What AI work actually costs, and why we publish shapes instead of prices
People notice that our packages describe shapes, two to three weeks, fixed fee, days per week, quarterly, and not prices. Some assume that means the prices are enormous. Some assume it means we will make them up in the room. Neither is right, and the honest explanation is more useful than a number would be.
Here is where the cost of AI work actually comes from, what moves it, and why a shape tells you more than a figure.
Where the money goes
Almost all of the cost of an engagement is senior time. The model calls, the infrastructure, the tooling are real but small by comparison for the scale of work we do. What you are paying for is the judgment of someone who has done this before: which system to build first, how to specify it so an assistant can build it fast, where the guardrails go, how to prove it works, and how to hand it over.
That is why fixed-scope work is priced by the week and ongoing work by the day. The unit is attention, and the price is a function of how much of it the shape requires.
What makes it cost more
Ambiguity. An engagement that starts without a clear owner, a defined process, or accessible data spends its first weeks creating those things. That is legitimate work, but it is work that could have been done cheaper internally, and we will say so.
Integration surface. A system that touches one tool costs less than one that touches six. Each connection is a place where the real world disagrees with the documentation.
Consequence. A system whose mistakes are cheap can ship with light gates and a small harness. One whose mistakes are expensive needs heavier controls, more evaluation, and more review, all of which are time.
Handoff depth. A team that can own the system needs less documentation and pairing than one that cannot. We will not skip the handoff, so a team with fewer hands to hand to costs more to leave safely.
What makes it cost less
A real owner in the room from day one. Accessible data. A process that is already defined well enough to automate. An engineering team with fast tests. A leadership team that has already decided, so the engagement is delivery rather than persuasion.
Many of these are things an assessment or a strategy sprint produces, which is why we credit those fees toward the build. The cheapest build is the one that starts with the questions already answered.
Why shapes, not prices
A price on a website has to be either so high it covers every case or so low it covers none. Both mislead. The shape does not: two to three weeks at a fixed fee tells you the engagement is bounded, that the number will be known before we start, and that it will not change unless the scope does. Days per week on a quarterly commitment tells you exactly how the ongoing work is metered and how long you are committing for.
Given the shape and the four cost drivers above, the price is a short conversation, not a mystery. We give a number in the first call, in writing, and it holds.
The comparison people actually want
The question behind the question is usually "compared to what?" Compared to a large consultancy, our engagements are smaller, shorter, and staffed by the people who deliver rather than the people who sold. Compared to hiring, a fixed-scope build costs less than a senior hire's first quarter and produces a running system rather than a job description. Compared to a platform subscription, the build is more up front and yours afterward.
None of those comparisons are universally right. The honest version of the answer is that the price is a function of the shape, the shape is a function of the problem, and the problem is what the first conversation is for.
If that conversation ends with a number you do not like, you will have lost an hour and gained a clearer picture of the work. That is a reasonable trade, and it is why the call comes before the quote.