Main idea

The skill expected of an employee working with an AI was first framed as a skill of formulation: knowing how to ask, how to frame, how to specify. It is the prompt that has drawn the attention, because it is the only visible gesture.

But what makes the difference between two results is decided before the request, in what the context already contains. Bringing in material, correcting a false interpretation before it gets picked up, validating a source, writing a rule, creating a template, indicating what you are working on right now, flagging a contradiction between two documents, and deciding what deserves to be kept — those are the gestures that determine the quality of the following answers.

The employee then stops being a user filling in fields. They tend a thinking environment: a well-kept context reduces the cost of the decisions to come, a badly kept context produces bad summaries faster.

Why it matters

It shifts what has to be trained, and what has to be recognized as work. Prompt training is organized in half a day and is hard to measure; tending a context is continuous work, with no deliverable of its own, and no job description covers it.

It also explains why two teams with the same tool obtain incomparable results. The difference is not in the tool, and the gap widens over time, since a well-kept context grows richer with every use.

Nuances and limits

Formulation keeps its share: asking badly from a good context gives a bad result. The point is not that the prompt does not matter, it is that it plateaus fast.

And tending a context can become endless work, filing material nobody will query again. The rule that bounds it is the same as for governance: tend only what will be drawn on again.

Open questions

• How does an organization recognize context-tending work that produces no identifiable deliverable?