Learn AI Before It's Too Late: What the People With Horses Got Wrong
- Jul 16
- 7 min read
TL;DR
19.2% of Canadian businesses use AI. 40% say it isn't relevant to them. That second number is the whole story.
Every general purpose technology hit the same wall: the car, electricity, the spreadsheet. The people resisting were not stupid. They were early, and being early feels a lot like being wrong.
Factory electrification is the real lesson. Buying the technology changed nothing. Redesigning the work around it changed everything. That took forty years.
“Too late” doesn't mean you get replaced overnight. It means competing against someone holding two years of judgment you can't purchase.
Start with 30 minutes a day on work you already do. No course required.
In 1865, the British Parliament passed a law requiring every self-propelled vehicle on a public road to be preceded by a man on foot, waving a red flag. Speed limit: 4 mph in the country, 2 mph in town. That law stayed on the books for 31 years.
It's easy to laugh at now. Don't. The men with horses weren't idiots. Early cars were loud, unreliable, absurdly expensive, and genuinely dangerous. The red flag was a reasonable response to an unreasonable machine. The trouble is that they were right about the machine of 1865 and completely wrong about the machine of 1895. By the time the law was repealed in 1896, the question was no longer whether cars worked. It was who had spent thirty years learning to build them, drive them, fix them, and route roads around them.
The 40% aren't wrong about today. They're wrong about the clock.
Statistics Canada's latest numbers, from the second quarter of 2026: 19.2% of Canadian businesses reported using AI to produce goods or deliver services in the previous 12 months. That's triple the 6.1% recorded in 2024. Real growth, fast growth.
Now the other number from the same survey. Two in five businesses, 40%, said AI use is not relevant to their business. Not “too expensive.” Not “we're waiting.” Not relevant. And it clusters exactly where you'd expect: construction at 9.2% adoption, wholesale trade at 7.9%, agriculture at 4.5%.
Those owners are making a reasonable judgment about the tools as they exist this morning, in their shop, for their specific work. Same judgment the man with the horse made. He wasn't wrong about 1865. He was wrong about how fast 1865 turns into 1895. Most of the reasoning holding that 40% in place is built on myths that are quietly costing small businesses money, and the data doesn't support any of them.
The factory that bought an electric motor and changed nothing
Here's the analogy that actually matters, and almost nobody tells it.
Electric motors were commercially available in the 1880s. Factories at the time were built around a single massive steam engine driving a central shaft that ran the length of the building, with leather belts dropping down to every machine. Machines were positioned by proximity to power, not by workflow. If you needed a lot of torque, you sat near the engine. The whole building was a diagram of the driveshaft.
The first wave of factory owners adopted electricity enthusiastically. They ripped out the steam engine, bolted in one big electric motor, and connected it to the exact same driveshaft. Productivity barely moved. The reasonable conclusion, widely reached: electricity is overhyped.
The gains showed up roughly forty years later, when a new generation of managers put a small motor on each individual machine. Suddenly the floor plan could follow the work instead of the shaft. You could arrange machines in the order the product actually moved. You could put windows where the shafting used to be. You could add overhead cranes, because the ceiling was free. Productivity exploded.
The technology hadn't changed much. The people had. It took a generation to unlearn the steam engine.
This is precisely where most AI projects sit right now. A business buys ChatGPT licenses, bolts it onto the existing process, gets 4% faster at writing emails, and concludes the whole thing is hype. The motor is electric. The driveshaft is still there. It's the same pattern behind why most AI projects fail in the first 90 days: the tool arrives, the process never changes, and everyone blames the tool.
The gap isn't technical. It's a learning gap.
A 2025 KPMG Canada snapshot found only 24% of Canadian employees had received any AI education or training. RBC named the underlying problem well: an “imagination gap,” a widespread inability among leaders, especially at smaller firms, to picture where this fits in their actual work.
Then set that beside a finding from the Business Development Bank of Canada: 97% of SMEs that have adopted AI report tangible benefits.
