How we work

Four phases, flat fees, a planned exit.

The engagement shape is the same whether we are building an agent, a private inference stack, or ordinary software at unusual speed. Each phase is priced flat before it starts, and each one ends with something you keep.

Discovery → pilot → production → handover. Every phase ends with an artifact you own. Illustrative.

Discovery

1–2 weeks

We map the workflow, the data, and the risk. Paid, fixed fee, and useful even if it ends here.

You keep

A scoped build order with risk conditions

Pilot

3–6 weeks

A working system on a real workload with real users — designed from the start to reach production, or to fail honestly and early.

You keep

The pilot, its evaluation record, and a go / no-go you can defend

Production

scoped per build

The system carries real load. Autonomy expands only as the evaluation record earns it; every gate we relax is a decision with evidence attached.

You keep

The system, its evals, its runbooks, its audit trail

Handover

planned from day one

Your team operates it without us. Handover is part of the build, not a renegotiation.

You keep

Code, weights, evals, documentation — all yours

Graduated autonomy, in practice

Every agent starts with approval gates on everything. We build the evaluation harness first, run it against real traffic, and relax one gate at a time — each relaxation a recorded decision with evidence attached. If the record says a step isn't ready, it stays gated.

This is also why we can say no cleanly: when the evaluation record shows a workflow is not ready for autonomy at all, you hear that early — in discovery or in the pilot — while the spend is still small.

Gates relax one at a time, with the evaluation record as the justification. Illustrative.

What we don't do

Each of these is a standing policy. They cost us some deals, and they are why the deals we do take reach production.

We don't sell strategy decks.

The deliverable is working software. When analysis is the job — value mapping — the output is a build order anyone could execute.

We don't run pilots designed to stay pilots.

Every pilot is scoped against a production path before it starts. A pilot that can only ever be a demo is declined.

We don't bill by the hour.

Our pipeline makes us fast; hourly billing would pay us to be slow. Phases are flat-fee, agreed before they start.

We don't hold your system hostage.

You own the code, the weights, the evals, and the infrastructure. Handover is planned from day one.

We don't take work we don't believe in.

If the honest value-to-risk answer is “don't build this,” that is the answer you get.

The first call is a scoping call.

Bring the workflow you are thinking about. We will tell you what discovery would cost, what it would answer, and whether we think the answer is worth the fee.