Governed work for people and their agents
Your people, their agents, one intent.
An intent is one piece of work — a line changeover, a migration, an audit response — written down as something a team and its agents can both work against. Intent Studio is where that gets built: questions, context and trade-offs, until it is sharp enough to act on.
A workspace for enterprise teams, running as a managed service or inside your own network. Try it now opens the app sign-in — book a demo if you’d rather be walked through it first.
In thirty seconds
The rest of this page is the same argument, made properly.
What it is
A workspace where a team and its agents build one shared understanding of a piece of work, before anything acts on it.
Who it is for
Enterprises where work crosses functions that do not share a vocabulary — a plant, a supply chain, a controls team and quality, on the same change.
What it does
Turns a thin instruction into something an agent can act on — where every line carries who said it, what it rests on, and whether anyone still disagrees.
Why it matters
You can widen what your agents decide for themselves, and still defend having let them.
Why agents fail
Agents don’t fail because the model is weak. They fail because the intent was thin.
Handed a thin intent, an agent does the worst possible thing: it proceeds. It fills every gap with an assumption — silently, confidently. Assumptions compound, because each one becomes the ground the next step stands on. By the time anything is produced, the plan rests on a stack of guesses nobody agreed to.
Where the gap opens
What was specified stays flat. What the agent built drifts a little at every step, and each step stands on the one before it.
Diagram — the gap is invisible while it is cheap to fix, and obvious once it is not.
That gap bills four ways at once
Hallucination
The agent asserts what it was never told and could not know.
Iteration
Round after round of “not quite, try again” — each one a full paid run.
Rework
The most expensive people in the building fixing an agent’s guesses.
Corrupted plans
Work that was wrong at step two and only visibly wrong at step twenty.
Why more assistants don’t fix it
A workspace of four people is not four conversations with four assistants.
It is four people and four delegated agents contributing to one understanding — each seeing what the others have done to it.
Four people, four agents, one intent
Every agent acts for a named person. Contributions converge on one record rather than scattering across five private threads.
Diagram — every agent acts for a named person, and can never exceed what that person could do.
Who changed what, and what it produced
Four contributors on one intent — people and agents side by side, each with their own revisions — and the knowledge bundle the work built.
Illustration — every contribution attributed by name, and the knowledge bundle the work produced.
The agent is in the room
Each member’s agent can reason over what the others — and their agents — proposed, contested, accepted or resolved, and bring that into the conversation where the work is actually happening. That is the coordination product, and it is the thing a per-user assistant structurally cannot have.
What an intent is
So what is an intent?
One piece of work — a line changeover, a migration, an audit response — held as a structured record instead of a thread. It is the one thing your people and their agents both work against, and the only new idea on this page.
What an intent is made of
Eight named parts, each one a question somebody has to answer before an agent can safely act. Nobody fills in a form — the answers accumulate as the work is discussed.
Illustration — the eight sections of one intent, each a question somebody answers before an agent acts.
What happens to it after it is written
Anyone on it adds to it — a person, or an agent that has just read something. Each addition is a claim, carrying where it came from.
A ticket holds whatever one person typed, once.
It changes shape as the work is understood. Sections get filled, claims get revised, and every version is kept.
A ticket is filed, then goes stale.
Disagreement is a state a claim can be in — visible, attributed, and open until somebody settles it.
A ticket hides disagreement in the comments.
It carries what an agent is allowed to act on, and how far.
A ticket tells an agent nothing about what it may do.
Which raises the next question: how does anything in an intent get solid enough to build on?
How an intent gets solid
Every line in an intent is a claim, and every claim has a state.
Nothing in an intent is text somebody typed. Each statement carries where it came from and how settled it is — which is how an agent tells something agreed from something still being argued about.
Proposed
Put forward by a person or their agent, carrying the passage or document it came from. Nobody has accepted it yet.
Agreed
A person accepted it, and the record says who. An agent cannot promote its own claim.
