Anthropic's Claude Managed Agents: Decoupling Brain from Hands Cut First-Token Time by 90%
AI Engineer · youtube · 2026-10-12
Priyanka Phatak, who leads Anthropic's Claude Managed Agents engineering team, walked through the architecture overhaul in part 1 of a 4-part workshop.
- She defines an agent as a model in a loop with tools, plus a harness and sandbox; harnesses encode model-specific assumptions that go stale with each release.
- The first version crammed session, harness, and sandbox into one hard-to-debug "pet" container. The redesign splits them into an append-only session log, a brain (harness plus Claude), and hands (sandboxes and tools)—one brain, many hands.
- Credentials are injected at an egress proxy so Claude never sees them, and flexible compaction dropped time-to-first-token by roughly 90%.
- She closes by demoing a joke-teller agent over the core APIs: environment, agent, session, events.
More from coding & agent
- Ice Cubes dev uses Codex to clear open-source repo backlog of bumps and fixes — Dimillian · 2026-10-12
- SEO team uses Grok bot groups as web crawler, sometimes beating ScreamingFrog locally — gaganghotra_ · 2026-10-12
- EXL3-QLoRA lets you fine-tune 2-8bpw quantized models, beating BNB NF4 on loss — Unstable_Llama · 2026-10-12
- Linus Torvalds is using Claude Code on a hefty Linux kernel performance improvement — PointmanW · 2026-10-12
- Claude-driven Blender 3D modeling from images makes this dev bullish on inspection gear — jwt0625 · 2026-10-12
- Work2Skill Turns Human GUI Demos into Reusable Agent Skills, Beating No-Skill Baselines — nex-agi · 2026-10-12