The AI Revolution Is Real. So Why Is It So Patchy?

William Gibson’s observation that “the future is already here; it’s just not evenly distributed” has never felt more apt. AI is, without question, one of the most significant technological shifts of my lifetime. I’ve been in software for nearly four decades and I’ve never seen anything move this fast. The capability improvements over the past three years have been, at times, startling.

And yet. If I look at the organisations around me (and I have a fairly direct view of this, running a software company that works with SMEs across the Midlands) the picture is genuinely odd. Pockets of extraordinary individual productivity. And then, right next to them, people carrying on largely as before. The revolution is happening in the same building on the same floor, and most of the floor isn’t in it.

Revolutions Are Always Patchy — and That’s Normal

There’s a well-documented story about the introduction of electricity in the United States. It took the best part of twenty years for electricity to spread across the country after it became commercially available. And even once factories had access to it, the productivity gains were initially modest and uneven.

The reason is telling. Factory owners who replaced their steam engine with an electric motor got a boost (they had a better engine) but their factory was still laid out for the steam age, with a single central power source feeding everything through belts and shafts. It was only when the next generation took over, reorganised the factory floor entirely around the new technology, and distributed power directly to individual machines that the real gains materialised.

The old guard couldn’t change fast enough. And they probably didn’t fully understand why they should. The electric motor was better than steam. Why would you throw out a perfectly good factory to take advantage of that?

The factory owners who replaced steam engines with electric motors got a boost. The real gains came when someone reorganised the whole factory around the new technology, and that usually took a generational change in leadership.

We are three years into a revolution that will probably take decades to fully play out. Patchiness right now isn’t a sign that something is wrong. It’s exactly what we should expect.

Prompting Puts the Work Back on Users

One thing that makes this revolution particularly uneven is that the current interface model (prompting) places significant demands on users. I’ve written about this in more detail elsewhere, but the short version is this: a blank prompt requires you to know what you want, know how to ask for it, and be willing to go back and forth iteratively until you get there. That’s a different set of skills from clicking through a structured interface, and it doesn’t come naturally to most people.

What this means in practice is that the people getting most value from AI right now are disproportionately those who are intrinsically motivated to invest personal time in learning to use it. That’s a small subset of the workforce. And a small subset of the workforce getting dramatically more productive is not the same thing as an organisation getting dramatically more productive.

The People Benefiting Most Are Exactly Who You’d Expect

Walk into most organisations and the AI story is actually quite consistent. There are one or two enthusiasts (almost always technical, often younger) who have invested enormous amounts of personal time into getting good at this. They’ve read the articles, tried the different models, worked out how to prompt effectively, and built up a set of techniques that work for them. Their personal productivity has gone up significantly. They find it hard to imagine working without AI now.

And then there’s the rest of the organisation. Not necessarily opposed to AI (often genuinely curious) but without the time, motivation, or frankly the inclination to invest that much personal effort. They’ve tried it once or twice, got mediocre results, and quietly went back to doing things the way they always did. That’s not laziness. That’s a rational response to an interface that rewards sustained investment of personal effort.

The enthusiast’s productivity gain is real. But it tends to stay personal. The knowledge doesn’t transfer easily because it’s largely tacit: accumulated intuition about how to frame prompts, which model to use for what, how to iterate. You can’t hand that to a colleague. And even if you could, you’d still be building on individual skill rather than shared organisational capability.

Your enthusiasts are achieving extraordinary personal productivity gains. That is genuinely not the same thing as your organisation getting more productive.

Even When Something Gets Built, It Rarely Gets Deployed

There’s a second problem that I see constantly, and it’s one that gets less attention than the skills gap. Even when people get good at AI and build something genuinely useful (a tool, a workflow, a prompt that works reliably) there’s no infrastructure to make it available to the rest of the organisation.

I’ve seen this dozens of times. Someone builds a brilliant prompt sequence for handling customer queries, or a process that uses AI to summarise internal reports. It lives in their personal account. Maybe they share it with one or two colleagues directly. It never gets to the rest of the team, because there’s no shared platform to deploy it on, no way to manage access, no mechanism to update it when the original creator leaves. The work just disappears into a personal subscription that gets cancelled when the budget comes under scrutiny.

This is the deployment problem, and it’s inseparable from the data problem. Most genuinely valuable business use cases require access to company data: your invoicing system, your CRM, your document libraries. Generic AI tools on individual subscriptions can’t reach that data. The enthusiast’s personal productivity gains are mostly on tasks that don’t require it: writing, research, summarising publicly available information. The hard stuff (the stuff that would actually move the organisation) is out of reach.

Smaller Organisations Are Doubly Disadvantaged

If this were equally hard for everyone, it might be acceptable. It isn’t. Large organisations have resources that smaller ones simply don’t.

Enterprise AI strategies at large companies look quite different from what I’ve described. They have dedicated AI teams, internal centres of excellence, people whose job is specifically to evaluate models, build internal tooling, and connect AI to corporate systems. They can negotiate enterprise contracts that include data residency guarantees and proper governance frameworks. They can afford to build the deployment infrastructure that individual subscriptions don’t provide. They’re running pilots with multiple thousand employees, iterating fast, and accumulating institutional knowledge.

Meanwhile, small and medium-sized organisations have enthusiastic individuals, a collection of personal subscriptions, and an IT team that is already stretched thin. The gap between what large organisations are building and what smaller ones have access to is not closing. If anything, it’s widening. The AI revolution has the potential to be the most significant accelerant to large-company advantage over smaller competitors that we’ve seen in a generation.

The AI revolution has the potential to be the most significant accelerant to large-company advantage over smaller competitors that we’ve seen in a generation.

What Actually Needs to Change

None of this is an argument against AI. The electricity comparison cuts the other way too: eventually, the technology became the foundation everything was built on. The factory owners who dismissed it as overhyped or too difficult weren’t vindicated. They were overtaken. The question isn’t whether to engage with AI; it’s whether you engage in a way that creates shared organisational capability, or just accumulates personal subscriptions.

The things that would actually close the gap are pretty clear. A shared platform that makes AI tools available to the whole organisation, not just those willing to learn to prompt. Proper integration with company data, so the AI can actually help with company work rather than just generic tasks. Infrastructure to deploy what gets built, so that when someone creates something useful it can reach the rest of the team. And the ability to track what AI is costing and where, because “AI” is not a fixed cost and it needs to be governed like any other significant operational spend.

None of these are technically exotic. They’re just not what you get from a subscription to a chatbot. The electricity didn’t reorganise the factory by itself. Someone had to decide to do it, and then do the work. That’s where we are.