Would you let an AI agent email your customers without reading it first?

I'm Erin, founder of AutomateNexus CRM. We build a CRM where AI agents work the pipeline: they score new leads, draft the follow-ups, book the meetings and flag the deals going cold.

The hardest question we've had to answer isn't what the agents can do. It's what they're allowed to do without asking.

Here's where we landed. An agent can read, research, score and draft on its own. Anything sensitive, like an email to a customer, waits for a person. The draft shows up with the agent's reason for writing it, in a queue we named "Waiting for your yes". Every action an agent takes is logged.

The agent proposes. You decide.

I can argue the other side too. If a person has to approve everything, how much time did the agent really save? And one wrong email to the wrong customer can cost more than a month of saved hours.

So where is your line?

1. Never. A human reads everything that goes out.

2. Low stakes only. Reminders and confirmations go on their own; anything written fresh waits.

3. Earned trust. After enough approved drafts in a row, the agent goes solo on that kind of message.

4. Full send. If it's wrong sometimes, that's the price of speed.

Reply with your number, and tell me what would have to be true for you to move one step up. I'm reading every answer, and the best arguments will shape what we build next.