Here is an embarrassing admission from someone who runs his own companies: I need a boss.

Not for direction. I have more direction than I can use. I need one for the thing a good middle manager actually does, which is notice that a task has been sitting untouched for nine days, walk over, and say without any drama: “Where are we on that? What is stopping you? Could you do the first twenty minutes of it before lunch?”

When you work for yourself nobody does that. The work you like gets done early and well. The work you dread, the awkward follow-up, the tax paperwork, the proposal you are not sure how to price, slides quietly from Monday to Friday to next month, and the only person who knows is you, at two in the morning.

So I am building the manager. I will be straight about where it is: the pieces work, I do not yet use it every week, and the gap between those two facts turns out to be the interesting part.

What a middle manager really does

We are rude about middle management, and I think it is because we picture the bad version. The good version is mostly four repetitive jobs:

  • Tracking. Knowing what everyone has committed to and when.
  • Chasing. Asking, at the right moment, without it being personal.
  • Unblocking. Turning “I don’t know where to start” into a first step.
  • Reporting upward. Telling the person above what is on track, what is slipping, and what needs a decision.

I would put sixty to eighty percent of that in the repetitive column. It is diligence on a schedule. It needs a good memory, no ego and endless patience, which is a fair description of an AI and a poor description of me on a Thursday afternoon.

How mine is designed

It reads the same sources my assistant does: the task board, the inbox, the calendar, the session notes I keep for each business. From those it knows what I said I would do, for whom, and by when. Then it does three things.

It tells me who is waiting on me. Not what is on my list. Who is waiting. That reframing changed more than I expected, because I will procrastinate on a task forever and I will not leave a person hanging once I can see their name.

It breaks down the thing I am avoiding. Avoidance is nearly always a sign that a task is too big or too vague to start. “Do the quarterly tax work” sits for weeks. “Open the ledger and list the six unreconciled items, ten minutes” gets done before coffee. A manager that notices a task has aged and offers the ten-minute version is doing the single most useful thing anyone can do for a procrastinator.

It reports, to me, about me. Once a week: what moved, what did not, what has slipped twice. The idea is that seeing “slipped twice” in plain text, without judgement, does what no amount of self-reproach does. There is no telling-off. There is just the fact, and the offer of a smaller first step.

One design choice worth copying: it does not write my to-do list. I write my own tasks by hand, deliberately, because a list a machine generated is a list I feel no ownership of. The manager reads the list, ranks it, chases it and reports on it. The commitments stay mine.

What I am building it for

The goal is to lower the emotional cost of being organised. I carry too much in my head, which means thinking about all of it, all the time, slightly. When something else holds the list and tells me when a thing matters, I can think properly about the one thing in front of me, because the other forty are being watched by something that does not get tired or anxious. On the weeks I use it, that is exactly what happens. The irony is not lost on me that the thing I most need a manager for is the habit of checking in with my manager. So the next piece is making it come to me on a schedule, rather than waiting to be asked.

That is what a good manager gives a team, by the way. Not pressure. Relief.

If you lead a team, the same thing scales

Everything above works for a team of ten as well as a team of one, with the obvious care. An AI manager that tracks commitments, chases on schedule and reports what is slipping takes the repetitive majority of the coordinating load off your human managers, who can then spend their time on the part that needs a person: the conversation about why it is slipping. The rule I hold to is that it chases tasks and never judges people, and that anything it sends to a human being other than me is a draft until someone approves it.

If you want one

Start small. Put your commitments somewhere an AI can read them, a task board, a document, anything. Ask it each morning who is waiting on you and which task has sat longest. When it names the one you are avoiding, ask it for the ten-minute version. Do that for a fortnight and see what happens to the pile.

And if you want the real thing, wired into your systems and running on a schedule whether you ask or not, tell me how you work and what you keep putting off. I promise not to judge. I am still building the manager who will.