Anthropic details multi-day Claude workflows for scientific computing
Anthropic published a research post on long-running Claude for scientific computing, describing how to structure agentic coding work that can run across many sessions instead of a single short chat. The piece focuses on practical workflow design: using test oracles to check progress, keeping persistent memory so the agent can resume effectively, and orchestrating sub-tasks when debugging or exploration becomes too complex for one straight-line pass. Anthropic says the patterns are meant for scientific computing, but the guidance is presented as broadly useful for other long-running agentic coding projects as well. The post also points to an internal C compiler project as an example of Claude operating across roughly 2,000 sessions. The development is best read as a workflow and orchestration playbook, not a new model release.
AI-generated summary with automated source checks. It has not been reviewed by a human; consult the linked evidence.
Read original at Latent.Space ↗
Originally published 10/12/2026, 6:18:17 PM