There's a question I've been asking myself for most of this year: as AI changes how software gets built, will the clients we serve still exist?
The honest answer is that one kind won't.
The end of buying hands
For twenty years the largest thing our industry sold was capacity. A client had a technical lead and a backlog; what they lacked was people. Vendors supplied them. The trade calls it staff augmentation.
Those clients are leaving, and they won't be back — not because the vendors failed, but because what they were buying got cheap. They bought hands to write code, and AI has pushed the price of a pair of hands toward zero. A technical lead who uses AI well needs perhaps a third of the people he needed three years ago. Earlier this year a colleague told me his largest client had cut a team of twenty by more than half. Not for a different supplier. They just didn't need them.
That's not bad news. It's a fact, and the useful thing to do with a fact is work out what's left.
Who keeps the difference
This is the part I think matters most, and it's the part easiest to get wrong.
AI lets the same outcome be reached in less time. That saved time is worth money. Whose is it?
If you pay by time, it's yours. Same result, fewer hours on the invoice. That's fair.
If you pay a fixed price, it's the engineer's. Same price, less of their time spent — the hours they got back belong to them. That's also fair.
The one place it must not go is to the company in the middle.
A company that captures the AI dividend — by encouraging engineers to bill hours the work no longer needs, or by quietly cutting what engineers earn on fixed-price work — has turned AI into a tool for squeezing both ends at once. The engineer loses any reason to use it well. The client sees none of the benefit. What's left is a layer that takes and adds nothing, and both sides will eventually leave it. They'd be right to.
Our answer is to fix the company's share and refuse to let it move. Of the engineering fee, 70% goes to the engineer doing the work, 10% to the Account Owner responsible for you, and 20% to the company. The 20% does not grow because AI arrived. It's what makes the other 80% flow to where the value is being made.
What "an engineer who can use AI" actually means
Everyone is trying to hire engineers who can use AI. I think the phrase is a mistake.
There is no such thing as an engineer who can't use AI. The tools are there; you open them and they work. That's not a skill.
What AI does is multiply whatever you bring to it. Bring an understanding of the business — why this feature exists, where the data comes from, who is affected when it breaks — and AI makes you ten times more useful. Bring nothing but the ability to type a prompt, and you produce code that looks right and doesn't work, faster than ever. Zero times anything is still zero.
So the definition is simpler than the industry makes it: an engineer who can use AI is an engineer who understands the business. There isn't a second kind.
That's why we insist on things that look like service-model choices but aren't: engineers talk to you directly, stay on your work for years, and where you want it, run your production environment. Those are not features. They're the only way we know to make an engineer understand a business well enough for AI to be worth anything in their hands.
Companies and engineers, honestly
Put those three together and you arrive somewhere uncomfortable.
A truly capable engineer — one who understands the business, uses AI well, and can hold a direct, long-term relationship with a client — doesn't need a company. They could take the work, deliver it, and bill for it themselves. Before long, AI will be matching engineers like that to clients directly.
But companies need engineers like that. Badly.
Those two sentences aren't in conflict; together they say what a company is for. The old question was "how do we hire the right engineers?" It's the wrong question now. The new one is: how do we make engineers more valuable — so valuable they don't need us, and still choose to stay?
There are things an individual can't offer that a company can. Holding a production environment means someone has to answer when the engineer is on holiday, or ill, or asleep. Cross-border contracts, data protection agreements, professional insurance. A replacement when a client needs one. And an environment where people get better: new engineers rotating in, knowledge staying put, a good developer becoming someone who is answerable for a business. That is what the 20% has to buy. If it doesn't, there's no reason for the 20% to exist.
So who the clients will be
Back to the question. There will still be clients. But not ones buying hands — ones who need someone to own the outcome. Mostly founders without a technical leader of their own, for whom the software is the business, who don't need more people but need a small team that understands what they do, keeps the system running, and is still there in three years.
Those clients won't want twenty engineers. They'll want three, for ten years. Smaller teams, more clients, each one deeper.
That's what we're becoming. Not because AI is forcing it, but because it's what this business should have looked like all along. AI just made the cost of every other way of doing it impossible to hide.