The bottleneck is no longer AI, it's how we work

For the past three years we have heard the same prediction with different dates attached: next year, artificial intelligence will turn every industry upside down. No company will be safe, all software will be rewritten, and whoever fails to move will disappear.

And yet here we are. Companies still buy from the same suppliers, approve invoices through the same circuit, write proposals from the same templates and hold the same Monday meetings. All of this with a technology on the table that could, objectively, let them do almost any of it differently.

It is worth understanding why, because the explanation is not where most people go looking for it.

The prediction that hasn't come true (yet)

When the first genuinely capable models arrived, plenty of sensible people assumed disruption would be immediate. It wasn't. And the reason is not that the technology disappointed: it is that the economy has enormous inertia.

People keep doing what they were already doing. They keep buying from whoever was already selling to them. They keep wanting to use their tools the way they learned to use them. A three-year contract, a certified process, a team that finally works well with the current workflow — none of that evaporates because a better model exists.

The lesson for anyone running a company cuts both ways. The bad news is that having the best technology guarantees absolutely nothing. The good news is that you have more room than you are being sold.

Inertia is not the enemy

This is worth saying plainly, because the dominant narrative treats it as a defect to be corrected: that resistance to rapid change is, to a large extent, healthy. It is what stops an organisation from reorganising itself every time something new appears, and what allows big transitions to happen without breaking things along the way.

Inertia is not the problem. Mistaking it for a decision is. Many companies believe they are waiting — evaluating, letting the business case mature — when in fact they are simply doing what they have always done by default, with nobody having decided anything at all.

And that is an enormous difference between companies that, from the outside, look equally slow.

The uncomfortable case: ourselves

There is a very simple test that almost nobody passes. Think about how much the way you use a computer has changed over the last twenty years. Probably very little: you still copy and paste from one application to another, still scroll down the inbox deciding which email hurts least to open, still keep a to-do list that you drag from one week to the next.

The interesting part is that this happens even to people with access to the best tools in the world, including the people who build them. Ask them, and none will say they prefer working this way. Yet by revealed preference, they do.

That rules out the easy explanation. It is not a lack of tools, and it is not a lack of conviction. Something in our heads associates working with that particular sequence of gestures, and dismantling it is harder than installing anything.

It is also, in large part, a product failure. We are in the moment before the iPhone: every technical piece exists — the touchscreen, the connectivity, the browser — and nobody has yet assembled the set of ideas that genuinely changes how a person relates to the machine. Whoever solves that inside a company will capture most of the benefit, and it will not necessarily be whoever has the best model.

What the companies actually moving do differently

When you look at organisations that are genuinely months ahead — not months ahead in announcements — a fairly boring and very consistent pattern shows up.

In almost all of them there is someone with decision-making power using the technology with their own hands. Not a committee that receives a quarterly demo, and not a team preparing a deck to reassure the board. Someone at the top who tries the models, crashes into them, and finds out first-hand where they work and where they don't.

The reason is simple: judgement cannot be delegated. When your feel for the terrain arrives filtered through three layers of people who want to bring you good news, you end up making decisions about a polished version of reality. In a technology that shifts every few months, that distance is paid for in lost quarters.

The other shared trait is that they start with work that already exists. Not with the flashy project that will make headlines, but with the awkward process that eats hours of expensive people's time: contract review, tender responses, tier-one support, the monthly report nobody wants to write. Far less epic, and it works far better.

The real limit is no longer intelligence, it's context

Here, in our view, is the most important reframing for a company this year.

Models are already smart enough for the vast majority of office work. What limits them is not their ability to reason: it is how little they know about you. An excellent model with no access to your documents, your internal conversations, your customer history and your business rules is a brilliant professional on their first day, with no context and no memory.

And context is precisely the ground where no human can compete. Nobody reads tens of thousands of pages in seconds and uses them accurately when making a decision. No executive reads the entire support channel, every sales call of the quarter, or the hundred reports their organisation produced last year. An AI with orderly access to all of that replaces nobody's judgement: it gives the person deciding a foundation that until now was physically impossible to have.

That is why the companies gaining traction are not the ones testing the most models, but the ones that first put their information sources in order and connect them sensibly, with permissions and traceability. Intelligence can be bought. Context has to be built.

Less epic, more postmortems

One final point on how to introduce all of this without the project collapsing on first contact.

Aviation became the safest way to travel thanks to a deeply unglamorous culture: every incident is reported, investigated without hunting for culprits, and the lessons shared with the whole industry. It wasn't solved by thinking very hard before take-off; it was solved by flying and studying every failure honestly.

The same applies to AI inside a company. You learn by deploying early in a bounded area, with a clear owner, watching closely where it breaks and writing down what happened when it fails. Pilots designed for six months in a room so they can launch perfectly almost never launch — and when they do, the model that inspired them is already obsolete.

It also helps to turn down the volume on the apocalyptic narrative. Much of the industry has communicated terribly for two years: existential risk figures on one side, promises to eliminate half of all jobs on the other. No wonder people use AI daily and distrust it at the same time. Inside an organisation, what decides whether adoption works is not the messaging: it is whether each person feels the tool gives them more decision-making power, more judgement and less mechanical work. When they feel it, adoption spreads on its own. When they suspect the goal is something else, no amount of training will fix it.

Nobody knows which year it will be

It is tempting to put a date on it. Some will tell you this is the year every business is up for grabs. It will probably take a little longer, and it certainly won't happen to everyone at once.

But the direction is not in doubt, and the advantage will not be set by who has the best model — before long, everyone will have access to similar ones. It will be set by who changes how they work first: who connects their knowledge, who decides with the information in front of them, and who gives their team tools that expand what they are able to do.

That cannot be bought. It has to be started.