Vibeleaderboard
← All Intel
Intel / article

Scaling Managed Agents: Decoupling the brain from the hands

Source
anthropic.com
Author
Anthropic Engineering
Date

Context

A crashed execution environment need not erase an 's record of its work. Anthropic's Managed Agents architecture separates the session history, the software that coordinates model and tool calls, and the where code runs. Its earlier design placed everything in one container, so losing that container could lose the session.

With those components separated, a replacement coordinator can read the durable event log and resume from the last event. A failed sandbox instead becomes a tool error that Claude can respond to, potentially by requesting a replacement. This makes recovery depend on explicit records and replaceable components, rather than keeping one particular container alive.

Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
  • agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
Key quotes

“Harnesses encode assumptions that go stale as models improve.”

“But when we used the same harness on Claude Opus 4.5, we found that the behavior was gone. The resets had become dead weight.”

“In our case, the server became that pet; if a container failed, the session was lost.”

“The solution we arrived at was to decouple what we thought of as the “brain” (Claude and its harness) from both the “hands” (sandboxes and tools that perform actions) and the “session” (the log of session events).”

Read the source www.anthropic.com
More from Anthropic Engineering
Recommended reads
Comments

Checking sign-in…

Loading comments…