Last May I told myself I'd take four fixed-price AI builds in a year and write down what I learned. The year is up. I shipped four. Some went better than others; one nearly didn't ship at all. Here is the version of the lessons that fits in a single essay — the rest are in the running changelog.
The clearest pattern across all four: the projects that worked had one decision-maker, one outcome, and one deadline. The projects that struggled tried to serve a committee. The model is a component; the product is the work; the politics are the killer.
On scoping. The hour-long scoping call is the most leveraged hour of the engagement. By the end of it, both of us either know what's being built or we don't take the next step. I'd rather lose a client at the scoping call than three weeks in.
On price. Fixed price is not about the number — it's about the conversation. When I name a price, I'm telling the client what scope I think this is. When they push back, they're telling me what scope they actually wanted. Most disagreements about price are really disagreements about scope.
On time. Four weeks is short. Six weeks is comfortable. Eight weeks is a different kind of project — usually one that needed splitting into two engagements. I learned this the hard way on build three, which was an eight-week project that should have been two four-week ones.
On the AI part. The model work is rarely the hard part. The hard part is the interface around it — the prompts, the empty states, the failure modes, the feedback loop, the trust signals. I spent more time on the moments when the model got things wrong than the moments it got things right.
On Friday demos. Every engagement runs the same Friday-demo cadence, and every time I think about skipping a week I regret it. The Friday demo is the forcing function. Skip one and the next one is twice as hard.
On saying no. I turned down more work than I took. Good projects with the wrong shape, good clients with the wrong scope, good problems on the wrong timeline. None of those calls have aged badly. Senior practitioners say no more than they say yes.
Four builds, four clients, no retainers, no subscriptions, no scope creep absorbed silently. The next four will look the same. If anything, they'll be more selective — Plans is taking more of my time, and the practice was always meant to be small.