Find your internal AI champions and make their methods reusable

Two people in the same role may use AI very differently: one drafts emails while another prepares a complete case for review. Compare their actual output and working conditions before inferring a productivity gap.

An internal champion can describe a useful task, supervise its execution and help another person repeat the method. That contribution is observable; it does not require a new status hierarchy inside the company.

An adoption survey cannot identify these people for you. Ask teams to show a task they already improved, the evidence behind it and the support they need.

The opportunity is to make useful individual learning available to colleagues without spreading unreviewed tools or exposing data.

To understand who ends up on the right side of the gap, look at what an agent actually demands.

The old chain of knowledge work was search, click, read, decide, act. The agent replaces it with a shorter one: instruct, supervise, validate. That reads like a productivity upgrade. It is a change of profession. You stop doing the work and start specifying it, and specifying work is a skill your company has never hired for, tested, or trained.

Define the expected result, method, constraints, available tools and stop conditions before delegating work. A person who can explain the exceptions is often a strong partner for that design.

This is the same wall I keep pointing at from another angle. Your company runs on rules nobody wrote down. Agents demand the writing. The people who can do that writing, who can turn "handle the supplier claims" into a sequence of decisions with checks and stop conditions, hold the decisive skill of the next decade. Prompting was never it. A prompt is a sentence. A working agent is a formalized method.

A saved-prompt chatbot and a tool-using agent have different powers. Some Custom GPTs can call actions; the label alone settles nothing. Inspect the actual tools, permissions and approval requirements.

People start with different experience of files, data and access rights. Give them a practical way to learn these foundations instead of judging ability from their first chatbot result.

Training needs protected time and real tasks. Motivated early users can help test the approach, while everyone affected by the new workflow needs support to use it safely.

If methods remain in personal accounts, the company inherits a continuity risk. Recognition and documented transfer help retain both the person and the capability.

Support the people doing the first work

Start with a small group whose work can be observed and tested.

Invite people with useful experiments to explain what they did. Review data handling before sharing the workflow more widely.

Use hands-on workshops on real tasks and measure whether colleagues can repeat the method afterwards. Workshop size or enthusiasm does not guarantee results.

Store the method, agent design, checks and stop conditions in company-controlled systems. Give a second person access and ask them to operate it without the original author.

The objective is shared capability that survives staff changes and can be improved by the team.

Bring one useful experiment to a 30-minute Strategy Session. We will examine its next controlled step.