If an AI agent does everything for us, where are we in the loop?
Here is the pitch for the newest kind of AI assistant. It briefs you in the morning. It keeps working between your messages. It does not bother you unless a decision truly needs you. Lovely. Who would not want that?
One of the big labs just opened a waitlist for exactly this kind of proactive agent, and at the same time it is making a more automatic mode the default for paying users, with automated checks that block the riskiest actions. You can still go back to approving everything by hand, but let us be honest about how many people will.
The ladder nobody announces
It starts with an agent that asks before it does anything. Then it asks only about decisions. Then it asks only about the decisions it considers important. See where this goes?
So if the agent only pings you when it decides you should be pinged, who is actually in charge?
Fair question, and the numbers from inside one of these labs make it sharper. Since March, its AI agents have logged more working hours than its human staff. By August, the agents each employee directs were running about six times as many hours as the employee. That is an estimate of runtime, not of value created, so do not over-read it. But here is the kicker: someone outside engineering would need more than half a working week just to read everything their agents produced.
So where are we?
Our role slides, step by step. From doing the work, to reviewing it, to approving it, to receiving a summary of it. Each step feels reasonable. The sum of them is a person who signs off on things they never really looked at.
Two things make this tricky. First, the whole deal only works if checking costs less than the time saved; if reviewing eats the savings, the loop is empty. Second, the better the agent gets, the harder it is to spot its mistakes, because everything looks right. People who build with these tools already warn against asking a model to judge its own work. The check has to come from somewhere else.
So are we still in the loop? Technically, yes. We are the signature at the bottom of someone else's decision. Whether that counts as control or just ritual is the question worth sitting with.