Read those two together and the picture is unambiguous. The tools work for nearly everyone who uses them seriously, and nearly nobody has been taught how to use them seriously. That is not a technology problem. It's a training problem.
The good news: training problems are cheap to fix. The bad news: they take time, and time is the one input you cannot buy retroactively.
What “learning AI” actually means
It doesn't mean a certification. It doesn't mean a weekend course. It means reps.
Thirty minutes a day on real work. Not toy problems. The quote you're writing anyway. The email you've been dreading. The spreadsheet you reconcile every Friday. Toy problems teach you nothing because you don't care about the answer.
Push it until it breaks. The valuable knowledge is the boundary: where it's reliable, where it invents things, where it needs a human check. You only find that edge by walking into it.
Learn the skill, not the vendor. Tools move faster than your business does. The most powerful model ever launched vanished three days later, and anyone who had wired their operation to that one model spent a week rebuilding. The judgment transfers between tools. The subscription doesn't.
Keep a list of what it got wrong. That list is your actual expertise. It's the thing no consultant can sell you, because it's specific to your business, your vocabulary, your clients.
Stop restarting. Week two is where it gets useful. Most people quit in week one because the first attempts are bad, and they conclude the tool is bad. The tool isn't bad. Your prompt is bad. That's fixable in an afternoon.
Then automate. Automation is what you do once you know what good looks like. Automating a process you don't understand just produces mistakes faster and at scale. This is the single most common reason automation projects die, and it has nothing to do with the software.
Too late for what, exactly?
Let's be honest about the risk, because the doom framing is lazy and mostly wrong. Your business does not get vaporized overnight. Nobody shows up Monday and replaces your team with a chatbot.
The risk is compounding. The competitor who started in 2024 doesn't just own better tools. They own two years of knowing what to ask for. Two years of knowing which tasks to hand over and which to keep. A team that isn't afraid of the thing. Cleaner data, because they finally had a reason to clean it. A habit of questioning their own processes, which turns out to be the most valuable side effect of the whole exercise.
You can buy the identical software tomorrow morning. You cannot buy the two years.
That's what “too late” means. Not extinction. Just being permanently one step behind someone who started earlier and never stopped.
Start this week
Pick one task you do every week that you don't enjoy. Give it 30 minutes a day for two weeks. Write down what worked and what didn't. That's the entire assignment. If, after two weeks, you still think it's not relevant to your business, you'll have earned that opinion honestly, which is more than most of the 40% can say.
If you're earlier than that and wondering where AI fits before you've built anything at all, the practical path for starting with AI covers the ground before this one.
FAQ
Do I need to learn to code to use AI in my business?
No. The highest-leverage skill isn't programming, it's knowing which parts of your work are worth handing over and how to describe them precisely. That's a management skill, not a technical one. Coding helps later, when you move from using tools to building systems.
How long before I see a return?
On personal use, days. On real workflow automation, expect 90 days before it's stable and measurable, which is exactly the window where most projects quietly die. The businesses that see returns are the ones that mapped the process before automating it.
My industry barely uses AI. Doesn't that mean I can wait?
It means your competitors haven't started either. That's the opportunity, not the excuse. Construction sits at 9.2% adoption in Canada. Low adoption in your sector means the gap you build now is larger and lasts longer than it would in a saturated market.
Where to go from here
At Neurotek AI, we start every engagement with a discovery session for exactly this reason: before we automate anything, we map what the work actually is and where the driveshaft is still bolted to the floor. That's the whole philosophy, and it's why we work the way we do. Automate. Delegate. Eliminate. In that order, and never before you understand the process.
If you want to see what that's worth in your shop before talking to anybody, run your own numbers with the ROI calculator. If you'd rather just listen for a while, Simon breaks down one AI story a week, in plain language, on Cogeco Outaouais 104.7 FM.
The red flag law lasted 31 years. Nobody remembers the names of the men who carried the flags.
Comments