Why AI Literacy Is Not a Training Problem
Organisations keep booking trainings and wondering why nothing changes. The problem was never the curriculum.
Six weeks after one of the better trainings I have run, I sat with the same team again. In the workshop they had asked sharp questions, built real prompts, left energised. Now I asked who had used any of it since. Two hands out of fourteen. The training had been rated 4.7 out of 5.
That number should worry you more than a bad rating would. It means the format worked exactly as designed, and the design measures the wrong thing. Trainings measure satisfaction on the day. Capability shows up in workflows, weeks later, or it does not show up at all.
The research behind this is old and keeps being rediscovered. Roughly ten percent of a capability shift is the technology. Twenty percent is data and tooling. Seventy percent is people and process. A training addresses the ten percent and sends everyone back into an unchanged seventy.
A skill decays in isolation. It compounds in an environment.
Literacy, it turns out, behaves less like knowledge and more like a property of the environment. Four things decide whether it takes root. A shared vocabulary, so people can talk about what they are doing without translating for each other. Explicit permission to experiment, which in most German organisations means someone senior saying it out loud, twice. A colleague within reach who is slightly further along. And context: the tool has to meet real work on Monday morning, or it meets nothing.
None of these four are on the agenda of a training.
What does work is slower and less photogenic. A champion inside each team, not a central AI office three floors away. Artefacts that leave the room with the participants: a canvas filled with their own process, a working prompt chain for their own report, not an example one. Leaders who visibly use the tools themselves, badly if necessary. Badly is fine. Invisible is fatal. And a rhythm of returning, because this field rewrites itself every ninety days and a one-off intervention starts expiring the moment the projector goes dark.
I still run trainings. They are a fine ignition. But ignition without fuel is a spark in an empty room.
If you hold a budget for AI literacy, here is the reallocation I would argue for: spend a third on the workshop, and two thirds on what surrounds it. The champions, the artefacts, the follow-up rhythm, the visible sponsorship. The teams that learn fastest are never the best-trained ones. They are the ones where trying things is normal.