Factory Floor · 02
What we run ourselves,
stood up inside
your organisation.
A governed multi-agent workforce for a defined function — code review, incident triage, compliance checking, research synthesis. In your environment, on your data, under your control. Every agent action logged; every claim carrying an evidence state.
You hold: the deployed workforce, the review discipline and escalation rules.
Why this is different from selling you agents
We are not describing a capability. We are deploying the thing this firm runs on.
Two founders deliver this company’s work by directing a governed AI workforce: a senior engineer seat that reads a brief against a version-controlled standards corpus and decomposes it, and nine seats that build in parallel, each to a written per-discipline standard. A review gate, an integration gate and a human decision sit between that work and any client. It is running today, and describing it openly is the offer.
So when we deploy a workforce inside your organisation, we are not implementing a vendor’s reference architecture. We map your function to our discipline set, then configure the review discipline, escalation rules and human decision boundaries for your context — because a compliance-checking workforce in a regulated firm needs a different boundary from a code-review workforce in a product team.
The governance is the deliverable as much as the automation. Anyone can wire agents together; the hard part is being able to answer “how do I know an AI did this properly?” with a documented method rather than a reassurance.
The line we do not move
Agents propose, build and verify. Humans decide anything a reasonable person could disagree about — and anything irreversible, legal, regulatory or safety-critical. Agents do not merge to default branches, deploy to production, delete data, or commit anything that speaks outside the box.
How it runs
Your function, mapped to a discipline set that already works.
The sequence matters: the operating model is configured before the workforce runs, not discovered afterwards.
- EngagementAI Workforce Deployment — 8–16 weeks
- Runs onYour environment, your data, your control
- MethodReview gate · integration gate · human decision
- GateEvery agent action logged; every claim carrying an evidence state
- Designed alongsideAI Operating Model Design
Pick one function, properly
Code review, incident triage, compliance checking, research synthesis. One function with a real owner beats four with a sponsor. The scope is what makes an eight-week deployment possible.
Map it to the discipline set
Which seats the function needs, what each is responsible for, and what its written standard has to say. Standards are short enough to be read and specific enough to be failed against.
Configure the gates for your context
What the review gate checks, what the integration gate checks, and exactly where the human decision sits. Escalation arrives with options and a recommendation, never as a bare question.
Run it under supervision, then hand it over
The workforce runs on real work while we are still accountable for it. Rejections against the written definition of done are the useful part of this phase.
Leave the handbook
How to operate it, how to change a standard, how to add a seat, how to stop it. Plus the audit trail design, so the record answers questions after we have gone.
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 deployed workforce, running in your environment on your data
- The review discipline, and the written standard behind each seat
- Escalation rules, and the human decision boundary as configured
- The audit trail — every agent action logged and attributable
- An operations handbook, including how to change it and how to stop it
What we won’t do
- Autonomous agents without human decision gates
- Workforces without audit trails
- Deployments without rollback or escalation
- Agents that merge, deploy, or spend without approval
- Opaque systems where you cannot inspect the review
The part most buyers underestimate
Speed without evidence is just risk with a schedule.
The workforce makes delivery fast. The review discipline is what makes the speed safe to buy — weeks not quarters, and an audit trail that still answers the question a year later. They are the same system, not a trade-off, and our own operation is the proof we are willing to publish.
Read the discipline in full →Or start smaller
Team Enablement puts the same discipline into your existing engineers without deploying a workforce — evals before prompt code, review as its own workstream, evidence states on every hand-back.
Want this running on your problem?
Thirty minutes with a founder. Tell us the function you would point it at and we will tell you how we would decompose it — and whether eight weeks is realistic.
Fixed fee · Quoted in writing before we start · NDA available