The sprint, day by day
Two weeks, one process. Below is exactly what happens and when — no stages like a "discovery workshop."
─
STEP 0 — Call (30 minutes, free)
│ I ask about the process, about any earlier attempt to automate it, and about what the process costs each month in people's hours. If I decide I'm not the right person for it, I'll say so right away.
STEP 1 — Mapping (day 1–2)
│ I sit with the process. I look at real data, not a sample. We agree on acceptance criteria and write them down before I write a line of code.
STEP 2 — Build (day 3–8)
│ I build. A short status every day — what works, what doesn't, what turned out different from what we assumed on day two.
STEP 3 — Deployment (day 9–10)
│ Launch in production, on your infrastructure. Documentation, handover to the person responsible, and instructions for what to do when something breaks.
STEP 4 — Maintenance (optional, from month two)
│ Monitoring, fixes, tuning as the process changes. A monthly contract, cancellable any time.
─
┌─[ What "it works" means ]
We set the acceptance criteria on day two, and it has three parts: a measurable quality threshold on real data, the volume it has to be hit on, and a person on your side who confirms the result.
The agent classifies the tariff code correctly for at least 92% of 500 line items from last quarter, verified by the head of the customs department.
If the threshold isn't met, you pay half.
┌─[ What I don't do ]
│ I don't run workshops about AI transformation.
│ I don't write strategies or roadmaps.
│ I don't build things that can't ship in two weeks — if your process doesn't allow for that, I'll tell you on the call.
│ I don't hire myself out by the hour to someone else's backlog.