Done Means You Can Check It

When someone says a job is done, there are two different things they might mean. One is that they believe it is done. The other is that you can look and see.

At Kreygo we build custom AI agents, or small teams of agents, around how a business already works. One rule runs through all of it: a task counts as finished only when it can be checked.

What does "checked" look like? It is something you can point to. A file that was changed. A message that was delivered. A confirmation that came back. If the only evidence is the agent telling you it succeeded, the task is not finished. It is still in progress.

We hold ourselves to the same rule when we build. Our process has five steps: discovery, proposal, build, staged go-live and handover. At the go-live step, each stage is accepted against an external check, meaning a check that does not depend on our word. The last step, handover, ends with the client's sign-off.

Why bother? Because the cost of an unchecked "done" is paid later and by someone else. A report that was never sent, a change that was never made, a record that does not match: these are annoying to find a week on, and they are hard to trace when nobody wrote down what "finished" was supposed to mean.

So we ask the question early. During discovery we talk through the workflow and the goal, and we write down what the agent will not do. Before anything starts, the scope is agreed in writing. A written scope gives both sides something to check the work against.

A few honest limits. A check proves the thing it checks and nothing more. We are not saying an agent is perfect, and we are not putting a number on how often anything happens. We are saying that, for each task, we can show you what finished looks like, and you can see for yourself.

If you are weighing up any automation, ours or someone else's, try this. Pick one task you want taken off your team's hands. Ask: how will I know it is done, without taking anyone's word for it? If the answer is clear, you are in good shape. If the answer is vague, settle that before you build anything.

That is the question we start every project with. If you have a task in mind, you can describe it to us and we will tell you whether an agent fits.