01 · Diagnose — free, no obligation
We spend a week or two learning how your business actually works — the workflows, the data, the systems, the commercial goals. You get a full statement of work, a wireframe of what we'd build, and a clear ROI case: what it's worth, what it will do, and what it won't. You keep all three whether or not we ever build it. If the honest answer is "don't do AI yet", that's what the document says.
More on the AI Readiness Diagnostic →
02 · Build — typically 4–12 weeks
A focused senior team — tech lead, senior engineer, part-time AI ops engineer — ships working software on a two-week cadence, in your tools, on your stack, behind your auth. The first paid step is a scoped proof of concept that proves the idea in practice before you commit to the full build. Evals are built alongside the features, so quality is graded, not guessed.
More on building production AI →
03 · Operate — ongoing, optional
Production AI changes underneath you: models improve, prices move, APIs deprecate. Operate covers monitoring, evals before any change goes live, model updates when better options land, and on-call for production issues — for businesses that don't want to staff that internally. Prefer to run it yourself? We hand over cleanly and document everything.
More on Managed AI Operations →
Already mid-project and stuck?
The Rescue service reviews a stalled or underperforming AI build, tells you what's salvageable, and gives you a route forward — even when the honest route is a clean rebuild. More on Stalled-project Rescue →
What it costs
The Diagnostic is free. Proof-of-concept work is scoped and fixed before it starts. Smaller production builds typically start in the tens of thousands, and every build is priced against the value in the ROI case — if the numbers don't justify the build, we won't recommend it. Two useful tools while you're weighing it up: the AI ROI calculator and our honest guide to what AI implementation actually costs.
The working principles
- Built around the business, not the demo. The measure of success is a number in your P&L, not a wow moment in a meeting.
- Your stack, your auth, your data. See the trust & security page for the full detail.
- Multi-model, no lock-in. Workloads route to the model that clears the quality bar at the best cost — and move when the evidence says so.
- Evals before opinions. Every change is graded against your quality bar before it ships.
- Senior people, end to end. The team that scopes the work builds it and answers the phone afterwards.
Common questions
What does an engagement look like end to end?
Three steps with one team: the free Diagnostic (SOW + wireframe + ROI case), a 4–12 week Build shipping every two weeks, and optional Operate support. Payment starts only at the proof-of-concept stage.
Do we have to commit to all three phases?
No. The Diagnostic carries no obligation and you keep its outputs regardless. After a build, some clients run the system themselves; others choose Operate. Each phase earns the next.
Who actually does the work?
The founders and senior engineers you meet in the first conversation. No handovers between a sales team, a strategy team and a delivery team.