Lire en français

“Should we train everyone in AI?” The question is poorly framed.

There is training that amplifies a profession. And there is training for a workflow — where people learn to control a machine, to verify, to document, without producing the final deliverable themselves.
Mixing the two creates more confusion than progress.

Not everyone is meant to become an agentic AI expert.

Training everyone in the same way does not produce the equality we imagine: uniform training widens gaps.
Those who are already comfortable pull ahead; those who are further away fall behind — or drop off, which is much harder to detect.

What I aim for is equity of progression: tailoring support to what each role truly needs to amplify, and creating distinct spaces, outside delivery, where people learn how to control a machine.
But equity of progression does not mean everyone will end up authorised to do the same things.

In a regulated environment, the right to validate is not distributed. It is earned, demonstrated, and maintained. Giving everyone the means to progress is an obligation. Granting everyone the same signing authority would be a breach.

The cost, beyond designing the pathways, is the conversation that follows: telling someone competent they are not yet in a position to validate what an agent has produced.
If it has to take place, it is uncomfortable. There is no preparing for it.