Designing agent approval screens: show the effect, not the narrative
ClickOk5811 · reddit · 2026-10-29
A Reddit deep-dive on designing human approval gates for agents. The author argues that narrative-style approval prompts (a well-written summary of a bad action reads reasonably too) force reviewers to trust prose, while a structured "effect view" — action, account, amount, data changes, reversibility, policy trigger, source ticket, plus an "ask the agent why" button — lets humans judge in seconds from facts.
Key points:
- Gate placement should weigh undo difficulty, whether mistakes self-announce, review cost, and daily volume; cheap-to-discard actions get logs/rules/sampling, irreversible ones get human review
- Limitations: better screens can't fix volume problems, shared blind spots between rules and agents, some domains need named sign-off regardless
- Track rejection rate alongside sampled audits — a gate that never rejects could be a good agent or a review catching nothing
The author ends by asking how others design approval screens and whether anyone has measured if reviewers actually read them.
More from coding & agent
- Retweeted thesis: Self-updating docs are an 'infinite market' like AWS — handotdev · 2026-10-29
- Vicki Boykis on how to learn ML in 2026: self-taught path and working with agents — vboykis · 2026-10-29
- Coinbase launches global AI trading competition with $2M prize pool — MurrLincoln · 2026-10-29
- vboykis: an agent eagerly helped me set an env var wrong and broke my results — vboykis · 2026-10-29
- Video pro: anyone claiming AI edits videos end to end is lying — juliey4 · 2026-10-29
- One Prompt Batch-Audits and Edits 188 Videos, Author Calls Anti-AI Like Anti-Electricity — yungcontent · 2026-10-29