent can edit the app from within itself) and a dynamic workflow that translated a complex Python codebase to TypeScript over a weekend with adversarial testing. He stresses the importance of verification loops — screenshots, video captures, real user flows, and regression tests — and praises the model's improved judgment, noting it now pushes back thoughtfully on code review feedback rather than defaulting to agreement. The conversation also covers cost considerations (the model is expensive, shifting how individuals and teams think about usage), interface questions (decoupling where work happens from where you discuss it, multiplayer collaboration), and the broader cultural shift. Krieger argues the model both raises the floor (non-technical users like recruiters can now build internal tools) and the ceiling (experts can tackle unprecedented complexity), while acknowledging a genuine sense of loss among engineers who loved the craft of writing elegant code.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.