AI is a payroll line, not a feature
The market still prices AI as a feature. Something you add to a product, demo on a stage, and charge extra for. That is the wrong line on the spreadsheet. AI is a payroll line. Within three years the winners won't have AI products. They'll have AI cost structures: same output, a fraction of the headcount, margins nobody else can match.
The difference matters because features compete on novelty and cost structures compete on physics. A feature can be copied in a quarter. A cost structure compounds. When your operation produces the same output with a fraction of the payroll, you can price where competitors lose money, and they cannot follow you without rebuilding themselves first.
Most AI announcements fail this test. Autonomy theater (agents that promise to replace judgment) sells demos, not outcomes. The real returns in AI are boring: removing the work nobody should be doing. In the operations I have run, that is close to 40% of the day. The reading, the routing, the rewriting, the reconciling. Nobody keeps a slide about that work, which is exactly why it is where the margin lives.
There are two phases, and most companies are stuck in the first. Phase one is trying the tools: everyone gets a license, productivity ticks up, the org chart stays the same. Phase two is rebuilding the operation around the tools: you take the playbook you wrote for people and turn it into one machines can run. I spent years making commerce implementations faster through people. The playbook was the asset: charter, methods, standards, written down. That same playbook is becoming one agents execute. Context in, implementation out.
I am not writing this from theory. At Petbee, which I co-founded and where I sit on the board, we closed our developer gap by going AI-native, with agent-driven development at the core. More than 1,200 pull requests shipped in 2026, on a team small enough to fit at one table. That is not a feature we sell. It is the cost structure we run on.
None of this means AI replaces people. I do not believe it will, for a reason no benchmark measures: trust. In the end, a person trusts another person. When your name is on the line, you want a real name and a full address answering for it. I take full ownership of my actions either way, and the executives who actually sign the contracts want exactly that.
And even with AGI, I do not think AI will match the creative power of an exceptional human brain. There will always be a human in the loop. The ones who master AI to scale their knowledge, their capacity, and their performance will thrive. The ones who refuse will compete against them.
So yes, companies will run with fewer people. But companies are built by brilliant people, and they are the asset everything else compounds on. Fewer people does not mean less care: the people who remain are the ones building your future, and taking care of them is mandatory.
The test I apply to any AI claim is the same test I apply to my own: what line of the P&L does it change, and by how much? If the answer is an adjective, it is a feature. If the answer is a number on payroll, it is the future arriving early.
The future isn't predicted. It's built by the ones already living there.