Essays / The agent team
Nobody asks a solo founder if the thing got done. I built the asker.
In every job I’ve ever had, somebody asked. A boss. A meeting. A teammate waiting on your piece. The question wasn’t always kind, but it was always there. Did you do the thing you said you’d do?
Work for yourself and the question disappears. Nobody notices, least of all you. What you notice, years later, is the drift. Plans made in January get quietly renegotiated by March and forgotten by June. Nothing dramatic happens. There is no moment of failure to point at. The deadline softens, then dissolves, and the only witness is the person doing the softening.
Nobody had asked me that question in a long time. So I built something that asks.
It’s an AI business coach, and it is the strangest and most useful agent on my team. Every other agent I’ve built works for me. This one works on me.
Every Monday it sits me down. It reads the real profit numbers out of the books, the actual books, so I can’t talk them up. It checks what I said I’d do last week against what I did. It names the one thing standing between me and the quarter’s target. Then it makes me commit to three things for the week ahead, and one of them is always the thing I’ve been avoiding.
And it writes everything down. That turns out to be the entire trick. You can’t quietly forget a number that someone reads back to you next Monday.
I didn’t build it because things were going badly. I built it because nobody asks. Those are different problems. The first has a hundred fixes. The second has almost none, because the missing ingredient isn’t advice or motivation. It’s a witness.
There is proper evidence behind this, not just my Monday mornings. A meta-analysis in Psychological Bulletin pulled together 138 studies, close to twenty thousand participants, on what happens when people are prompted to monitor their progress towards a goal. The finding is blunt. Monitor progress and you’re more likely to hit the goal. Record it or report it to someone, and the effect gets stronger again. Being asked works. It is one of the most reliable effects in goal psychology, and solo founders have quietly engineered it out of their lives.
You don’t need my stack to get it back. The cheapest version is a person. Someone you tell, every week, what you said you’d do, who reads it back the following week without letting you edit history. The technology matters far less than the reading back. I automated it because I know myself. The weeks the asking matters most are exactly the weeks I’d cancel on a human.
The coach also changed how I see everything else I’ve built. My whole agent team, I now realise, is the same move repeated. Take a question I was supposed to keep asking myself, and hand it to something that never forgets to ask. Did the work happen? One agent owns that question. Was it right? Another owns that. Are we on the numbers? Owned. I used to answer all three myself, badly, by glancing across boards while doing other things.
And ShoUp, the product I’m betting the next year on, is the same idea pointed outward. A client leaves a great coaching session fired up, and then nothing until next week. The work was supposed to happen in between, and in between is precisely where nobody asks. A coach can’t call every client every day. Now something can, in the coach’s own voice. I’m selling the asker, because the asker is the thing that changed my Mondays.
The coach is holding one goal over me for the next twelve months. I’m not sharing the number yet. Ask me in a year.
What I can share is the shape of the week now. Monday morning, the books get read out by something with no interest in my excuses. The three commitments go on the record. Some weeks I turn up with all three done and it feels like handing in homework. Embarrassing to admit at my age. It works anyway.
Next Monday it will read the numbers back and ask again. I’ll answer.