The AI Wave Looks Small From Here. And That, Friends, Is The Problem.

Why the view of AI from afar is deceptive - and the data proves it.

A conversation worth repeating

A couple of weeks ago I had lunch with the CEO of a well-run, mid-sized 200-person business. When the conversation turned to AI, I said I was genuinely fearful that most businesses - maybe even entire industries - were still treating this moment as some kind of fad or novelty. Yes, from afar the waves being created by AI might appear small, but up close, and from where I'm standing, those waves are most definitely not small. In fact, right now, they're forming into a giant tsunami that will devastate everything and anything before it. I said to my friend that I sometimes feel like I'm shouting into the void just trying to get this point across.

"AI can already do most white-collar jobs. Not soon. Now."

I said the quiet part out loud. Sure, I said, there are still some office / professional jobs which still must be carried out by a human, but those few aside, I'm finding it hard nowadays to name many, if any, which couldn't already be performed by AI - using frontier models available today.

My friend did what any good friend does and pushed back. He named a bunch of roles - admin, sales, management - and said, go on then. Explain how AI replaces those jobs entirely.

The answer to that question is the point of this post, because I think some people are missing the point at which AI integrates into business, and the comment I hear the most; you mean a chat bot, right? Errm, not really, no.

The like-for-like fallacy

When people assess whether AI threatens a role, they instinctively imagine a like-for-like fight: first they think of an AI chatbot, then they imagine that against an experienced human being. And yes, a chatbot versus your very best salesperson, head to head, working a single account is 1-0 to the humans. The judgement, the relationships, the thirty years of instinct, empathy, the human connection are all real advantages. No arguments here - that scenario isn't so much a fight, it's more a bloodbath.

But that is not the fight that's coming. It isn't even the fight that's here today.

Take sales for instance, the role my CEO friend pressed hardest on. The correct picture is not a single AI bot versus a salesperson in a one-on-one fight. It's a squad of agentic AI systems built on the latest models - planning, researching, contacting, emailing, scheduling, following up and adjusting strategy, around the clock and working an entire market simultaneously. A human sales team works a pipeline of dozens of accounts. An agentic system works thousands - each contact researched, each message personalised, every follow-up on time, indefinitely. Learning, recording and contextualising every single contact and making that data readily available for the next sales cycle. Industry benchmarking of deployed sales agents makes the mechanism concrete: a lead arriving at 11pm on a Saturday now receives the same sub-fifteen-minute, contextually informed response as one that arrives mid-morning on a Tuesday. No human team can offer that, not at any salary. And guess what - your customers are going to get used to having that on tap - they're gonna prefer that - getting answers to important questions, and receiving a rapid response to a query when they need it, not just when you're available.

AI technology doesn't just level the playing field; it changes what the field is. Volume of contact, volume of completed work, consistency that never dips. The human doesn't lose the duel. The duel simply stops being how the hypothetical war is fought.

Run the same logic through admin, reporting, bookkeeping, scheduling, customer service, procurement, research, first-draft-everything. Role by role, the like-for-like comparison flatters the human - and the competitive reality doesn't care, because nobody serious is deploying AI like-for-like. And this is even before we examine what each version costs.

What the numbers actually say

None of this rests on opinion. The data, however, gathered from the past eighteen months, tells a consistent story - adoption is accelerating sharply, returns are real, and the effects on business have already started.

On adoption: British Chambers of Commerce research from March 2026 found 54% of UK firms actively using AI - up from 35% in 2025 and 25% in 2024. That is not a gentle curve; it is exponential growth in just under two years. Meanwhile 37% of surveyed organisations deployed agentic systems in the past twelve months, against just 8% in the previous two years combined. This tells us the quiet phase is ending.

On returns: companies report average returns of around 171% on agentic AI deployments, and roughly three-quarters of executives whose organisations introduced agentic AI saw a return within the first year. These are not laboratory numbers; they are why your fastest-moving competitor will not be slowing down.

On work itself: the World Economic Forum projects 92 million roles displaced and 170 million created globally by 2030 - a net gain, but a brutal reshuffle for anyone standing in the wrong place. The early signals are already measurable: US companies adopting AI have reduced junior hiring by around 13%, and employment among the most AI-exposed roles for 22-to-25-year-olds has fallen by double digits since capable AI became widely available. Anthropic's own chief executive has warned that AI could eliminate roughly half of entry-level white-collar positions within five years. It's this writer's humble opinion that this is likely an understatement. Nevertheless, and no matter what one thinks of any single forecast, the direction of every serious estimate is the same.

