Managing AI agents like a manager: parallel agent loops, Skill libraries and AGENTS.md
dotey · x · 2026-10-28
dotey poses a good question: human managers are measured on developing people — what's the equivalent metric for someone managing agents? His answer: good agent managers should build up knowledge bases and Skills, and optimize AGENTS.md to dramatically improve how well agents understand projects and execute tasks.
In his follow-up he argues multi-loop, multi-agent parallelism is inevitable:
- Like an IC promoted to Manager, you must split tasks across several "employees" while aligning direction, unblocking, reviewing, merging — and owning failures.
- Human attention is scarce while agent execution is async and slow — his tasks average 30+ minutes, so idling while waiting is wasteful; the move is to run several agent tasks in parallel, dispatching and verifying continuously.
- Everyone will eventually be pushed from executor to agent manager.
Related event: Debate over Multi-Agent Loops: Everyone Will Become an Agent Manager(3 posts)→
More from coding & agent
- Zero-AI-at-runtime design generator hits run 8, new aesthetics expected by run 10 — Afinetheorem · 2026-10-28
- Gunning for 60fps on a 200MHz CPU: dev shares Dreamcast optimization challenge — AIandDesign · 2026-10-28
- EnvLunch turns executable environments into free verifiable agent training data — KingsCollegeLondon · 2026-10-28
- Is agent recovery an underrated problem? Devs debate checkpointing and idempotency — CuteClass9864 · 2026-10-28
- Matt Turck amplifies Headless launch: AI-built data pipelines shouldn't die in a laptop Python file — mattturck · 2026-10-28
- Claude Code v2.1.311 auto-compacts 200K+ token sessions, adds runner idle-hold flags — caolonghua · 2026-10-28