How we work · the build shape
One system.
Production quality.
Six to twelve weeks.
Architecture with failure-mode analysis from day one. The evals harness built first. Continuous integration on every commit. Monitoring, alerting and drift detection tied to business impact. And a handover pack a team that was not in the room can operate from.
You hold: the production system, the evals harness and monitoring, alerting and drift detection tied to business impact.
Where this shape sits
This is the contract shape behind most of the Factory Floor.
Every class of system on the Factory Floor except the workforce deployment is contracted this way: production LLM systems, decision-support tooling, and the larger evaluation and red-teaming infrastructure builds. Two Consulting Desk services also run under it — governance and assurance, and an AI strategy and roadmap that continues straight into its first phase.
It is not in the main navigation because you do not usually arrive shopping for a contract shape. You arrive with a system you want built, or a decision you want answered. This page exists so that when the engagement letter says Production Build, you already know precisely what that means.
Fixed scope and fixed fee, quoted in writing before work begins. Where the scope genuinely cannot be pinned down yet, we will say so and propose a Feasibility Sprint instead of pricing a guess.
The gate that shapes everything
A team that was not in the room can operate it. Every architecture and documentation decision is made against that test, which is also why handover is a workstream rather than a final week.
How it runs
Observable by default. Rollback written first.
The sequence is fixed because each stage is what makes the next one honest.
- Duration6–12 weeks from green light
- FeeFixed, quoted in writing before work begins
- MethodFailure-mode analysis · evals-first · observable by default
- GateA team that was not in the room can operate it
- Delivered byTwo founders directing nine engineering seats
Architecture, with failure modes named
Not just the happy path. What happens when a dependency is slow, a model is confidently wrong, retrieval is empty, cost spikes — each with a designed response rather than an incident.
The evals harness, before the feature code
Golden sets, adversarial cases, latency budgets, cost ceilings. Carried over from the Feasibility Sprint where there was one.
Build with CI on every commit
Evals run as tests. Security reviewed as its own workstream rather than as a checklist at the end. Integration is separately gated, because nine parts that each pass alone is the normal state of a broken system.
Monitoring tied to business impact
Alerting and drift detection on what the business cares about. A wrong number should be caught by the system, not by you.
Handover as a workstream
Architecture decision records, runbooks, an eval maintenance guide, and the rollback plan that was written before the first commit. Documented by the technical-writer seat as the work happens.
The two lists that matter
What you hold, and what we will not do.
Both are in the engagement letter before you sign it. The second list is the one worth reading twice — it is where most disappointment in this market actually comes from.
What you hold at the end
- The production system, running and observable
- The evals harness, wired into continuous integration
- Monitoring, alerting and drift detection tied to business impact
- Architecture decision records, runbooks and an eval maintenance guide
- The rollback plan, written before the work started
What we won’t do
- Systems without observability from day one
- Black-box models you cannot audit
- Builds without a rollback plan
- Handoffs to teams who were not in the room
- Fixed price without scope clarity
Who actually builds it
A senior engineer seat decomposes it. Nine seats build it. A human accepts it.
Every hand-back carries an evidence state — Verified, Built, Designed or Intended. “Should work” is not a state. A review gate, an integration gate and a human decision sit between the work and your production, and nothing is deployed, published or purchased by an agent.
Read the delivery model →After handover
A Lab Retainer buys standing capacity for evolution, eval maintenance and security updates. Team Enablement leaves the discipline inside your own engineers.
Ready to build it properly?
Thirty minutes with a founder. We will tell you how we would decompose it and what six to twelve weeks would realistically cover — before you commit to anything.
Fixed fee · Quoted in writing before we start · NDA available