Today I had one of those conversations that stayed in my head afterwards. We were talking about rolling out AI in an organisation. Not the shiny demo version, not showing someone a new agent and getting the inevitable “wow”, but actually making AI work across an organisation. And at some point I realised how many different people you need around the table to do this properly. Architects, security, data, developers and engineers, Internal IT, legal and privacy, HR, Learning & Development, adoption and change, communication, finance, business owners and leadership. And yes, don’t forget the project managers. Someone still has to keep this whole circus moving in roughly the same direction. 😉

And maybe that explains why AI transformation can be incredibly difficult.

The technology is only one part of it. Every one of those disciplines looks at AI through a different lens. Architecture needs to understand how everything fits together. Security needs to understand the risks. Data needs to know whether there is anything reliable underneath that brilliant AI idea. Internal IT eventually has to make all those ideas work inside the reality of identities, access, devices, applications and support. Legal and privacy have their questions. HR looks at roles and skills. Learning & Development looks at what people need to learn. Finance wants to understand where the value is. The business wants to move faster. Project managers are trying to turn twenty moving pieces into something resembling a plan. And leadership has to make decisions while the technology underneath those decisions keeps changing.

None of those perspectives is wrong. In fact, you need all of them. And that is exactly what makes transformation hard.

You need speed and governance at the same time. Experimentation and security. Innovation and standardisation. Technology and behaviour change. Short-term results and a longer-term direction. You need to give people room to explore while also deciding where the boundaries are. You need to build new things while questioning whether some of the old processes should exist at all.

And somehow all of those conversations need to come together.

That is why I increasingly struggle when people talk about an “AI implementation” as if we are rolling out another piece of software. We aren’t. We are changing work. And changing work touches almost everything.

But there is another part of this that I think we talk about far too little: what all of this is doing to the confidence of the people using AI.

AI has become incredibly good at sounding like it knows what it is talking about. You ask a question about strategy and get an answer that sounds like someone who has been advising boards for twenty years. Ask a technical question and suddenly it feels like you have a senior architect sitting next to you. Ask something about HR and three seconds later there is a perfectly structured people strategy. Five pillars included, obviously. 😉

Sometimes I read those answers and think: wow. And I work with this technology every day. So imagine what that can feel like for someone who doesn’t. The answer appears immediately. It is beautifully written, confident and structured. There is no hesitation, no awkward silence and usually no visible insecurity. It just answers.

And I think we underestimate what that can do to people.

We naturally associate confidence with knowledge. If someone sits opposite you in a meeting and confidently explains something using exactly the right terminology, you assume they probably know what they’re talking about. AI can create exactly that feeling. Except there is one rather important difference: it can sound completely confident and still be wrong.

Imagine you have worked in your field for fifteen years. You have built experience, made mistakes and learned what works and what doesn’t. You know the exceptions. You recognise things that aren’t written down anywhere because you’ve simply seen them before. Then you ask AI a question about your own field and within seconds it gives you an answer that might even be better structured than the one you would have written yourself.

What does that do to your confidence? Maybe AI knows this better than I do. Maybe everyone can suddenly do what I do. Should I challenge this answer? What if I’m wrong? What is my expertise still worth if somebody without my experience can produce something that looks this good in thirty seconds?

I don’t think those are strange questions at all. I think a lot more people are going to have them. And maybe this is where we need to change the AI adoption conversation. We spend a lot of time teaching people how to use AI, how to prompt, how to give context, how to create better output and how to use an agent. All useful. But perhaps the next phase isn’t mainly about getting better at asking AI questions. Maybe it is about getting much better at judging the answers.

Because producing something and understanding whether it is good are two completely different things.

A junior employee can now produce a strategy document that looks surprisingly similar to something written by a senior executive. Someone who has never created a business case can produce one in minutes. Someone without a communications background can create a polished communication plan. Someone who isn’t an architect can suddenly have a pretty sophisticated conversation about architecture.

I actually love that. It lowers barriers, helps people learn faster and allows people to contribute outside the traditional boundaries of their role. But access to expertise and actually having expertise are not the same thing. And that distinction becomes harder to see when everything looks equally polished.

Maybe that means the value of expertise is moving. Knowing information by heart becomes less important when everyone has access to a powerful AI assistant. Writing the first version becomes less valuable when everyone can generate one in seconds. But knowing whether something makes sense, knowing which question hasn’t been asked, recognising the assumption that doesn’t fit reality, understanding the context of your organisation, your people or your customer and knowing when something that sounds brilliant will never work in practice? That becomes more valuable.

Judgement becomes more valuable. And so does the confidence to use it.

I think we need to teach people that too. Not just “here is how you work with AI”, but also: you are allowed to disagree with it. You should question it. Check things. Bring in your own experience. Ask another expert. Understand the context. And sometimes simply say: “This sounds fantastic, but I don’t think it’s right.”

That is AI literacy too.

And suddenly I am back at that ridiculously long list of people we apparently need around the table. Maybe that is exactly the point. The more AI can do, the more perspectives we need to decide what it should do. The more convincing AI becomes, the more important human judgement becomes. And the more work we give to machines, the more deliberate we need to become about the responsibilities we keep with humans.

So no, AI transformation isn’t an IT project. But Internal IT is essential. It isn’t an HR project, but HR is essential. It isn’t a security project, a data project, a change project or a business project either. It is all of them.

And maybe that is exactly why AI transformation is so incredibly difficult. Not because the technology doesn’t work. But because making AI work means bringing together technology, processes, data, risk, value, leadership and people. All moving at different speeds, with different responsibilities and sometimes completely different ideas of what success looks like.

The technology can move incredibly fast. Organisations don’t. And people shouldn’t always have to.

Maybe successful AI transformation isn’t about getting everyone to move at the speed of AI. Maybe it is about creating an organisation where all those different disciplines can move together, make better decisions and keep enough human judgement in the system to know when to speed up, when to slow down and when to question the answer in front of them.

Because while AI is getting better at giving us answers, our biggest challenge might be learning how to decide together which answers are actually right.