Contested
Somebody disagrees, and said why. The disagreement sits on the record instead of in a comment thread, and the work carries on around it.
The three states, on one intent
One agreed claim carrying the review it came from, and two the team is still arguing about — the changeover window, escalated because it has been raised twice and reopened.
Illustration — one intent’s understanding: what is proposed, what is contested, and the evidence behind each claim.
Readiness follows from those states rather than from a score. An intent reads as vague, actionable or ready — a number would only invite an argument about the number instead of about the work.
How much agents decide alone
You set the line. Then you move it.
Every other product picks a point on this and hard-codes it. Copilots stay propose-only, so the human is the throughput ceiling forever. Autonomous agents act by default, which is unsellable into a regulated process.
Propose
where you startNeeds: Nothing. This is the safe floor.
Suggest claims, questions and evidence. A person agrees to everything.
The agent may draft the changeover plan and flag the tooling lead time — but not book the shutdown.
Settle the immaterial
Needs: That materiality is defined by policy, not by the model’s judgement.
Close low-consequence items alone; escalate anything material.
The agent may settle the spares list on its own, and still escalate the shutdown length to a person.
Settle within a domain
Needs: A track record that is measured, and specific to that domain.
Act unsupervised inside an intent type where its record supports it.
The agent may resolve scheduling conflicts across every changeover it has run before.
Build and prepare
Needs: That the gates still hold — attestation and approval unchanged.
Produce artifacts and ready them for release without a human in each step.
The agent may produce the full changeover plan and the validation schedule, ready for sign-off.
Act, with gates
Needs: Halt and revoke are instant, and every step is reconstructable.
Carry work through, stopping only where policy demands a named human.
The agent may run the changeover end to end, stopping only at the safety sign-off.
At this setting, on INT-0142
The agent may draft the changeover plan and flag the tooling lead time — but not book the shutdown.
Use the arrow keys, or pick a position. Set per workspace, per intent type, per risk class.
What holds wherever you set it
Nothing happens that is not attributable, grounded, bounded, and revocable.
That envelope is what makes the line safe to move at all — and why widening it is a decision you can defend afterwards rather than a leap of faith.
What you keep
Every intent leaves behind more than its artifact.
The claims and the evidence under them, the decisions and the reasons given, the systems touched, the prior intents it built on. Each one is a node and an edge.
Those accumulate into a knowledge graph of how your organisation actually works — not an org chart or a wiki, but a grounded record of real decisions, real constraints and real outcomes. Every intent after it starts richer than the last.
It is also what makes widening autonomy defensible: you can only grant more where you can measure what happened, and this is the measurement.
What the second intent starts with
One intent, and the things it leaves behind. The next intent that touches any of them arrives already connected — it does not re-establish the lead time, re-litigate the part, or re-run the validation.
Diagram — the knowledge graph around two intents. Drawn, not captured: this is the shape of the record, not a screen.
Who this is for
Every team has intents. Almost none of them look alike.
The inputs and the outputs differ enormously — documents, telemetry, code, machines. What does not differ is that somebody has to work out what is actually being asked for.
Manufacturing & plant
Plan a line changeover · qualify a second supplier · schedule a maintenance window
→ A change plan the line, the supplier and quality all worked from
Field service & infrastructure
Plan an outage window · commission new equipment · agree what done looks like
→ A work plan with its approvals attached
Operations & finance
“Take a quarter of the cost out of the Q3 plan — work out what has to change”
→ An analysis and a plan, evidenced
Software & engineering
Ship a control-system change with the line · settle a contested requirement · run an RCA
→ A spec an agent or a developer can act on
The ask
See it on one of your own intents.
The fastest way to understand Intent Studio is to walk something you are actually working on through it. Half an hour, your intent, no slides.
A demo is a conversation with the people who build it. Try it now opens the app sign-in.
Next up — how it actually works.
Product