The executive's first ninety days with AI

Most leaders meet AI in one of two ways. Either they try it once, receive a fluent but generic answer and conclude it is overrated, or they are shown an impressive demonstration and conclude it will change everything by next quarter. Neither conclusion survives contact with real work.

Ninety days is long enough to form a judgement based on your own use rather than someone else's claims, and short enough to keep the effort focused. The sequence in this guide is deliberately ordered: learn first, delegate second, and decide what to refuse third. Each phase depends on the one before it. Leaders who start by delegating — rolling tools out to their teams before they have used them seriously themselves — tend to set policies they do not understand and expectations that cannot be met.

The aim at the end of ninety days is not expertise. It is informed judgement: a clear, personal sense of where AI improves your work and your team's work, where it does not, and where it should not be used at all.

Three things need to be settled before day one. They take an afternoon and they prevent most of the problems that follow.

The first month is about personal fluency. The goal is to use AI on your own work, every working day, until you have an accurate feel for what it does well and where it fails.

Choose five recurring tasks from your own week. Good candidates are tasks that take real time, involve reading or writing, and where you can easily judge the quality of the result. For most executives the list looks something like this.

Use the tool for each of these at least once a week, and do the work carefully. The difference between a poor result and a useful one is almost always in the request. A one-line question produces a generic answer. A request that gives the tool a role, the relevant context, the material to work on, the steps to follow, the rules to respect and the format of the answer produces something much closer to what you need.

By the end of the first month you should notice three things. First, the tool is fastest where you already know what good looks like, because you can judge and correct the result in seconds. Second, it is weakest where accuracy depends on facts it does not have — your organisation's figures, recent events, specific people. Third, it is confidently wrong more often than you would expect, which is the most important lesson of the month.

Keep particular note of failures, because they shape every later decision.

With a month of experience, you can now make informed decisions about delegation — in two directions. Some work can be delegated to the tool, with you reviewing the result. Some of your new knowledge can be delegated to your team, so that they adopt AI with the benefit of what you have learned.

Start with the tasks from your log that consistently produced useful results with light review. These are the candidates for routine use. For each one, write down the request that worked best, the checks you make before relying on the result, and the situations where you would not use the tool for that task. That short note is a working procedure, and it is far more useful to your team than general encouragement to "try AI".

Then work with your team. Share the procedures. Ask each person to choose two or three of their own recurring tasks and run the same thirty-day experiment you have just completed, keeping the same kind of log. Agree how results will be reviewed and by whom.

Some tasks delegate well, and some do not. The pattern is usually consistent across organisations.

The second month is also the right time to agree a few simple norms with your team, before habits form on their own.

The final month is about deciding, deliberately, what you will not use AI for. This is the phase most leaders skip, and it is the one that most determines whether adoption goes well.

Refusal is not a lack of ambition. It is a statement about what matters. When a leader says clearly that certain work will always be done by people, it gives everyone else permission to use AI confidently for everything else, because the boundaries are clear.

The refusal list differs between organisations, but most experienced leaders arrive at something close to the following.

Write your list down, share it with your team, and explain the reasoning for each item. Revisit it in six months; some items may move as tools and policies change. But make the list explicit, because an unspoken boundary is one that will be crossed by accident.

At the end of ninety days, a leader who has followed this sequence should be able to say the following with confidence.

That last point matters most. It means the next conversation about AI in your organisation — with your board, your peers or your team — will be based on what you have observed rather than what you have been told.

Four mistakes recur in the first ninety days, and all of them are avoidable.

The tools will change, probably faster than any guide can keep up with. The sequence will not. Each new capability is worth meeting the same way: learn it yourself on real work, delegate what proves reliable with clear procedures, and decide explicitly what it should not do.

Leaders who adopt AI well are rarely the earliest or the most enthusiastic. They are the ones who formed their own judgement first — and then gave their organisation the clarity to use these tools with confidence and care.