Service 01
From first sketch to paying customer.
We build working SaaS products with AI coding agents under our own review. Not a prototype that survives a demo — software that runs in production, has tests, and that someone pays money for.
What you get
A working product in production, with the things around it you only miss when they are absent: a test suite, a deploy pipeline, error reports that actually reach someone, and documentation a next developer can work from.
In the first week there is something you can click on. Not because speed is a sales argument, but because a working screen is the only reliable way to find out that the question was different from the one on paper.
What has to be right from the start
Three things are nearly impossible to fix later, so we do them from day one: the separation between customers, the money arithmetic, and who is allowed to see what.
Multi-tenancy belongs in the database, not only in the application. In booxx the isolation sits at two layers at once — a tenant filter in code and row-level security in PostgreSQL — because a single forgotten `where` otherwise shows another customer's data. That is not a bug but a breach.
Amounts are calculated in whole cents and tested on the outcome, not on the code. A rounding error that sits there for two years before anyone finds it costs more than the entire project.
How we build
AI agents write the bulk of the code. We set the architecture, review every change, and put up the gates it has to pass: tests green, typecheck green, build green. An agent that edits a test to make it pass does not get through.
On our own products that work continues overnight: a crew of agents that removes dead code, fills test gaps and measures search visibility, each in its own worktree behind its own gate. That is not a showpiece — it is the reason maintenance does not pile up until it hurts.
What AI does not do is judge. The decision not to build something, the trade-off between two data models, the conversation where it turns out the real problem sits two departments away: that is human work and it stays that way.
When we are not the right choice
If you need a team of ten, we are not it. If there is a tender with an ISO certificate attached, probably not either. And if the answer is already fixed and someone just needs to type: you can get that cheaper.
Where this does work: a scoped product or a scoped module, one client who can decide, and a timeline in which quarters are too long.
Frequently asked questions
So does the AI write the whole product?
Most of the lines of code, yes. The architecture, the reviews and the decision about what does and does not go in are ours. In practice that is the same ratio as a senior working with a team — this team just delivers faster.
What if the AI makes a mistake you miss?
That is why there are gates. Tests, typecheck and build must be green before anything is merged, and wherever money is calculated there are extra tests that check the outcome rather than the code. Mistakes still slip through occasionally — the difference is that they show up before a user finds them.
Who owns the code?
Yours. The code lives in your repository, on your infrastructure, with your keys. We do not deliver a platform you get locked into.
How fast is fast?
Pilot-Next went from first line of code to paying users in weeks. That is not a promise for your project — scope, integrations and the number of decision-makers set the pace far more than the code does. What we do promise: in the first week you see something working.
Let's talk
What are we building?
One conversation is enough to know whether we fit. Tell us what you have in mind — we will tell you how fast it can happen, and whether we are the right people for it.