Why AI adoption is inevitable, and why starting now beats starting later

In December 1975, a Kodak engineer took a photo of a woman named Joy Marshall. You couldn't see her face.

Joy was a lab technician. The man behind the camera was Steven Sasson, a young engineer, still new to the company. He had been given a job: see what could be done with a new kind of electronic sensor.¹

So he built a camera. The lens came from a movie camera. It ran on 16 AA batteries and saved its pictures to a cassette tape. It weighed about 3.6 kilograms and was the size of a toaster.¹

The photo took 23 seconds to save. It was black and white, and only 100 pixels wide. The camera could see very dark and very light. Everything in between came out as static.¹

That's why Joy had no face.

That toaster is now known as the first digital camera.¹

Kodak's executives watched it work, more than once. They saw no market for it. Sasson wasn't allowed to talk about it in public, or show it to anyone outside the company. He got a patent in 1978. The camera never went into production.¹

You know this story. Everyone tells it. But most people skip the part that matters.

On 19 January 2012, Kodak filed for bankruptcy protection in New York.² Its chief executive said the company had already cut 47,000 jobs since 2003.³ When it came out of bankruptcy in 2013, it had given up cameras, film sales and consumer photo developing.⁴

Now the part people skip. That same January, Kodak filed a presentation with US regulators. One slide listed its innovations.

Among them: "the invention of the digital still camera".²

Kodak had the idea early. It looked slow and silly, so it went on a shelf. And the business was never rebuilt around it in time.

We want to be fair to Kodak. One decision doesn't sink a company that big, and many smart people worked hard to save it. But the pattern matters, because we're living through it again.

A new technology arrives. At first it looks small, clumsy and a little silly. Sensible people wait. Then it gets cheap and good, faster than most people expected. And the ones who started early have a head start that is very hard to catch.

So this month we're asking one honest question. If AI will be cheaper and better next year, why not wait?

The forty-year wait

Kodak was one company. But a whole country once made the same wait, and it lasted forty years. It starts with a light bulb.

Electric light bulbs were on sale by 1879. By 1881, there were power stations in New York and London.⁵ Electricity was clearly a big idea. So you'd expect factories to get much more productive, quickly.

They didn't.

In 1990, the economist Paul David wrote a famous paper called "The Dynamo and the Computer". A dynamo is an early electric generator. He found that in 1899, electric motors gave less than 5% of the power that drove American factories. It took about two more decades for that share to pass one half.⁶

The big jump in productivity came later still, in the 1920s. David found that roughly half of that jump in manufacturing could be linked to the spread of electric motors.⁶ That's about four decades after electricity first went on sale.⁵

Why so slow? Because factories climbed three steps, one at a time.

First, they swapped the tool. They took out the steam engine and put an electric motor in its place.⁵ Everything else stayed the same: the same building, the same long shafts and belts running across the ceiling. Little changed.

Next, they redesigned the task. Each machine got its own small motor, and the heavy shafts disappeared.⁶

Last, they rethought the whole business. Factories could be built on one floor, with the work laid out in a smooth line.⁶ Workers needed new training, because they were now making more decisions themselves.⁵

The technology was ready long before the work was ready for it.

AI looks set to climb the same three steps. Swapping the tool is asking a chatbot to write the email we would have written anyway. Redesigning the task is letting AI read and code the invoices, while a person reviews only the odd ones. Rethinking the business is a finance team that closes the month in days and spends its time on advice.

Many businesses today are on the first step. That's fine. It's where everyone starts. But the big rewards sit on steps two and three, and those take time to reach.

Economists call this the J-curve.⁷ When a big new technology arrives, businesses have to invest in things that are hard to measure, like new skills and new ways of working. For a while, the numbers look weaker than they really are. Then the learning pays off, and they rise quickly.

So it pays to get the slow part done early.

This time, it's much faster

Electricity gave factories forty years. AI isn't giving us that long.

In late 2024, researchers asked American adults aged 18 to 64 a simple question. Have you used generative AI, the kind that writes and answers, like ChatGPT? About two years after ChatGPT launched, 39.6% said yes. By the same study's count, the internet was at about 20% two years after it became widely available.⁸

Canadian businesses are moving too. In the third quarter of each of the last three years, Statistics Canada asked them one question. Do you plan to use AI in the next 12 months? Two years ago, seven in ten said no. Today it's about half.

