Where to draw the human-approval line for agent actions: one operator's take
high-onsalt · reddit · 2026-11-08
A developer running a production agent shares how they gate human approval: anything irreversible or external (emails, record updates, API spending) requires sign-off, while read-only actions run freely.
The hard middle ground is writes to internal systems — too many approvals turn the agent into a slow form, too few and one bad tool call corrupts data. They ask whether others gate by action type or by a risk score.
More from coding & agent
- Joke post "Electricity. iPhone. OpenClaw." — harness leaderboard shows OpenClaw dead last — rm-rf-rm · 2026-11-08
- Perplexity CEO explains why Teammates beats rival agents: harness, model-neutral, search — AravSrinivas · 2026-11-08
- Indie Founder Launches Colabli, an MCP Gateway with Persistent Business Memory — Fragrant_Match1599 · 2026-11-08
- Idea renders Hollywood-grade 3D animation from your AI chat, starting at $15/mo — wordgrammer · 2026-11-08
- IRC or email? Two devs debate the right UI metaphor for talking to coding agents — yacineMTB · 2026-11-08
- Awesoft tests idiotproof: Opus now drives video workflow end-to-end — wordgrammer · 2026-11-08