How Anthropic Uses Claude Fable 5 With Mike Krieger
TL;DR
Fable 5 feels less like a chatbot and more like a teammate: Krieger says he now gives Claude long, multi-hour tasks, including overnight builds, and trusts it to recover from failures like a downed service by scaffolding substitutes and finishing the job.
The workflow shift is as important as the model jump: Instead of asking for a quick feature stub, Krieger now starts with architecture conversations, diagrams, HTML mockups, and markdown plans, then runs multiple parallel Claude sessions to execute chunks of work.
Verification is the new core engineering skill: At Anthropic, confidence comes from tight review loops such as screenshot galleries, video captures, staging flows with real accounts, regression paths in text, and mock backends that Claude can keep in sync as the code evolves.
Software engineering is not over, but the craft is changing fast: Krieger says ownership, intent, production judgment, and incident response still matter, even as coding itself collapses into product management and delegation, creating both excitement and a real sense of loss for engineers who loved hand-writing elegant code.
Fable 5 is expensive, but Krieger argues total task cost matters more than per-turn cost: His claim is that the model often saves money in practice by getting the whole task right in one pass instead of requiring 9 or 10 corrective turns, and even a personal weekend app fit within modest paid usage.
Anthropic is building toward agent-native software and dynamic workflows: Krieger demoed a personal media tracker that Claude could modify from inside the app itself, and described using dynamic workflows to port a complex Python codebase to TypeScript over a weekend with staged translation, testing, and adversarial checks.
The Breakdown
Mike Krieger says Anthropic's new Claude Fable 5 is the first model he trusts with real overnight delegation, to the point where he can hand it a complex task before bed and wake up to a finished result, often with workarounds, documentation, and follow-through already handled. The bigger shift is not just better code generation, but a new way of working where software teams manage fleets of agentic sessions, verify outputs with screenshots and video, and spend more time on intent, architecture, and judgment.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.

Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.