Multi-agent in production: what works, what gets abandoned, and who controls the flow
fkj83 · reddit · 2026-10-11
A Reddit thread targeting practitioners running multi-agent setups in production. The OP suspects the popular "orchestrator spawns a swarm of chatty agents" pattern mostly burns tokens and is hard to debug, and wants input from teams with real users or internal processes depending on it.
Core questions posed:
- What's your architecture (fixed pipeline, router, supervisor) and who controls flow: your code/state machine or the LLM?
- Did you start multi-agent or single agent + tools? If you split later, what forced it: context size, tool overload, accuracy, latency?
- What did you abandon as too expensive, flaky, or unmaintainable?
- Where do humans sit in the loop (approvals, escalations, review queues)?
- How do you handle evals and tracing across agents?
More from coding & agent
- Codex agent battles Verizon's broken billing system in surreal hour-long browser session — rickasaurus · 2026-10-11
- WoWBench ranks AI agents playing vanilla WoW in an open world; a Grok bot tops the leaderboard — djcows · 2026-10-11
- Anatomy: one-line prompt turns Claude into an interactive 3D machine wireframe generator, fully open source — _AustinCalvert_ · 2026-10-11
- Claude Code fast mode reportedly blocks usage despite 97% of weekly credits remaining — RileyRalmuto · 2026-10-11
- REA: an agentic reverse-engineering toolkit for dissecting arbitrary hardware — jazir55 · 2026-10-11
- exe team's open-source web-based coding agent is standalone and useful — davidcrawshaw · 2026-10-11