Your Prompt Library Is Working Against Your Team

By Martin Bergmann | AI Project Lab | July 2026

Most AI adoption programs built by project managers start the same way: gather the best prompts, package them into a library, and hand them out on day one. It feels like good program design. Lower the barrier, give people a head start, make it easy to say yes.

Cognitive science says this is exactly backward.

In 1978, psychologists Slamecka and Graf ran a deceptively simple experiment. Participants who generated a word by completing a partial cue retained it significantly better than those who simply read the same word. The effect held across decades of replication and extended well beyond vocabulary: information you produce yourself is encoded two to three times more durably than information you receive passively. The effort of generation is not a friction in the learning process. It is the process.

This is the Generation Effect, and it has a direct implication for every PM designing an AI adoption program right now.

When you hand your team a prompt library on day one, you are optimizing for immediate output quality. That is a reasonable instinct — early wins build momentum, and nobody wants to watch a team struggle. But what you are trading away is long-term retention. The team member who receives a polished prompt for summarizing meeting notes will use it once or twice and then forget it when the context changes slightly. The team member who wrote a bad prompt, saw it fail, and revised it three times has built something that will transfer.

The training program that looks roughest in week one is often the one that lands deepest in month six.

This matters beyond productivity. Real AI literacy — the kind that lets someone adapt when the tool changes, apply reasoning to a new problem, catch an output that is confidently wrong — cannot be handed to people. It has to be generated by them. A prompt library trains people to retrieve. Struggle trains people to think.

What to do instead

This is not an argument against sharing good prompts. It is an argument about sequence.

The design principle is simple: the first attempt always belongs to the learner. Ask your team to write their own prompt before you show them a better one. Let them see where it falls short. Then introduce the example — not as the answer, but as a comparison. The gap between what they wrote and what works is where the learning lives.

If you are designing an AI onboarding session, build in at least one task where the participant has no template, no example, and no starting point. The discomfort you see is not a design flaw. It is the investment.

A prompt library is a useful reference. It is not a training program. The distinction matters because organizations that confuse the two will develop teams that can follow AI instructions but cannot think with AI tools.

That is not AI literacy. That is AI dependency.