The agents that actually stuck for me do one narrow thing, not everything
Cold_Hall_5384 · reddit · 2026-09-27
After months of building agents, the author concludes that do-everything agents were always brittle—impressive in demos but unusable in practice because it was impossible to tell when they went off the rails. The agents that stuck are embarrassingly narrow: one watches a folder and tags files, one drafts replies to a specific type of email, one summarizes a daily report.
Narrow agents are easy to trust because their behavior is predictable and failures are instantly visible. The author asks others running agents in production: are yours narrow and boring, or have you made a broad one actually hold up?
Related event: Practical Lessons: Narrow Agents Survive, General-Purpose Agents Fail(2 posts)→
More from coding & agent
- Baseten's Guide: Evaluate Open Models for Coding Agents on Your Own PRs — baseten · 2026-11-16
- Google Research open-sources PaperOrchestra, a multi-agent framework for writing AI papers — google-research · 2026-11-16
- ffmpeg-skill gives coding agents local video editing via 42 FFmpeg tools and MCP — kajisho5 · 2026-11-16
- Agents Now Reply to Slack Messages on Their Own, Resolving a Prod Incident Ghost-to-Ghost — iamrobotbear · 2026-11-16
- Engineer's Interview Prep Method: Use Claude to Find Which Answers Undersell Your AI Work — brandon_galang · 2026-11-16
- DeepSeek open-sources its app frontend "chat-demo", already at 3.9k stars — NielsRogge · 2026-11-16