Recipe 3

Learn from your peers¶

Try this first. Hold a show-and-tell, then pair your most experienced users with your least experienced.

Summary¶

Agent use varies a lot, even within one lab. General training courses often do not reach the people who use agents least. What worked was peer learning: a round in which everyone shows what they actually do, followed by pairing the most experienced users with the least experienced.

Steps¶

Hold a two-hour session where everyone shows how they use agents.
Pair your most experienced users with your least experienced.
Before expanding use, hold a session on what could go wrong.
Keep one shared skills repository, including skills that help people write new skills.
Hold an open drop-in meeting for several groups.

How to check the output¶

After three months, ask again who uses agents every week, and for what.

Examples¶

In one group, about a fifth of members hardly used AI and about a fifth used it heavily.
A first hands-on workshop on programming with agents was too fast for beginners, and they dropped out.
Pairing experienced users with beginners increased use clearly.
A later session on what could go wrong was rated among the most useful.
Skills also cut costs: an agent that follows a tested skill uses far fewer tokens.

Known problems¶

Training designed by and for heavy users leaves beginners further behind.
In the centre's survey, the most common obstacle was time to learn (53% of respondents).

Skills¶

Each skill page has a brief. Give it to your agent to build a version for your own institute.

package-to-skill: turns the documentation of a software package into a skill an agent can load.

Best practices on aiforscience.eu that this recipe follows:

Match the method to the task: new users need to learn when not to use an agent.

Change history¶

2026-10-30: Renumbered to follow point 3 of the submitted Comment (formerly "Pairing and shared skills"). The shared skills repository example moved to recipe 1.
2026-10-28: Updated to match the presubmission draft of the Comment.
2026-09-25: Updated to match the first full draft of the Comment.
2026-09-25: First draft, from the first round-table.