The methodHow it works
The riskiest part,
built first.
A fixed-price week. It ends with the hard part built and tested, or with a written finding that it should not be built. Both are results. Only one of them is expensive to get any other way.
What happens
01
Intake
One call, before any money moves. You describe the problem. We ask what would have to be true for it to work, and which of those things nobody has checked.
02
Naming the riskiest assumption
We say it out loud and write it down in one sentence. If we cannot get it to one sentence, that is a finding in itself, and it usually means there are two projects here rather than one.
03
The build, on your real data
We build the least amount of software that can prove that sentence wrong. Real data, your systems, your constraints. Sample data proves nothing, because the thing that breaks a project is almost always the shape of the data nobody looked at.
04
The decision meeting
You see it run. Not a deck about it running. We tell you what held, what did not, and what we would want to know next.
05
What you leave with
A working prototype, or a written no. Either way: the code, the data work, and a roadmap. You own all of it. There is nothing to renew and nothing to unwind.
What you actually get
- The prototype, or the written finding, whichever the week produced.
- Every line of source, in your repository.
- The roadmap: what to build next, in what order, and what it depends on.
- The assumptions we did not get to, named and ranked.
- A fixed price, agreed before we start.
The toolingA human at the wheel
We use AI the way we use a compiler.
AI writes a great deal of the code here, and it makes a prototype week possible at all. It does not make the decisions. Somebody who has shipped this kind of system before decides what to build, what to throw away, and when the thing that looks like it is working is not.
The judgment is the product. The typing was never the expensive part.
Where we say noNo agent reaches production without a regression suite and a kill switch. If a deterministic script would do the job, we write the script. We will not point an agent at an untested warehouse and call the output insight.
The limitsFour of them
When not to do this.
You already know what to build.
If the requirements are settled and the risk is execution rather than uncertainty, you do not need a week of proving. You need a team. That is a different conversation and a cheaper one.
You want staff augmentation.
We are not a body shop. If what you need is three more engineers inside your existing process, hiring them directly will cost you less than routing them through us.
There is no measurable outcome.
If nobody can say what would make this a success, a prototype will not settle it, because there is nothing for it to settle. Get to a number first, even a rough one.
The hard part is political.
Some projects are not blocked on whether the technology works. Software does not fix those, and neither do we.
Prove the
hard part first.
Bring the part you are least sure about.
Start with a call