Researching AI agents that take real actions — would love your experience

I’m researching how companies are using AI agents that can actually take actions in business systems, rather than just generate text.

I’m especially interested in:

• Payments / financial workflows
• IT / cloud infrastructure
• Ecommerce / orders and refunds

I’m trying to understand:

What can your agents actually do today?
What actions are they NOT allowed to take autonomously?
Why do those actions still require human approval?
Have you ever had an agent make a wrong action? What happened?
What would make you comfortable giving the agent more autonomy?

I’m not selling anything — I’m doing customer research and trying to understand where the real problems are.

If you’re building or deploying agents in production, I’d really appreciate hearing about your experience.