The demo is going great. The vendor is beaming — he shoots a wink to his sales colleague across the room. "This one's in the bag." An executive watches an AI tool summarise her inbox in seconds. She's nodding along, licking her lips, imagining all the time her team will save. The CFO is calculating ROI on a napkin, and his bonus in his head.
And absolutely nothing is going to change.
This happens thousands of times a day. Welcome to the AI theatre — a place where artificial intelligence initiatives take the stage but the plot never actually advances. You've seen it. Hell, you're probably living it. The endless proofs-of-concept. The "quick wins" that never compound. The innovation workshops where everyone's doing AI but nobody's changing because of it.
McKinsey's recent 2025 State of AI report puts it starkly: 88% of organisations now use AI in some form. Adoption is pretty much solved. But only about 30% have scaled it beyond isolated experiments. That's a 58-point canyon between "we're doing AI" and "AI is transforming how we work."
The Faster Horse Fallacy sits right in the middle.
You know the legendary-even-if-unsubstantiated Henry Ford quote: "If I had asked people what they wanted, they would have said faster horses." Well, most companies are doing exactly that with AI. They're applying 21st-century technology to 20th-century workflows. Making the old way slightly faster instead of asking whether the old way should exist at all.
Paving the cow paths
We have a history of doing this, us humans. Picture a city built on cattle trails. Cows wandered wherever they wanted. Their paths became roads. Then someone paved them. Now you've got asphalt serving a logic that was devised by bovines[^1] — winding, inefficient, making zero sense for cars.
Urban planners call this "paving the cow paths." Companies do it with AI constantly.
The AI equivalent: you've got a twelve-step procurement process designed for a world where humans needed to verify, cross-check, and approve at every stage. Along comes AI, and you use it to speed up Step 4. Maybe automate Step 7.
Congratulations. You now have a bad process that runs 10% faster. You've paved your cow path with machine learning and called it transformation.
The three horizons
Not all AI value is created equal. Think of it as three horizons, each harder to reach and more rewarding.
The first horizon: efficiency
This is where 90% of AI projects currently live (and die). Cut costs, save time. Write marketing emails faster. Turn a doc into a slide deck. Summarise a Zoom call. You know the drill.
The trap here is that efficiency has brutal diminishing returns. The first 20% improvement comes easy. The next 10% is harder. The next 5%, harder still. And you're doing the same work, just marginally quicker. You haven't created new value — you've commoditised existing work.
Every competitor can buy the same efficiency tools you can. Racing to do the same thing 15% faster isn't a strategy. It's a treadmill. And mate, nobody wins on a treadmill. You just get tired.
The second horizon: amplification
Now we're getting somewhere.
Forget about doing the same tasks faster — this horizon is about doing tasks you couldn't do before.
A customer support agent with an AI copilot doesn't just answer tickets more quickly. She handles complex queries that would have required escalation. She spots patterns across thousands of conversations that no human alone could track, then proactively generates relevant documentation with a single command to her AI helper.
She becomes genuinely more capable, not just more efficient. See the difference?
The difference matters. Efficiency replaces human effort — this is what so many are scared of. Amplification multiplies it — this is where we should focus. It's certainly where my focus is.
The third horizon: transformation
Here's where it gets interesting — and where almost nobody I've seen has got to. Yet. (Kinda scary, kinda exciting, kinda the whole point.)
Transformation isn't about writing marketing emails faster (Horizon 1). It isn't even about writing better, more personalised emails (Horizon 2). It's about questioning whether "email campaigns" are the right model at all.
What if you replaced campaigns with autonomous agents creating hyper-personalised customer journeys in real-time? Not batch-and-blast, but continuous engagement that responds to each customer's behaviour as it happens. The entire concept of a "campaign" dissolves. You're not optimising the old game. You're playing a different one.
But most companies are stuck optimising for faster horses. The automobile has already been invented, people.
Breaking through the horizon
So how do you bring the curtain down on the AI theatre? Two shifts — one in how you design work, one in how you think about it.
The design shift: should it exist?
Stop asking: "How can AI help us do this task?"
