Running several narrow agents instead of one do-everything assistant: what's working
FitForWhatsNext · reddit · 2026-10-16
The author shares weeks of hands-on experience splitting daily work across four narrow agents — social posting, email watching, website/career, and research — each with its own instructions, and none able to post or send without human approval.
- What works: narrow roles make output far more consistent, and the approval step catches bad drafts before they go out.
- Unsolved pain point: handoffs and shared context between agents — when an email matters for social, things fall through the cracks.
- The author asks the community for patterns: shared notes file, a coordinator agent, or something else.
A practical template for a "many narrow agents + human approval" personal automation setup, with the handoff problem as the open question.
More from coding & agent
- Claude Code and Codex now autocomplete the user's next action — tokenbender · 2026-10-16
- Hermes Agent ships official Reverse Engineer Anything plugin with one-command setup — Teknium · 2026-10-16
- MLExpert: 35B MoE Agent Trained to Autonomously Handle Machine Learning Engineering — Shangheng Du · 2026-10-16
- Cowork-Gym: Alibaba Synthesizes 64.9K Agent Work Environments, Boosting Qwen by 8.7 Points — Anni Zou · 2026-10-16
- MMSearch-OPD: Joint RL and On-Policy Distillation for Long-Horizon Multimodal Search Agents — PekingUniversity · 2026-10-16
- A gallery of 214 projects built on one model shows the 'post your build' ad flywheel — gaganghotra_ · 2026-10-16