AI Agents That Need Constant Babysitting Aren't Agents
Few-Garlic2725 · reddit · 2026-11-18
The author argues an agent requiring constant babysitting isn't really an agent—it's a suggestion machine with extra steps. Constant checking doesn't remove work, it redistributes anxiety. The viable business case may be much narrower: one high-confidence workflow failure the agent reliably detects or prevents. Key question: how do you test whether an agent reduces work rather than creating supervision work?
More from coding & agent
- Capability Is Not Authority: The Hidden Risks of Deploying AI Agents — Adjust-the-rhythm · 2026-11-18
- SciML author shares multi-model coding workflow: role-based routing for writing and review — ChrisRackauckas · 2026-11-18
- Rebuilding the Metabase MCP Server: Lessons in MCP Tool Design — Ramirond · 2026-11-18
- Dev wants a single client to manage all agent threads with Slack-style swipe UI — curious_vii · 2026-11-18
- Adaptive UI in a Claude CLI Enterprise Assistant: Prompting Isn't Enough — Shivam__kumar · 2026-11-18
- Lindy rebuilt its AI teammate engine with coding agents — faster at half the cost, 9 lessons — mattbeane · 2026-11-18