AI Is Not a Strategy: How Leaders Create Real Value With AI

When the AI boom first began, I worked at a company where everyone suddenly wanted AI in the platform.

What would it do? Nobody was entirely sure. What problem would it solve? That part seemed almost secondary. It simply needed to be there so we could say that our product had AI.

This was happening everywhere. Companies were rushing to add an AI feature before they had worked out whether their users needed one. In many cases, the grand innovation ended up being some version of ChatGPT rebuilt with different branding. It looked current in a product demo, but there was not always much thought behind why it existed.

That is the part of AI adoption that still makes me uneasy. It is easy to become so distracted by what the technology can do that we forget to ask what it should do.

I am not an AI sceptic. I use it every day, and its place in my work has grown considerably since I first wrote about this topic. It helps me with code, research, writing and problem-solving. My team uses it too, but we did not introduce it with an ominous announcement about transformation or a new expectation that everyone should suddenly work twice as fast.

We use it for repetitive tasks that nobody particularly enjoys. We use it to put together quick proofs of concept and to assist with work that still belongs to us. Nobody on the team has expressed concern that they are being replaced or quietly measured against the speed of a machine. AI is simply another useful part of how we work.

One of the most valuable ways we use it is while writing features and tickets. Before work begins, we use AI to question us about the idea and push on the gaps in our thinking. We have used a grilling skill from Matt Pocock’s open-source collection for this. Instead of accepting the first version of a requirement, it keeps asking questions until the missing decisions, edge cases and assumptions become visible.

It saves us a considerable amount of time, but speed is not the part I value most. The requirements are more complete because we have been forced to think about things we might otherwise have missed.

That is very different from asking AI to create a ticket and blindly copying whatever it produces. The value comes from the conversation. We guide it, question it and reason with it until both sides are working with the same understanding. The AI may help us see the gaps, but we are still responsible for deciding what belongs in them.

That responsibility matters because AI can sound exceptionally convincing while being completely wrong.

I have seen it happen many times. It can produce code that looks sensible until you understand what it is actually doing. It can offer a technical answer with absolute confidence while quietly inventing part of the explanation. It can turn a rough thought into polished writing that no longer sounds like the person who had the thought.

The polish makes it tempting to trust. Sometimes the answer looks so finished that checking it feels almost unnecessary. That is exactly when understanding matters most.

You cannot hand your judgement to AI simply because it presented its answer neatly. You need enough knowledge of the work to challenge what it gives you. You have to ask why it made a decision, point out where its reasoning does not fit and keep talking until you understand each other.

Sometimes it feels less like operating a tool and more like working alongside an extremely knowledgeable colleague who occasionally makes things up with remarkable confidence.

That is one way my view of AI has changed. I still believe it is a tool, but “tool” no longer quite captures how I experience it. It has become more like a companion because I use it across so many different parts of my life.

Recently, I started a sourdough starter. Each day, I give my AI an update about what it looks like and what it has been doing. It explains that the strange thing I am seeing is still normal, tells me what stage the starter may be in and helps me work out how to feed it next.

In that moment, it feels a little like having a professional baker teaching me as I go.

Before AI, I would have found a website explaining how someone else created their starter and tried to follow it exactly. The instructions might have been perfectly correct for their flour, their kitchen and their climate. The moment mine behaved differently, I would have searched for more answers, found several that contradicted one another and possibly given up in a small cloud of flour and disappointment.

AI lets me continue the conversation from where I am. I can explain what happened, ask the next question and adjust as I learn. That ability to respond to context is what makes it more useful than a static set of instructions.

But a companion still needs to be understood, challenged and given current information.

If a company feeds its product documentation into an AI assistant so that people can ask questions, the work is not finished when the assistant launches. The product will continue changing. If nobody maintains the documentation, the AI will confidently answer questions using an increasingly outdated version of reality.

The same applies to the model, the surrounding systems and the way people are using them. An AI implementation needs an owner. Someone must keep checking whether the information remains relevant, whether the answers are still accurate and whether the tool continues to solve the problem it was introduced to solve.

Without that ownership, AI does not remove work. It hides the work until the consequences become difficult to ignore.

This is why leaders need to begin somewhere other than “We should add AI.”

They need to understand the problem first. They need to be honest with their teams about what the technology is meant to do and what it is not meant to do. They need to create room for people to experiment without turning every efficiency gain into a new performance expectation. They also need to make it clear that somebody remains accountable for the output.

The most useful applications of AI in my work have not been the flashy ones. They have been the moments when it took repetitive work off someone’s plate, helped us find a missing requirement, allowed us to test an idea quickly or gave me enough context to keep a sourdough starter alive for one more day.

None of those things required pretending AI would change everything.

AI can be deeply helpful. It can make expertise easier to reach, create space for better thinking and help us work through questions that might otherwise stop us. But its usefulness still depends on the judgement, context and care surrounding it.

Before adding AI to a product or a team, leaders need to ask whether it solves a real problem, whether their people understand why it is being introduced and who will remain responsible when its confident answer turns out to be wrong.

If those questions do not have clear answers, the organisation probably does not have an AI strategy yet. It just has AI.