AnthropicfeatureClaude Managed Agents Update (Budgets, Regions, Skills, Advisors)
Four changes to Claude Managed Agents: a spend cap per session, control over where the work runs, skills loaded from your own repository, and a stronger model an agent can consult mid-run.
Details
- You can set a spending budget per session; a session that hits the cap pauses and emits a budget_reached event, and raising the budget lets it resume where it stopped
- A new model.inference_geo setting picks where the work runs: global runs wherever there is spare capacity at the standard rate, us keeps it in-region and bills at 1.1x
- Managed Agent sessions now load skills from the .claude/skills/ folder in your attached repositories automatically at start, the same folder Claude Code already reads
- You can give an agent an advisor, a stronger model it calls mid-session for a second opinion, by adding one line to the roster
Best for
Teams running Claude Managed Agents on real workloads where cost and data location matter
When to use it
An agent session can run up an unpredictable bill, or your compliance rules require the work to stay in the US
Who benefits
Developers and platform teams operating Managed Agents in production rather than testing them
Join the newsletter to get ahead with AI.
I share strategies, new modules, free guides, and tools to get you better at using AI in your work.
Leading a team? See team programs →