Over the same two years, the group that does plan to use AI more than doubled. It went from 10.6% to 25.2%.⁹

Among the businesses that still have no plans, 79.1% say AI is "not relevant" to what they sell.⁹ If you run a bakery or a plumbing company, that's true. AI doesn't bake the bread or fix the pipe. But every business also has quotes, invoices, emails, scheduling, bookkeeping and marketing. That's where many owners lose their evenings, and that's where AI helps first.

And it keeps getting cheaper

There's one more reason AI will spread everywhere. The price keeps falling.

Stanford's AI Index tracks this carefully. It looked at the cost of using AI as capable as GPT-3.5, the model behind the first ChatGPT. AI is priced by the token, which is roughly a short word or part of a word. In November 2022, a million tokens cost $20.00. By October 2024, the same level of AI cost $0.07.¹⁰

That's more than 280 times cheaper, in under two years. Sasson's toaster became the camera in every pocket. AI is making that trip in a fraction of the time.

When something useful gets that cheap, it doesn't stay optional for long. We saw it with electricity.

Why not wait, then?

So, back to our question. If AI keeps getting cheaper and better, why not wait?

Because the software is the cheap part. The expensive part is learning to use it well. Which tasks do we give it? How do we check its work? How does the team get comfortable with it? That learning doesn't get cheaper if we wait. It only starts when we start.

Two surveys show what waiting costs. One looked at big companies, one at small.

Boston Consulting Group surveyed 1,250 senior executives in 2025.¹¹

Read that chart carefully. Most companies have little to show yet. To us, that looks like step one: the tool is in, and the work hasn't changed.

The small group at the top reported 1.7 times the revenue growth of the group at the bottom. BCG says the leaders put their gains back into people and technology, so the gap keeps growing.¹¹ This is a survey of how executives see their own companies, so we'd treat it with some care. But it matches the electricity story. The gains go to the people who redesign the work, and redesign takes time.

The Business Development Bank of Canada looked at smaller firms. In early 2026, it surveyed 1,500 Canadian business owners and decision makers. Only 30% of small and medium businesses were using generative AI. BDC says those that did were 24% more productive.¹² We can't be sure AI caused all of that difference. More productive businesses may simply be quicker to try new things. But it's a gap worth noticing.

Picture two small accounting firms on the same street. These numbers are made up, for illustration only.

Each spends 10 hours a week on admin. The first starts with AI today. For a few months it saves almost nothing, because the team is learning. By month six, it saves 2 hours a week. By the end of the year, after redesigning a few tasks, it saves 5.

Add that up, and the first firm has about 100 hours back in year one. The second firm never gets those.

Then the second firm starts. The tools are better by now. But its people still have to go through the same clumsy first months. Meanwhile, the first firm is putting its saved hours into its next task. It's on its third redesign while the other is on its first.

You can buy the tool later. You can't buy the practice.

Test yourself

That was a lot of numbers. Let's see which ones stayed with you.

A small example from our own work

Finance has been through this kind of change before, many times. We've lived one small version of it.

Changing the system was the easy part. A tool that reads the invoices for you is different. It changes who does what. That's the factory story again, on a small scale.

"We'll wait until it's more mature"

It's a sensible instinct. Nobody wants to waste money on something that won't last. And AI does make mistakes, which is why the plan below has a checking step.

But the tools will keep changing whenever we start. The habits won't: choosing the task, checking the work, agreeing who signs off. Those carry over to every new tool.

And your team may already be using it. In the US, about one in four workers had used generative AI for work by late 2024.⁸ With no simple rules, people use it anyway, on their own. That's riskier than starting on purpose.

The first time we used Claude in Excel to run a P&L by product, in a previous role, we were amazed by how fast it worked. But there was one problem. It didn't give us exactly what we needed.

After a few rounds, it started to make more sense. In the end, it gave us exactly the result we were looking for.

The takeaway: AI may not give you what you need on the first try. But if we keep at it, we can master it. Then it's not only about speed. It's about quality too.

One way or another, this is the only path we see in the months and years to come.

Start small: a 30-day plan

Starting is smaller than it sounds. About 20 minutes a week is enough.

None of this is exciting. That's fine. A year of small, checked steps gives us a team that knows how to work with AI. No competitor can buy that off the shelf.

Back to that photo

Kodak held the future in its hands in 1975. It was slow and heavy, and it had no face.

Thirty-seven years later, the company wrote that camera into a presentation, as one of its innovations. By then it was a line on a slide.

Our first weeks with AI will look like Joy Marshall's photo.

Take the photo anyway.