AI is an amplifier. You still need a point of view.

Imagine a colleague who can produce twelve versions of your proposal before you have found the good mug. Useful? Absolutely. Ready to decide what your customer needs, what your team can promise, and whether the whole proposal should exist? That is a different conversation.

This is how I prefer to think about AI: an amplifier of human capability. It can give an idea more room, help explore unfamiliar territory, and take some repetition out of a working day. The value depends on the person choosing the direction and checking what comes back.

An amplifier turns up everything

Give AI a clear problem, useful context, and a sensible definition of success, and you have the ingredients for a productive experiment. Give it an unresolved argument dressed as a brief, and you may get a very polished unresolved argument. Now in bullet points.

Speed does not settle questions about purpose. If nobody knows who a report serves, generating it faster simply gets you to the confusion earlier. Before opening another tool, ask who will use the result, what decision it supports, and what would make it useful to them. These are human conversations worth having.

Calling AI an amplifier is an intention for how to use it, not a promise that every job will stay unchanged. Tasks can disappear, responsibilities can move, and some roles may shrink or be redesigned. People deserve an honest discussion about those possibilities. A cheerful workshop cannot substitute for that discussion.

Keep your depth. Give it some neighbours.

There is still value in being excellent at something. Knowing your craft helps you recognise when an answer is impressive nonsense. But being able to connect that craft to other kinds of work makes you more useful: a designer who can ask good research questions, an operations specialist who can explain a customer problem, a manager who can test an idea before assigning a project.

You do not need to become six full-time professionals wearing one increasingly tired face. Try one adjacent skill at a time. If you write, try mapping the process behind the thing you describe. If you manage projects, try interviewing the people using the result. If you analyse numbers, practise explaining the decision those numbers should inform.

AI can be a practice partner here. Ask for alternative questions, a rough structure, or a critique of an early attempt. Keep enough contact with the underlying work to know whether you are learning something or simply receiving nicer-looking homework.

Experiment in turns, not in seventeen tabs

Being adaptable does not mean interrupting yourself all day. Give an experiment a boundary: one task, one question, one short session. For example, can an AI-generated outline help you prepare a clearer project update? Compare it with your normal approach. Note what improved, what needed repair, and whether the repair was worth it.

Then stop and decide. Keep the method, change it, or retire it without a farewell ceremony. Repeating this across different tasks builds a useful repertoire. Constantly switching between tools mostly builds a very personal relationship with your browser's memory warning.

Learn the task before handing it over

A good handover requires understanding. You need to know the inputs, the awkward exceptions, the audience, and the point where a mistake becomes expensive. You do not have to perform every step forever. You do need enough knowledge to explain the work and judge the result.

Take a customer email. You can ask AI to draft it, but you should already know the customer's concern, the facts you can support, and the commitments you are allowed to make. The instruction should include those boundaries. Then you check the draft against them.

A useful delegation note has five parts:

• The job: what needs to be produced and who will use it.
• The context: the relevant facts and approved source material.
• The boundaries: what must not be invented, exposed, promised, or changed.
• The standard: an example or clear description of an acceptable result.
• The handback: what needs a human decision before anything leaves the room.

Keep private or sensitive information within your organisation's approved tools and rules. For an early experiment, made-up or anonymised material may be enough. There is no prize for using the most sensitive spreadsheet in the building.

Review is part of the work

A fluent answer deserves the same scrutiny as any other draft. Check facts against their sources, challenge assumptions, and look for missing context. For important decisions, involve someone qualified to judge the subject. Asking the same tool whether it is sure is a conversation, not independent verification.

The NIST AI Risk Management Framework treats trustworthiness as something to consider throughout the design, use, and evaluation of AI. My practical translation: deciding how to check the work belongs in the plan from the beginning.

Leave room for the human part

If AI saves you time preparing a meeting, use some of that time to listen properly in the meeting. If it helps you explore ten ideas, spend your attention choosing one for a reason. If it clears repetitive work, talk with the team about what deserves that space.

I want better tools to make us more curious, more capable, and more available to each other. That requires an active person at the centre: someone willing to try, able to judge, and responsible enough to say, “This looks good, but we are not sending it yet.”