Ask what the agent will be able to see and do once connected, which actions can require your approval, where your data goes and whether it trains their models, how long they keep it, how you turn the agent off and take back its access, what it costs at your real volume including usage fees, and who owns your prompts, workflows and settings if you leave.
Seven questions, and what a good answer sounds like:
| Ask | A good answer |
|---|---|
| What will it be able to see and do once connected? | A specific list, and a way to narrow it |
| Can we require approval before it sends, pays or deletes? | Yes, per action |
| Where does our data go, and is it used for training? | Named companies, and training off in writing |
| How long do you keep our data, and how do we delete it? | A number of days, and a deletion process |
| How do we turn it off and revoke its access? | Minutes, from our side, without calling you |
| What does it cost at our volume, including usage? | A monthly estimate with the usage math shown |
| Who owns our prompts, workflows and settings if we leave? | You do, and you can export them |
If the answers are vague, the AI Agent Evaluation is built for exactly this.