Back to the cube

AI & work / 1 min read

The work after the demo.

A compelling first result is the beginning of an operating model.

A new set of questions

A good demo makes a possibility tangible. I can see a workflow taking shape and imagine how it might help. The next step, though, changes the questions. What happens with incomplete information? Who notices when an external service stops responding? How does someone correct a result without starting again?

These questions are design material. They reveal what the system must explain to its users and what the people operating it need to see. I want to answer them while the workflow is still small enough to change easily, before a polished interface makes uncertain behavior feel settled.

Define a complete attempt

For an AI-assisted workflow, I would write down what counts as one attempt. It needs a beginning, an allowed set of actions, a stopping point, and a result that can be checked. If it stops halfway through, the next attempt should know what already happened. Repeating a request should not silently repeat an external action.

I also want an explicit path for uncertainty. A system that asks for a decision at the right time can be much more useful than one that confidently fills every gap. The request should include the evidence, the unresolved issue, and the smallest choice needed to move forward.

Observe the outcome

My final question is simple: how will I know that the intended result actually happened? A successful internal step might mean a draft was created. The useful outcome might require someone to receive it, review it, or act on it. I would keep those states distinct. That makes the workflow easier to operate and gives the next improvement a concrete place to begin. The work after the demo turns possibility into something someone can depend on.