# What’s one part of an AI agent workflow you still don’t trust enough to automate? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/79add7e107e62e4a

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-09-21 12:28 ingested 2026-09-22 02:35

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

## retained text



```
AI agents can handle a lot of tasks now, but there always seems to be one step where people still prefer to check the work manually.

It could be:

* Making important decisions
* Sending messages to customers
* Updating a CRM
* Researching information
* Scheduling meetings
* Approving transactions
* Handling unusual requests
* Using external tools

**What’s the one step you still keep under human control, and why?**

I’m more interested in real-world experience than demos.
```



document fetched 2026-09-22 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1wmbdvn/whats_one_part_of_an_ai_agent_workflow_you_still
