The people who get the most out of automated systems are not the ones who use them most. In the working sessions our councils ran this year, the highest performers were consistently the ones who used them most selectively, and who had a clear internal rule for when to stop.
What the good ones actually do differently
- They frame the task before they delegate it, rather than iterating a vague prompt.
- They verify against a source, not against how convincing the answer sounds.
- They know which decisions they will never delegate, and can say why.
- They take the result to a colleague as their own work, not as the tool’s output.
Everyone in the room had access to the same tools. The difference in output quality was almost entirely down to who checked, and how.
Why it needed its own standard
We resisted creating this standard for a while. The argument against was that it is a way of working rather than a discipline. What changed our position was employer evidence: organisations were making hiring and promotion decisions on this capability while having no consistent way to describe or test it.
The standard runs to four levels rather than six. At Level 1 you use the tools competently and know where they fail. At Level 4 you set the organisational practice for how people and systems divide the work — including what is off-limits.
The accountability point
The clause that generated the most discussion was the simplest one. Whoever submits the work owns the work. An automated first draft changes nothing about who answers for the result, and the assessment is written to make that explicit.