Company in
Point us at a company — its website and its careers page. Nothing else required from them.
The architecture · company in → AI workforce out
Drop a company in.
Get its operation back, automated.
We don't ask what a company does — we work it out from what it has already told the world. Then we build the AI that runs it, reason where judgement is needed, and hand back only the humans that truly remain. Every role becomes an ai66: the job, distilled and rebuilt as a working system.
The pipeline
Seven moves from a website to a working AI workforce.
Every vacancy — current and historical
Today's openings show what they're hiring now; years of past postings reveal every function and output the company has ever defined. A job advert is the one public document where an organisation states a problem, attaches a budget and names the decision-maker. That's the real shape of the operation.
spine of the researchDeep research — every term a node
Each system, skill or framework a JD names — SAP, Lean, GMP — is a pointer into a universal node we research once, to advanced level, and reuse for every company that ever touches it. Competitor structures, the org chart, filings, regulator and news triangulate the rest. Every claim is evidence-graded and falsifier-gated — nothing renders that can't be re-derived from a source.
the graph that compoundsReverse-engineer the org
From all of it we build a blueprint of the entire operation, decomposed down the capability spine — every asset, skill, role and how they connect.
Asset → Skill → Role → Team → Dept → OrgFill roles — AI builds the system
The bulk of the work becomes programmatic, codified automation that just runs — deterministic, durable, no AI in the loop per task. Cheapest and most reliable.
AI builds itAI reasoning — only where it's needed
A live, grounded agent handles the genuine-judgement slices: ambiguity, synthesis, language, the board report. Used sparingly, on real data, with every figure traceable.
AI runs itHuman remains
What's left is the irreducible core — accountability, relationships, regulated sign-off, the calls a machine shouldn't own. Fewer people, each doing only what truly needs a human. We never put a verdict against a named person — the analysis stops at the function.
human keeps it
Photography · Unsplash
The split — what the blueprint decides
Every task lands in one of three buckets.
This is the whole point. Once the operation is mapped, every task is sorted — and that sort is the answer to "what can AI do for us?"
AI builds a system
Programmatic
Codified automation that runs itself. No model in the loop per task — deterministic and durable.
the bulk · cheapestAI used where needed
Reasoning
A grounded agent for the judgement slices only — on live data, every number traceable. Used sparingly.
a thin slice · on demandWhat's left is human
Irreducible
Accountability, relationships, regulated sign-off. The people who remain get the work that's truly theirs.
fewer · expertThe unit
Every role becomes an ai66.
One vacancy → one ai66: the codifiable bulk of the job running live, the residual human core named explicitly, evidence and falsifiers attached, served from one role-agnostic engine. A hiring wave is just a countable set of them — "automate these N, hire these M, here's the evidence."
Asset→Skill→Role→Job→Team→Dept→Org→Sector
Capture the spine once and it drives everything — the automation, the reasoning, the candidate training, and the org chart, all rendered from the same structure.
What a vacancy is worth
Four ways a single vacancy pays.
The advert isn't just the research seed — it's the qualified lead and the channel. Each role can exit the graph four ways, and the wave is the deal.
E1
Capability sale — or Monty
The role's codifiable bulk delivered as a working system, priced against the advertised salary. Or deployed as Monty: a named, metered virtual employee — every action evidence-linked, every customer-impacting move gated by a human's approval. The only virtual employee with an audit trail.
E2
Placement — the anti-recruiter
The right human for the residual core. We match reality to reality — the role as it will actually exist, not the JD fiction — and we're the only recruiter who knows the job isn't the JD.
E3
Training
Close a candidate's gap to the residual-human spine — aimed only at the appreciating skills, on the organisation's actual software stack.
E4
Assessment → implementation
A capability assessment against our data and theirs, then the build — scoped per wave.
The proof is live
Not a slide — two running rigs and a queryable spine.
ENSEK (a mid-market energy platform) and Thermo Fisher's UK operation (enterprise scale) are both reverse-engineered and live, every role distilled with its evidence behind it.
165roles distilled
136claims · all evidence-linked
84spine tables · MotherDuck
2live rigs
Independent, not affiliated; company marks used for identification only; every figure on every rig is re-derived from a cited source, or it doesn't appear. The capability graph compounds — each company researched makes the next one cheaper, because most of its nodes are already enriched.
Inside your own walls
Or run the whole engine on your own data, your own keys.
The same engine deploys inside your tenancy, calling your model keys — or a local model on your own silicon. Your operational data never crosses our wire. We build the tool; you run it. Like Microsoft ships Excel, not like a vendor hosting your spreadsheet.
The output
"You wanted to hire X, Y and Z."
Here are the systems we'd build, the thin slice where AI earns its keep, and the only humans you truly need — running on a realistic replica of your own data. anaigent does the build; itsorted does the call.
Get it sorted →