# How are you capping the total damage an agent can do, not just each action? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/c776d9a6fbbf152f

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-06 17:55 ingested 2026-10-07 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1wz9bxs/how_are_you_capping_the_total_damage_an_agent_can/)

## retained text



```
Most setups I've seen put limits on single actions: refunds under $100, emails to under 50 people, that kind of thing. But 40 refunds at $90 each are all individually fine and still add up to $3,600 nobody approved.

For those of you with agents taking real actions (money, customer emails, tickets, CRM changes):

* Do you cap totals over time (per day, per week, per customer)? Or just per action?
* If you cap totals, what happens when the next action would go over: block it, queue it for a person, or something else?
* Do actions waiting for approval count toward the total, or only completed ones?

Curious what's actually working in practice versus what sounds good on paper.
```



document fetched 2026-10-07 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1wz9bxs/how_are_you_capping_the_total_damage_an_agent_can