Start asking: "If we had infinite intelligence at marginal cost, would this task even exist?"
Yeah, it's a weird thought experiment. But it breaks you out of incremental thinking.
Take expense reports. The standard approach is to use AI to review them faster, flag anomalies automatically, and maybe pre-fill some fields. Faster horse.
But why do expense reports exist? Because we didn't trust employees with company money and needed verification after the fact. They're a control mechanism designed for a world without real-time oversight.
What if AI agents handled pre-approved spending limits, auto-booked travel within policy, and flagged anomalies before money was spent? The expense report doesn't get faster. It stops existing. The need disappears.
That's the difference between optimisation and transformation. One makes the old thing better. The other makes it irrelevant.
Transformational thinking is a muscle just starting to be flexed. McKinsey reports that 62% of companies now experiment with AI agents — not chatbots or copilots, but autonomous agentic systems that chain together complex tasks without constant human prompting.
The old model: Human starts task → Uses AI tool → Human reviews output → Passes back to AI tool → Human reviews output → Repeat.
The agent model: Human defines objective → Agent swarm executes → Humans manage by exception.
It's the difference between being a player on the field and being the coach. Your job becomes setting goals, defining constraints, and handling the edge cases that require judgment.
The mindset shift: unlearning what you know
Here's the harder part.
Every organisation runs on some level of status-quo power — beliefs that feel like immutable truths but are actually just habits from a different era.
"Customers need to talk to a human to feel heard." (Do they? Or do they need their problem solved quickly and competently?)
"Quality requires human oversight." (At every step? Or just the steps where human judgment genuinely adds value?)
"We need weekly status reports to stay aligned." (Do you? Or do you need visibility into progress — which real-time dashboards already provide?)
These beliefs made sense once. They might still make sense in some contexts. But they're not laws of physics. They're organisational choices, calcified into "it's just how we do things."
Here's the problem: organisations have immune systems. They reject changes that threaten existing structures, even when those structures have outlived their purpose. Someone suggests reimagining a process from scratch, and suddenly everyone discovers reasons why every step is essential — usually reasons that amount to "because that's how we've always done it." (I've sat in those meetings. They're exhausting.)
Transformation requires identifying which constraints are real — regulatory requirements, physical limitations, genuine customer needs — and which are legacy artifacts. Then you need the courage to challenge the latter.
That's much harder than buying new software. It's changing behaviour.
And listen, if you're just starting your AI journey I'm not saying you should go for horizon three or nothing. Often the cow path is the right route to start on. If your invoice processing works well but takes four days instead of four hours, simply automating it might be exactly right. Start there. Learn. Build confidence.
The danger isn't in consciously choosing efficiency — it's in mistaking efficiency for transformation and wondering why your AI investments never compound into a real advantage.
The courage to reimagine
You know what's safe? Running a pilot. You know what's risky? Actually transforming. Nobody gets fired for the former. But they absolutely can for failing at the latter.
That's the uncomfortable part: transformation is risky. Adoption is easy. You buy AI software. You run pilots. You demonstrate activity. Everyone feels innovative. (The inevitable self-congratulatory LinkedIn posts practically write themselves.)
Scaling to transform is hard. It means dismantling processes that work. It means telling teams that the way they've done things for years might be obsolete. It means betting on new approaches before you have proof they'll work. It means awkward conversations with people who built their careers on the old way.
The winners over the next 12 months won't be the companies with the best AI models. The models are already commoditised — the same foundation models are available to everyone. The winners will be the companies with the courage to redesign their workflows from the ground up. The ones who reach the third horizon.
And that's how I spend many of my days — helping companies build jets instead of strapping rockets to stagecoaches. I work with leadership teams to move beyond the AI theatre — past the pilots that go nowhere, past the efficiency plays that commoditise your work, toward the transformation that actually changes how you operate. It starts with one uncomfortable question: "If you were building this company today, with everything AI makes possible, would it look anything like what you've got?"
Most of the time, the honest answer is no. And that's where the real work begins.
[^1]: It's incidental, but that's the first time I've ever used the word "bovine" in an article. Achievement unlocked.