The Real Lesson in AI Implementation

The implementation literature suggests we've been solving the wrong problem.

We’re still putting the cart before the horse. We start with which tech, which model to use, which platform to buy and which vendor has the best capabilities. When projects struggle we assume the technology isn’t mature enough, it’s a problem of adoption or we simply haven’t implemented it well enough, but the implementation literature tells a different story if we read between the lines.

Read enough solid AI implementation guides and you begin to notice an odd pattern. They spend surprisingly little time discussing models compared with workflows, governance, ownership, operating models, cross-functional teams, decision-making and organisational learning. Even the sections on agents become discussions about coordinating work rather than improving model capability.

Rightly so.

But...these implementation guides consistently stop one level short of the natural abstraction.

None of them set out to explain how intelligent systems change organisations. Their purpose is much more practical, which is understandable. They’re trying to help leaders implement AI successfully, yet they keep arriving at remarkably similar conclusions because the hard problems turn out not to be selecting a model or writing a prompt, they’re deciding who owns decisions, how work should flow, where authority sits, what controls are needed and how people and intelligent systems work together.

What’s striking isn’t that one implementation guide reaches these conclusions, it’s that they all seem to converge on broadly the same organisational advice despite coming from different authors, vendors and institutions. That’s exactly what you’d expect if the primary constraint wasn’t the technology but the organisation trying to absorb it.

I don’t think that means organisations should stop implementing AI while they redesign themselves. In fact, I think the opposite. The operating model isn’t discovered in workshops before implementation begins. We cannot redesign an organisation around assumptions. We have to redesign it around evidence, and implementation is how that evidence is created. It’s discovered through implementation itself. Every deployment should leave the organisation with two assets, a working capability and a better understanding of how intelligent systems change the work around them. Implementation isn’t separate from organisational redesign, it’s the mechanism through which the redesign is discovered.

At some point we have to acknowledge what’s sitting in front of us. You cannot absorb AI in the way these implementation guides consistently recommend without changing the organisation and its operating model. The implementation literature isn’t just teaching organisations how to deploy AI. It’s documenting organisational adaptation without quite recognising that’s what it’s doing.

Once we accept that, implementation stops being a technology project supported by organisational change. It becomes the mechanism through which organisations discover how they need to change. That’s why I increasingly think the more interesting question is no longer “How do we implement AI?” but “How do intelligent systems change the design of organisations?”