Design agent approvals around risk, not org charts
usehive · reddit · 2026-10-17
The author argues agent approval flows are designed backwards: they follow the org chart instead of risk, so read-only lookups, routine updates, and consequential actions all wait for the same manager for the same two days.
The deeper point: the real rules were written by practitioners, not platform teams, and appear in no API contract—testimonials need human review, escalation after two contacts, sub-$50 refunds auto-approved for paid, never-refunded orders. Those unwritten rules ARE the approval flow; risk lives there, and an agent that can't see the rules can't follow them no matter how clean the permission model looks.
Proposed design: let reads through, let routine updates through with logging, and make consequential actions wait for a named human with the reason attached. Closing question: what's the highest-risk action your agent can take today without asking anyone?
More from coding & agent
- CoT isn't faithful reasoning: researchers push latent reasoning plus verifiable intermediate plans — AlexTensor · 2026-10-17
- Claude runs TradingView end-to-end: scans, backtests and writes indicators on its own — nikola_mr64990 · 2026-10-17
- AI Writes Binary Directly: Model Generates an FL Studio Project File — TinfoilTricorn · 2026-10-17
- Infographic: Turning Claude Into a Full AI Operating System With 8 Features — nikola_mr64990 · 2026-10-17
- Productivity is going multiplayer again — this time agent-human-agent in one workspace — aishashok14 · 2026-10-17
- MVP talk at AI Engineer: what does the system do, and can you change it safely? — altryne · 2026-10-17