Engagement · Assessment first
You want an agent that does the work, not one that answers questions.
Most AI pilots stop at a chat box. The hard part is what happens after the answer — the permission to act, the record of what was done, and the ability to undo it.
What usually brings people here
The demo worked. Production did not.
These are the failures we hear about most often. If two or more sound familiar, the problem is usually the operating layer around the model, not the model.
- The pilot impressed everyone and then nobody could put it in front of a customer.
- It gives a confident answer and nobody can tell where the answer came from.
- It cannot take an action, so a human still re-types everything into the real system.
- Nobody will sign off on letting it act, because there is no approval step to sign.
- It worked in one team's hands and broke the moment anyone else used it.
- When it got something wrong, there was no way to see what it had done.
How we approach it
An agent is an accountable actor, not a smarter prompt.
The interesting engineering is not getting a model to answer. It is deciding what the agent is allowed to do, proving what it did, and being able to reverse it.
Permission before capability
What the agent may touch is designed first. A capable agent with undefined authority is a liability, not an asset.
Every action leaves a record
Actions are written to an append-only ledger. Months later, someone who was not there can still reconstruct what happened.
Reversible by design
If a step can change the real world, there is a defined way to undo it — decided at design time, not during the incident.
What an engagement produces
A mission with named agents, and a ledger that survives audit.
This is the shape of the artefact, not a customer's project. It is the standard we would hold the work to, so you can judge that standard before anyone discusses scope.
Illustrative sample from the OneAI agent stack. It is not a delivered customer engagement, and we do not present it as one.
How it starts
We assess before we agree to build.
The first outcome is a verdict, not a proposal. Sometimes the verdict is that an agent is the wrong tool, and saying so early is worth more to you than a scoped project that should not exist.
Describe the scenario
What decision or task you want handled, who is accountable for it today, and what system it has to touch.
Feasibility assessment
We look at the data, the permissions, the integration surface and the failure cost. This is where most of the thinking happens.
A verdict you can act on
Viable, viable with conditions, or not worth building. You get the reasoning either way, and you own it.
A pilot, only if viable
Scope, boundaries and success criteria agreed in writing before any build starts.
Who does the work
Founder-led, which is why we run 2 at a time.
Engagements are led by the founder directly rather than handed to an account team, so the person who designed this stack is the person on your project. That is a real constraint, and we would rather state it than queue you behind something.
One principal, start to finish
The same person runs the assessment, the design and the build. Nothing is re-explained to a delivery team that was not in the room.
Permissions designed first
You get the authority model in writing — what the agent may touch, what needs approval, and what it may never do.
The agent, and its ledger
Not just a working agent: the audit configuration, the rollback path and the integration that puts it inside your system.
Handover, not dependency
Documentation and a walkthrough so your team can run it. If you would rather we kept operating it, that is a separate agreement, agreed up front.
What we are not saying
- 2 engagements at a time is a real ceiling, not a scarcity tactic. If they are all running, we will tell you when rather than take a deposit.
- We have no customer case studies to show you yet. Everything on this page is our own system and our own standard, labelled as such.
- If the honest answer is that you do not need an agent, that is the answer you will get — even though it is the smaller engagement.
Tell us what you want an agent to be responsible for.
The more concrete the scenario, the more useful the assessment. A real task with a real owner beats a category.
Your AI should not stop at a demo.
Route intelligence through Core, govern it with TheOne, persist the work in OneMission, execute through OneClaw and keep verified outcomes in OneField.