THE FOUNDING ARGUMENT · AUGUST 2026 · PAPER 7
The most important decision you will make about AI was already made once, in 1955, by people who had never heard of your business. It is yours to make again.
CAPACITI AHI INC. · TORONTO · CAPACITIAHI.COM · PATENT PENDING
In the early 1960s there were two laboratories near the Stanford campus, about a mile apart, paid for out of the same government office. They spent a decade proving opposite things.
One was run by John McCarthy. He had put the words artificial intelligence together for the first time in 1955, and when he opened his lab in 1964 he had a stated goal: build a working artificial intelligence inside ten years. A machine that does what a person does. The person comes out of the loop.
The other was run by Doug Engelbart. Two years earlier he had written a paper called “Augmenting Human Intellect,” and his goal was the reverse. Use the machines to make the person in the loop more capable. Never take them out. Engelbart’s lab went on to invent the mouse, the word processor, video conferencing and the hyperlink. Most of what you did at your desk this morning came out of that building.
Both men were serious. Both were funded by the same person, out of the same agency, at the same time. They disagreed for the rest of their careers about what any of it was for.
McCarthy won the naming fight.
The field got called artificial intelligence. It did not get called augmented intelligence. And sixty years of headlines, funding rounds, university departments and product roadmaps followed the name.
That is worth sitting with, because it has a practical consequence. When you read that AI is coming for your people, you are not reading a law of nature. You are reading the long echo of a naming argument settled in 1955.
Names set the default question.
Because the field is called artificial intelligence, the question every vendor has been trained to answer is how many fewer people will I need? If Engelbart had won, the default question would have been how much more can each of my people do?
Same technology. Different question. And you get very different products, because products are built to answer the question the buyer actually asks.
The machinery does not care which one you ask. You should.
People mash two separate questions together, and pulling them apart is the most useful thing in this paper.
The first question is how it is built. Enormous models and raw computing power, or smaller systems with more structure built in? That is a real argument among serious people and it is not settled. It is also, respectfully, not your problem. You do not ask what alloy is in the wrench.
The second question is what it is pointed at. At the work, or at the worker?
That one is entirely your problem, and nobody can answer it on your behalf.
Here is the plain version, and it works on any vendor who ever sits across from you. Ask them: when this is running properly, does my team do more, or do I have fewer people?
Then watch. Most cannot answer, because most have never been asked. The ones who say “fewer people” are being honest, and now you know exactly what you are buying. The ones who say “both” have not thought about it.
| Pointed at the worker | Pointed at the work | |
|---|---|---|
| The question asked | How many fewer people do I need? | How much more can each person do? |
| What you buy | A substitute for capability you used to own | A multiplier on capability you already have |
| Where the value settles | With whoever owns the machine | Inside your business |
| Three years on | You rent your own capability back, at their price | You own something a competitor cannot buy |
I want to be careful here, because “put people first” is the sort of thing everybody says and nobody costs out.
The version I find hardest to argue with comes from Erik Brynjolfsson at Stanford, and it is about bargaining power rather than sentiment. If a machine is built to imitate what your person does, your person has nothing left to bargain with, and the value moves to whoever owns the machine. If the machine is built to extend what your person does, your person is still in the transaction, and the value stays where they are.
He calls the first path the Turing Trap. Not because it fails. Because of where the money ends up when it works.
Put it in a service business. A company that hands its diagnostic expertise to an outside algorithm has quietly turned an asset into a subscription. It now rents its own capability back, at a price somebody else sets, and that price only ever moves one way. A company that pushes the same expertise out to every technician instead owns something no competitor can purchase: unusually capable people, at scale, in a market the competitor does not understand.
Both companies bought AI. Only one of them still owns anything at the end of it.
Researchers from Harvard, Wharton and MIT ran a study on 758 consultants at Boston Consulting Group. Given AI tools, they finished twenty-five percent faster and produced work rated forty percent better. The most junior people gained forty-three percent. Least experienced, biggest jump. That single statistic is the entire promise of this technology.
Then there is the number nobody quotes.
The people who took what the machine gave them without questioning it performed nineteen points worse than the people who used no AI at all.
Worse than nothing.
Read it twice, because it inverts the thing everyone worries about. The danger is not that the machine becomes too clever. The danger is that your people stop arguing with it. It sounds certain, it is four o’clock, they are tired, and the checking quietly stops.
That is a management problem with a management solution. It is also the reason we build the way we do.
This part is harder to explain than it ought to be, so let me avoid the vocabulary altogether.
We do not build chatbots. A chatbot knows what is on the internet. What we build knows your business: your products, your procedures, your regional conditions, and the specific things your best people learned the hard way over thirty years.
There are three roles, and it is easiest if you picture them as three people.
The coworker has three habits that are built in and are not up for negotiation. It shows you where every answer came from. It says I don’t know rather than producing something confident and wrong. And it hands the question upward when it is out of its depth.
What that looks like on a Tuesday. A second-year technician is standing in a mechanical room in front of equipment he has not seen before. In the old arrangement he phones the one person who would know, that person is on another call, and the customer is told somebody will come back tomorrow.
Instead he describes the symptoms out loud, hands still on the tools. The coworker walks him through the diagnostic sequence the retiring veteran would have used, names the two likeliest causes, and tells him which to rule out first. It also tells him where that came from.
He fixes it on the first visit. The customer never learns he was a second-year technician, because for the twenty minutes that mattered, he wasn’t one.
Engelbart could make one person more capable at their desk. That was the achievement, and it was enormous.
What he had no way to do was catch what that person learned and hand it to the next one.
That is the piece we added, and I would argue it is the whole business.
When your senior technician solves something nobody has seen before, that solution does not leave with him. It becomes part of what the system knows. The next technician who meets the same problem, in a different truck, in a different city, next year, gets the benefit of it. Every problem solved makes the next one cheaper to solve.
We call it the accretive loop. Capture, learn, see.
The commercial consequence is the part to hold onto. The knowledge compounds inside your business, not inside ours. You own the fuel. We own the engine. A competitor can license the same engine tomorrow and still start from nothing, because what makes yours worth anything is three decades of your people’s judgment, and that was never for sale.
This is also why it matters now rather than eventually. For every five skilled trades workers leaving North America, two replace them. By 2030, 2.7 million people retire out of the trades, and every one of them is carrying something that was never written down. We have been losing that for a hundred years, one retirement lunch at a time, and calling it normal.
You cannot hire your way out of that arithmetic. You can only capture faster than it leaves.
Three tests. They are cheap, they take an afternoon, and they work on any vendor in the market.
Apply all three to us. We would rather be measured on them than on a demonstration.
McCarthy and Engelbart never settled it. One of them won the naming rights. The other one built the mouse.
Sixty-four years later the same argument turns up in your business wearing a subscription agreement. Nobody is going to resolve it for you, and the vendor asking for your signature has already chosen a side, whether or not they have mentioned which.
The technology will keep improving either way. That part is not up to you and never was. The only piece that is genuinely yours is what you decide to point it at.
Point the intelligence at the work. Not at the worker.
Dennis Gray, Founder · Capaciti AHI Inc. · Toronto, Ontario, Canada
Superhuman work. Human in command.
This is the argument. The other six are the working papers underneath it — the deployment playbook for standing this up beside your existing operation without the organisation rejecting it, the economics of why captured knowledge compounds, what it does to a dealer network, and how the trust architecture holds up to somebody who audits it for a living.
They take a name and an email, and they arrive on the next screen. No sequence, no list to escape from.