I have been thinking about what a software company becomes when software itself stops being particularly scarce.
For most of my career, the economics were straightforward. We had ideas about how computers could make training humans better. Turning those ideas into working software was expensive. It required developers, infrastructure, testing, support, operations and time.
So we sold hours. Then seats. Eventually SaaS.
That was how we paid for the next swing.
Looking back over more than twenty years of Thinking Cap, though, I don’t think selling seats was ever really what we were trying to do. We have always made one fairly unique thing: our best current attempt at how computers can make training humans better.
The LMS was one manifestation of that. The business model financed it.
AI is making me wonder whether those two things need to remain so tightly connected.
Mozart needed an orchestra
Mozart could hear music in his head, but to hear the symphony in the world he needed an orchestra.
Software has worked much the same way. One person could conceive an extraordinary system without having any practical ability to build it alone. The idea had to be distributed among developers, designers, database people, testers and infrastructure people. Eventually the pieces came back together and, if everything went well, the originator got something resembling what they had imagined.
This gave large organizations an enormous advantage. They could gang up on implementation.
AI changes that. Mozart suddenly has an orchestra available whenever he wants one. He can hear four bars, stop, change the score and hear it again. The cycle between conception and realization that used to take months can happen in hours.
That doesn’t make everybody Mozart. In fact, the companies building the most capable AI systems seem to be demonstrating the opposite. They are trying to automate enormous amounts of intellectual labour while paying extraordinary sums to secure a remarkably small number of particular human minds. They have the orchestra too — apparently they still care rather a lot about who is writing the music.
Creativity loves company
None of this makes me think the future belongs to solitary geniuses sitting in rooms talking to machines. Creativity loves company.
A symphony orchestra may contain vastly more aggregate musical ability than the Beatles. You still can’t assemble enough excellent musicians to manufacture Lennon and McCartney.
Another independent mind brings something different from additional processing capacity. Someone sees what I don’t see. They disagree. They misunderstand my idea in an interesting way. They make something that makes me jealous. They take a swing that makes me want to take a better one.
John with AI would have been formidable. Paul with AI would have been formidable. John and Paul, both with AI, would have been something else.
So perhaps the future technology company looks less like a pyramid and more like a band: a small number of exceptional people with genuinely independent minds, surrounded by enormous machine capability.
The test pilots
Developer doesn’t seem quite the right description for what those people do anymore. I increasingly think of them as test pilots — not because they operate the controls, but because the job is taking a new machine toward the edge of what’s understood, finding out what nobody predicted, and bringing that knowledge back.
What happens if intent becomes a first-class part of a company operating system? Build it and find out. What happens if email becomes evidence flowing into organizational memory? What happens if the durable thing isn’t the document but the knowledge and intent that caused it to exist? At some point I don’t want to theorize about these questions anymore. I want to fly the aircraft.
That may be the increasingly scarce human occupation: operating at the frontier where neither the human nor the machine yet knows the answer.
And if that’s the scarce thing, we need to think differently about the cost of people. If ten extraordinary people surrounded by AI can continuously produce ideas of enormous value, their cost isn’t overhead surrounding the product. They are what is being funded. The objective isn’t to accumulate inexpensive developers. It’s to attract the test pilots, keep them together, give them extraordinary tools, and get everything else out of their way.
We own the machine
I keep coming back to a mining analogy.
We don’t own the coal face, and increasingly I don’t think we need to be terribly protective of every piece of coal we have already extracted. We own the machine that chews new rock. For us that machine isn’t CapCom, Tapestry or the LMS — those are things the machine has produced. The machine is the people, accumulated knowledge, doctrine, tools, history, judgment, customers, mistakes, AI capability, and the habit of continually putting ideas into contact with reality.
Someone may eventually be able to reproduce yesterday’s CapCom remarkably quickly. That bothers me much less if we remain capable of producing tomorrow’s.
Customers turn the wheel
This is where I think about TSMC. Obviously TSMC has secrets — that’s not the interesting part of the analogy.
The best customers bring a foundry difficult things to make. You choose it because what you’re doing is difficult enough that there will be problems you haven’t anticipated, and you want those problems occurring inside an organization unusually good at solving them. The customers turn the wheel.
I increasingly see the roughly two million learners touched by our systems the same way. One customer has a ridiculous organizational hierarchy. Another has an unexpected regulatory requirement. Another has a strange integration. Another operates at enough scale to break an assumption. Another wants to do something that simply doesn’t fit our current conception of an LMS.
Those aren’t annoyances surrounding the product. They are part of its intellectual supply chain. Most problems should disappear into deterministic software and models. Some won’t. Those are interesting. They reach the edge of what the existing machine knows and occasionally deserve to travel upstream to the test pilots.
Two million learners aren’t merely seats — they’re an enormous sensor network telling us where reality disagrees with our ideas.
