Work you have proven starts running in your own systems. A person still checks every result.
This is the step most people want to start with, and the one that fails hardest when they do.
Step by step, including the awkward exceptions.
Not just the happy path. The exceptions are where unattended work goes wrong.
By different people, on different jobs, with the same result.
Once is an anecdote. A dozen times across a few people is a pattern.
You know what it costs, what it saves, and what going wrong would cost.
That third number is the one that decides how tight the review point needs to be.
A named person inside your business, responsible and able to change it.
If nobody owns it, nobody notices when it quietly stops working.
Inside tools your business already pays for and has approved.
Rather than adding another subscription and another login for somebody to manage.
Nothing reaches a client, a ledger or a decision without a person approving it.
Designed in from the start, not bolted on later when somebody gets nervous.
What ran, when, what it produced, who signed it off.
So when somebody asks in six months, there is an answer.
One obvious way to turn it off, with the manual process kept intact.
If it needs to stop on a Friday afternoon, it can stop on a Friday afternoon.
Documented plainly enough that your owner does not need to call us.
If they do, we have built you a dependency rather than a capability.
Agents run the repeatable steps. A person reviews every result before it counts.
If you have heard the word agent and wondered what it means in practice, this is it: a set of steps you have already proven, running on their own inside your systems, stopping at a review point you control.
The word matters less than the two rules around it. Nothing runs without a review point, and nothing runs that your own team cannot explain, change or switch off.
Nothing sends work out with no person accountable for the result.
That is not caution. It is the whole design, and it is why this step is safe to take at all.
If your team cannot explain to a client what happened, we have not finished.
Anything we build has to be explainable in plain words by the person who owns it.
We will not build on top of work that has not been proven manually first.
Speeding up a broken process only gets you to the wrong answer faster.
Each build is quoted as its own fixed scope piece, with the review points, the record keeping and the handover included rather than treated as extras. Everything runs in your own systems, under your own accounts.
When AI has changed how the work gets done, the commercials need to catch up.
The check will show you where you actually are.
Get your free AI Readiness Check