Yesterday I spent the day at the C-Vision International CIO & CISO Think Tank in London with the Chief Architect Network. The agenda was the agentic enterprise, and by lunchtime the language in the room was familiar. Operating model. Human in the loop. Human above the loop, around the loop. And a phrase I’ll be chewing on for a while: it isn’t artificial intelligence, it’s assisted intelligence.

All worthy conversations. And yet I woke up this morning with a lightbulb moment that hasn’t let go.

We are talking about moving work into AI that we have never properly understood in the first place.

To see why, it helps to look at how we got here. For decades we have deployed technology on its own terms. Finance, supply chain, customer management. The business defined the data. The data fed the platform. The processes were built and configured the way the technology needed them to be: hosted, secure, backed up.

The front office systems pushed clients, colleagues and citizens towards self-service. The exceptions landed with teams who understood the business rules well enough to work around them. We all know the sound of that model. A sigh, then “the system says no.” Or “can you raise a ticket?”

The technology did the transactions. The people did everything else.

Now listen closely to the AI rhetoric. The word that keeps surfacing, sometimes said out loud and sometimes just implied, is migrate. Take what humans were doing around the technologies of the past and move it into an AI stack. Extend technology beyond transactions into language and knowledge.

Which assumes we know what humans were doing.

I’m not convinced we do, because we have been here before. In the early 2000s I led the outsourcing of back-office processes from onshore teams to offshore ones. What became very clear, very quickly, was that the onshore teams were doing far more than the core systems required. They were also doing far more than most people understood.

There were shadow spreadsheets. There were contacts you called to get things done. There was a whole heap of knowledge about the country, the company, the products and the system workarounds. None of it was in the process maps. The migration exposed all of it. And because I was leading it, I was the one who had to explain why.

At the time, information governance teams were pushing for better management of corporate knowledge. They worried about personal C drives mushrooming with knowledge files nobody else could find. Every attempt to classify, codify and centralise it ran into the sheer scale of the problem. More often than not, it was quietly deprioritised.

That knowledge didn’t go away. It just stayed in people.

Years of restructuring programmes taught me something else. Knowledge-based roles are captured far less faithfully than processing roles.

We spend days reclassifying roles, debating delegated authorities, trying to pin down decision rights. The role profiles go out for consultation. Within weeks of the new structure going live, the gap between someone’s profile and what they actually do can be striking.

So suppose the plan is to load up a set of AI agents with letter-of-the-law role profiles and point them at the guardrails of our policies and standards. If we then expect the operating model to perform better, are we kidding ourselves?

We will get adherence to the standards. But will we lose what was keeping the business running in the background? Those are the things people did that often they themselves weren’t even conscious of.

There is a second assumption underneath the excitement, and it’s financial. Where exactly is the saving in an organisation that is already lean?

Much of the transformation work of the last few decades has been about building lean operating models. We have automated. We have outsourced and offshored to drive down labour costs. Now AI arrives to take its slice of the savings too.

The business case may start as a cost case. But taking cost out of an organisation that has already had cost taken out many times is a hard sell. AI will clear a backlog faster and will replace some roles. Add governance, oversight, security, compute and subscription costs, though, and trimming a team won’t pay it back. It needs scale.

Scale means growth. Organic growth is hard, and it takes more than AI. It takes product, price, place and promotion, done better than competitors who aren’t standing still either. Otherwise it means acquisition and consolidation.

That’s why I keep coming back to this. AI is a game changer for the business model, not just the operating model. Yet most of the conversations I heard were about the operating model.

Which brings me back to the loop everyone was talking about. Human in the loop is a worthy discussion. But listen to how it is framed. Too often the human is there to make the technology more effective, to catch its errors and keep it compliant. The human serves the tech.

Business starts and ends with people. A mindset where humans exist to optimise the loop, producing generic experiences for everyone involved, may matter far less than we think.

AI is more sophisticated than any technology we have deployed before. My fear is that we are bringing in the new tech without having become any more sophisticated in the way we deployed the old.

That distance, between the business we wrote down and the one people are actually running, is the human gap. AI doesn’t close it. Moved too fast, it widens it.

So before anyone signs off the agentic roadmap, there are harder questions to sit with.

• Do we really understand our business model, and how AI might change it?
• Do we understand what our operating model actually does, not just what the role profiles say it does?
• Are we ready to onboard AI into the way we work around here, and the way we intend to work around here?

None of that is a technology question. It isn’t the CIO’s or the CTO’s responsibility alone. Nor does it hand them the head of the table by default. That seat, as ever, has to be earned.

Like every technology before it, AI is an enabler. The question is what it’s enabling. Is it the business we wrote down, or the one people are actually running?

Whatever next?