Which is why we can’t have many secrets
If this is the model, secrecy becomes actively counterproductive. I don’t mean customer information, credentials, vulnerabilities, or anything entrusted to us. I mean our ideas and our work.
If our future depends upon attracting unusual minds and organizations with unusual problems, those people need to see what we’re doing. We need to scream it from the rooftops. Explain Tapestry while we’re building it. Show our experiments with intent, evidence and organizational memory. Explain why we think deterministic software should surround models. Publish things that work and things that don’t — not six months later when Marketing has turned them into product announcements. While we’re doing them.
The work is how the people we want find us. Some extraordinary person somewhere should encounter it and think, these people are working on the things I think about. Some organization with a genuinely difficult problem should think, these are the people I want looking at my problem. Those are exactly the inputs the machine needs: great minds and hard problems.
Once I look at it that way, being copied seems considerably less threatening. If somebody implements something we publish, they have something we already had. Perhaps they improve it and we learn something from them.
Competition doesn’t disappear. I wouldn’t want it to. Maybe it becomes more like a game. The Beach Boys make Pet Sounds. The Beatles hear it and have to answer. You don’t prevent the other band from hearing your record. You want them to hear it.
A different bargain
This makes me wonder whether operating software and funding its continued invention need to remain the same transaction.
Today a customer pays us and somewhere inside that payment are servers, backups, security, support, development and R&D. Those are increasingly different things. There is real work involved in operating systems for millions of people. If an organization wants us to take responsibility for that, pay us to run your shit. If they don’t, fine. Run it yourself.
Perhaps there is another relationship in which organizations fund the people advancing the system because they want those people to remain together, encountering their problems and everybody else’s, and producing whatever comes next. That starts looking less like SaaS and more like an industry-funded R&D collective.
AI could greatly expand the territory where that makes sense. If some piece of learning infrastructure is fundamentally solved, why should fifty companies pay fifty groups to keep rebuilding it? Put more of that in the commons. Spend scarce human attention on what isn’t solved.
Someone will ask the obvious question: what stops a better-funded rival from taking everything we publish and simply outrunning us with it? Nothing — and I don’t think that’s the vulnerability it looks like. They’d be taking yesterday’s coal, not the machine that cuts it. If a rival can absorb our open work faster than we can produce the next thing, we don’t have a secrecy problem. We have a test-pilot problem. The actual defense was never the lock on the software. It’s whether the people who keep finding the next thing are still in the room with us. Lose them, and no amount of secrecy would have saved us anyway. Keep them, and no amount of openness will cost us.
I can imagine our industry supporting a number of small collectives with genuinely different schools of thought. They might use one another’s software and steal one another’s good ideas. Competition moves toward contribution. Who found something? Who moved the frontier? Who attracted the people everybody else wishes they had? Who is producing enough useful thought that the ecosystem wants to keep them alive? That’s still competition. It might be ferocious competition.
And there is a human reason I want this
AI is driving us apart. It is, and I think pretending otherwise is silly.
Every month I discover more things for which I once needed another human and now don’t. Research, programming, editing, analysis, explanation, even conversation. These are enormous gains. But those little dependencies were also part of the glue that forced human beings into one another’s lives.
I don’t think the answer is to preserve unnecessary work so we continue needing each other. I’d rather build institutions around wanting each other.
Creativity loves company. Another person has their own ambitions, tastes, experiences, blind spots and ideas. They can surprise me without first having to learn what kind of surprise I would enjoy.
Maybe the future company doesn’t say we need all these people because there are too many tasks for one person. Maybe a much smaller collective says the machines can perform an astonishing amount of the work, but these are the people we want in the room while we decide what is worth doing.
Small groups who seek one another out because they make one another more creative, surrounded by machines that let them make almost anything they can conceive, competing with other groups doing the same thing and showing one another their work.
I find that future enormously appealing. And if none of it appeals to you, don’t worry. There will be soma, feelies and Obstacle Golf aplenty.
Maybe this is what we were doing all along
The more radical this sounds, the more familiar it feels.
Thinking Cap has spent more than twenty years taking repeated swings at the same question: how can computers make training humans better? We sold hours because that paid for it. Then seats. Then SaaS. Those were financing mechanisms. They weren’t the purpose.
Maybe SaaS remains the best way to fund this for another twenty years. I don’t know. But I don’t want to confuse the financing mechanism with the thing being financed.
If software becomes abundant, give more of it away. If hosting remains difficult, charge for hosting. If customers want to host it themselves, let them. If another group has a great idea, use it. If we have a great idea, tell everyone.
I don’t particularly want to build an organization whose value depends upon a mountain of software nobody else is allowed to touch. I’d rather build one whose value comes from an unusual group of people who keep finding things worth doing.
Give them machines, hard problems and one another.
Then see where they fly.