Your agents need a manager
AI agents can now work through the night. The scarce resource in the morning is the person who has to review it all.
Some mornings I open my laptop to find more work finished than I could have done in a week. Agents ran overnight on code, research and drafts. Each one stopped at some point and left something for me: a question, a result to check, a decision it was not allowed to make.
It is a remarkable way to start a day, and it creates a problem nobody has designed for yet.
What happens when agents do the work?
The bottleneck moves. Building used to be the slow part, so plans were shaped around how much a team could make. Now an agent can make a great deal, quickly, and every piece of it still needs a person to read it, judge it and say yes.
Review is the new constraint. Twelve finished pieces of work waiting for approval is a full afternoon, and it arrives on top of the meetings, the calls and the rest of a normal day. Most people find out the hard way, with a backlog of results they have not had time to look at, and agents sitting idle because they are waiting on a reply.
Who keeps track of the agents?
Right now, nobody does. Each tool shows its own runs in its own window. The person who started them carries the rest in their head: which ones are blocked, which ones are done, which ones will be ready by lunch.
That is a management job, and it needs the tools managers have always had. A clear view of what is waiting on you. An honest estimate of how long the reviews will take. A way to say “hold the new runs, I have enough to read today.”
What should that feel like?
Like a good chief of staff at the start of the day. One page that says what finished overnight, what needs you, and whether today still fits once the reviews are counted. Nothing more urgent than it really is. Everything else can wait in the side room until you want to look.
I am building this into the to-do app I mentioned in August, because a day’s work now includes the work your agents hand back. The list that keeps your day has to know about them too.
Where does this go next?
Teams will soon have more agents than people. The organisations that do well will be the ones that design the human side of that arrangement with the same care as the technical side: who reviews what, how much in a day, and what is allowed to happen without anyone looking. Those are questions of judgment and of design, and they are the questions I find most interesting to work on right now.