Teaching people to use these tools well. It is less about features and more about judgement: what to delegate, how to check the output, when not to use it at all. A two-hour session on prompting produces very little; ongoing support tied to people’s real work produces a lot.
Technical view
Effective programmes are role-specific rather than generic, use the organisation’s own tasks as examples, cover verification practice and failure modes explicitly, and provide a shared library of vetted prompts and patterns. Include policy content on what data may be used where. Measure capability through observed work quality rather than course completion, and provide a route for questions after training, since most learning happens at the point of use.
Related
Part of AI Terminology Index, section: Product and Adoption.