← sources

how do you let non-engineers approve risky agent actions without killing the flow?

source: reddit (reddit:AI_Agents)published 2026-11-15 17:29ingested 2026-11-16 02:34

retained text

I'm building a platform for agencies running agents for clients (Zelvara). The part I keep going back and forth on is approvals. Right now each agent can mark actions as "ask first," and a person approves them from an inbox. Draft and live are separate, so changes get tested before release.

Things I'm unsure about: how much to batch approvals, what to do when nobody answers, and how to keep audit logs useful instead of noisy.

One gap I know about: per-user OAuth tokens aren't done, so users of an agent share one token per connection. If you've handled that in production, I'd like to hear how.

Product link if you want to poke at it: zelvara-ai.vercel.app

document fetched 2026-11-16 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1xx1c68/how_do_you_let_nonengineers_approve_risky_agent