WTB systems studio
Registry / 001
Every entry is a real configuration of context, capabilities, activation, authority and verification—grouped by what it actually is, not by what it would be flattering to call it.
10 deployed, 1 in cutover, 1 prototype, across 9 operating environments. Every entry carries its own status: deployed entries are running in production with client identities withheld; anything not deployed says so on its own page, and implies no outcome or integration. Entries are components, not clients — several belong to the same environment. 3 of 12 publish evidence you can check without taking our word for it; the rest currently rest on our own account. Of the 3, all show a mechanism in source; none yet shows operation in production or an outcome recorded by someone other than us.
8 of 12 entries contain no model at all. An LLM is optional — this is where that claim is checkable.
Interprets and returns. Holds no authority and routes nothing.
A deployed language capability that reads expense documents and returns structured fields. It interprets genuine ambiguity, holds no authority, and writes nothing — which is why it is not an agent.
A deployed pricing capability that suggests a resale price from comparable items already sold. It returns nothing rather than a guess when the evidence is too thin, and a human sets every price.
Executes a predefined path. No interpretation, no choice between actions.
A deployed deterministic automation that expires stale hold records in a live booking system. It runs unsupervised on a schedule, with no model and no decision in it—because the problem needs neither.
A deployed check that reads generated narration and refuses to let it reach voice synthesis when it carries a fault. Every rule it enforces was written after that exact fault shipped.
A deployed automation that sends consignment deadline notices at three fixed points around the end date. It sends the matching notice or nothing at all — there is no judgment in it, and none is needed.
Chooses, recommends or routes a permitted action from context.
A deployed agent issuing fiscal donation receipts for a non-profit operator surface, choosing per request whether it may act and refusing when the grant is absent.
An agent that verifies and fulfils commerce orders from an operator upload, with an explicit dry-run boundary before any live mutation. Cut over to the shared runtime; the first live fulfilment is still pending an operator.
A confidence-routed classifier that labels inbound message intent, resolves what it is sure of, and queues everything else for human review.
A deployed agent that turns retail orders into accounting invoices, deciding for each one whether to issue, skip or hand it to a human — and holding a lock so a duplicate webhook cannot invoice twice.
A deployed agent that updates live sites, then judges its own work from the outside and reverses it when the site comes back worse.
Combines automations, agents, humans and controls to deliver an outcome.
A deployed publishing system for a financial-education library. The build recomputes every figure an author declares and refuses to produce a lesson whose arithmetic does not hold.
A deployed managed product that turns a reference image and product packshots into brand imagery. Every model step is gated by an operator, and both gates exist because their absence already caused a failure.
Start with the operating problem
A first call is thirty minutes and costs nothing. We say honestly whether we are the right people. You would be talking to the people at We Think Beautiful in Brussels — the ones who do the work, not an account layer above them.