And on the gap: a third of UK SMEs still have no plans to adopt AI at all. Analysts put the value being left unclaimed by UK small and mid-sized businesses at an estimated £78 billion. The most commonly cited barrier is not budget - it's skills and knowledge, named by 60% of businesses. Which means the moat protecting the unprepared is not money. It's simply that nobody has shown them where to start. That is a fixable problem - but only if it's fixed before it matters.

So why does the wave look so small from afar?

Here is what makes this moment genuinely dangerous for the unprepared: almost everything described above is invisible from inside a normal business right now. Customers are still buying, clients are still sending emails and calling up on the phone. Traditional business ticks on just like normal. So where exactly is this fire (ahem, giant wave)? Well, your competitors are not announcing their plans or the transformative work they're doing right now in private. The early winners are silent, internal and unglamorous - a sales function here, a back office process there. A significant cost saved by removing a bunch of menial tasks, taken from those who didn't like doing the job in the first place. First a trickle and then a stream. You'll notice it when a rival wins a contract by delivering a significantly cheaper quote than you or when one of your competitors is suddenly everywhere and beating you more regularly than normal. Maybe your first experience of this is when a customer is apologising for not giving you the order only because someone else got back to them much faster than you did. At 9pm no less. On a Sunday.

Ok, so not a wave, let's go with structural shift instead.

The disadvantage compounds. A competitor who started in 2024 hasn't just bought better tools - they've spent two years learning what works in their business: the data hygiene, the staff habits, the judgement about which processes pay. It's organisational knowledge which cannot be purchased retroactively at any price. Every quarter of "wait and see" widens the gap that money alone can't close.

The businesses hurt worst by structural shifts are never the ones that saw them coming and bet wrong. They are the ones that decided it was hype and stopped looking. Tuning out the AI industry's noise felt like the sensible, grown-up response for a couple of years - and for a couple of years, it was. But that time is gone.

So what does a valid defence look like?

First of all, this post is not an argument for a knew jerk reaction; panic buys bad software. Panic reaches out to the current vendor and grabs the thing that has AI written on it. The defensible position is real, reachable, and - if started early enough - surprisingly undramatic. It has four parts.

1. Know your exposure. Map your operation. Map the opportunities for cost savings and increasing revenue. Map it the way a competitor armed with AI would: where do you charge for time a machine now does in minutes? Which revenue lines depend on customers lacking a faster alternative? What task takes several hours a week and has no direct impact on revenue. Most businesses have never asked. The answers are usually clarifying, occasionally uncomfortable, and always better known than unknown.

2. Adopt in parallel, start now. This is not a transformation programme, nor is it a tear down everything that exists today and hope that the AI can replace it. It's also not a platform purchase. It starts small and with single process, one pinch point, built and run alongside how you work today, proving itself before anything depends on it. The first process matters less than the muscle it builds - because that muscle takes months to develop, and it needs to exist before it's needed.

3. Own what you build, and stay nimble. The counterpart to adopting is not getting trapped. Long vendor contracts and locked-in platforms turn this year's defence into next year's liability, because the technology improves quarterly and your systems need to improve with it. Remember ChatGPT version 3, merrily hallucinating information and inventing "facts" back in 2022? You wouldn't dream of using that model today, so build it so the engine can be swapped out the moment a better one becomes available.

4. Bring your people with you. Staff are already using AI quietly - the ones using it well are an early advantage, not a risk. The roles that endure will be re-formed around judgement, relationships and oversight, with volume-work delegated to machines. Businesses that start reforming now will do it calmly, by choice. Those that wait will do it in a hurry, under pressure, and badly.

So what happened at lunch?

My friend was never a convert and still isn't now. But by the end of it he'd lost some of the bewilderment, but he did then ask me a pertinent question: "So what would you actually do then, in my business?" And that was the question I'd been hoping he'd ask from the very start.

Sources: British Chambers of Commerce AI adoption research (March 2026); World Economic Forum, Future of Jobs Report 2025; published enterprise agentic AI deployment and ROI benchmarks (2025–26); UK SME adoption gap analyses (2026).