Claude Code demo: Opus 5.5 plans while Jev makes bounded decisions
An Anthropic product manager demos a 90-minute autonomous coding run where Opus 5.5 handles open-ended planning, Jev handles bounded decisions, and the code layer manages budget, retries and human approval.
Anthropic's Claude Code product manager: "Claude Code worked for 90 minutes on this task" in this 10-minute demo, one prompt becomes hundreds of lines of code, tests, corrections, and a pull request Prompt → Plan → Edits → Tests → Verified Output the model is only one piece context, tools, permissions, state, and recovery keep the work moving Opus 5.5 handles open-ended research, planning, and writing Jev handles bounded decisions code owns the budget, retries, files, and human approval anyone can call a frontier model almost no one builds the operation around it watch the workflow then save the full Jev + Opus 5.5 architecture below ↓
This post quotes another post: x.com/leiroops/status/2102281900823839023