Dispatch, verification, runaway control: the hard parts of coding-agent orchestrators
W1141175 · reddit · 2026-11-13
A developer rebuilding a coding-agent orchestrator (Claude Code + Codex behind a task queue, one git worktree per task, a script spawning sessions per task) reports the model layer works fine but the plumbing doesn't, and asks how others solved it:
- Dispatch: currently a flat for-loop over pending tasks — for 5–20 concurrent agents, do people reach for Redis+Celery or just a SQLite table with a worker loop?
- Verification: a worker returns "done" with no cheap way to confirm — test suite as gate, second-model review, or diff inspection: what actually catches bad output in practice?
- Runaway control: an agent that misreads a task can spend an hour refactoring unrequested things — needs a hard budget and kill path, not a prompt sentence saying "be careful"
Also asks whether the community is converging on a repo (mentions T3 Code and several open-source orchestrators) or everyone is still rolling their own.
Related event: Dev Tackles Orchestration Challenges of Multi-Coding-Agent Systems(2 posts)→
More from coding & agent
- Claude Code reads 40% of all developer docs, Mintlify data shows — handotdev · 2026-11-13
- Student's AI agent defended his school's network from a cyberattack after he left for college — heyneighbor · 2026-11-13
- Threedog.js: open-source Three.js engine lets Claude Code build, playtest and debug its own games — Gullible_Ranger1402 · 2026-11-13
- SGLang Summit agenda: LinkedIn serves ads model at 10K+ QPS under 80ms — DavidBennett__ · 2026-11-13
- Dev builds an OP-1-inspired skill for agentic coding, posted on GitHub Next — narphorium · 2026-11-13
- vork: open-source headless 3D/VFX engine for AI agents, renders black hole bending light in 18 min — ravisparikh · 2026-11-13