I've never actually done one from scratch.
Newer people produce work they don't understand
Newer people use AI to produce work they couldn't produce on their own. The work may look competent, while the person builds less of the understanding they would need to judge it, fix it, or do something different when the situation calls for it.
- Newer people can't do core tasks without the tool.
- Their output looks better than their understanding.
- Senior people say the new hires are fast but aren't sure they get it.
Who: two newer people and the person who trains them. Ask the newer people to walk through one recent piece of AI-assisted work and explain why it's right. You're probably looking at this pattern if they can't explain the reasoning or spot what would make it wrong. Probably not if they can explain it and defend it.
The Change Barriers entry Craft loss, which describes the same thing as a barrier to a change. In this pattern, it's something to look for in one workflow.
Keep one junior person in a room they would normally miss, and have someone explain the thinking as it happens.
Pick one skill AI now handles and give a junior person one real instance of it this month, with an experienced person reviewing how they thought it through.
Before advancing someone, ask for one piece of work that shows the skill with the model not doing the hard part.
In robotic surgery, trainees got far less hands-on practice than under the older methods, and the ones who became skilled mostly did it by working around the official training (Beane, 2019). Established pattern, from a field study. Entry-level employment fell for workers aged 22 to 25 in most of the jobs most exposed to AI (Brynjolfsson, Chandar, and Chen, 2025). Emerging.