Why Leaders Go First: Practice, Honesty and Direction

Most organisations adopt AI tools through a mix of enthusiasm and caution, and staff watch their leaders to learn which one to follow. What a leader does with the tools, and what she says about them, shapes the culture more than any policy does.

Try it yourself, on approved material

A leader who directs AI use without having used it is working from other people's accounts, and those accounts tend to be tidy. Before asking a team to adopt a tool, take a real task from your own work and do it with the approved tool, using material you are permitted to put into it. Choose something with a little awkwardness: an incomplete brief, a document with a gap, a question where you already know the answer. Note what you asked, what came back and what you had to fix.

The words "approved material" matter. The trial is not a licence to test the tool on confidential or personal data to see how it copes. If the task you want to try needs such data, that is a question for whoever owns the approval.

Check the facts yourself

AI tools produce fluent text whether or not it is correct, and fluency is not evidence. When a leader quotes a figure, a date, a clause or a reference that came from an AI draft, she is the person who put her name to it. Check the claims that matter against the source: the contract, the ledger, the primary text of the rule. Asking the tool to confirm its own answer does not count as a check, because the same weakness can produce the same error twice. If you find one mistake in a draft, treat it as information about the whole draft and widen the check, rather than patching the one error and moving on.

Be open about where AI helped

The better habit is a short, specific statement: AI helped draft this section; I checked the figures against the ledger; our counsel reviewed the legal wording. That tells readers how much trust to place in each part. It also shows the team that disclosure is ordinary and carries no stigma. Do not swing the other way and label work as "written by AI" to deflect responsibility. The author is accountable for the final text, whatever drafted it.

The same honesty applies when you describe results. Showing only the three impressive outputs from a pilot creates a picture that the next person to use the tool will not recognise. Show a fair mix, including outputs that needed heavy rework, and say how the examples were chosen. If the team has quietly discarded poor samples, ask for them back before you present.

AI literacy is part of the job

Leaders who approve, fund or direct AI use need enough understanding to ask good questions: what the tool is for, how it fails, who checks it, and what happens when it is wrong. That is not the same as understanding how models are built. In EU law, Article 4 of the AI Act places a duty on organisations to take measures supporting AI literacy among their staff and others operating AI systems on their behalf. It is an obligation of means rather than of result. It is proportionate to role, context and the people affected, it applies to AI systems generally and not only high-risk ones, and it does not mandate any certificate. Whether or not that law reaches your organisation, the principle is sound practice: depth should follow reliance and consequence.

Write a direction statement people can use

A direction statement says what AI is for in your area and what it is not for. "Use AI responsibly to improve service" cannot be acted on, because nobody can tell whether a given use complies. "AI may draft and summarise internal material that staff then check; it may not decide eligibility or be given personal data" can be acted on. A good statement is short, names the purposes, names the exclusions, and says who to ask when a case is unclear. Pair it with a plain map of what is approved, what needs approval first and what is off limits, so that staff are not left guessing.

Worked example

Harbourlight Housing Association, an invented organisation, has an approved assistant for drafting tenant letters. The director of customer services, Amara, tried it on three of her own letters before briefing her team, using template text with no tenant details. She found that it had invented a policy reference in one letter. At the team meeting she showed that example beside a good one, said which parts of her own letters she had rewritten, and announced a one-page statement: the assistant may draft letters from templates, staff must check each policy reference against the housing manual, and tenant records may not be entered into it. Amara had not claimed the tool was safe. She had shown how she used it and what she still did herself, and that made it easier for others to be candid.

As of 24 September 2026.