Building software gets cheaper every month. Standing behind it does not. That gap is the whole reason a company still hires anyone to build its systems, and it is the part of our work we would defend hardest if someone asked us to cut a price.
The code is no longer scarce. Anyone with a good agent and an afternoon can produce something that looks finished and behaves, most of the time, roughly as intended. What has not become cheap is the answer to a simple question: when this breaks at two in the morning, whose name is on it?
Why generated software rarely leaves the desk it was made on
We see the same pattern in companies of every size. Someone in finance or operations builds a small tool for themselves. It works. It saves them an hour a day. Then the moment arrives when they have to hand it to a colleague, and they hesitate.
The hesitation is correct. They are not worried the tool is bad. They are worried about being the person answerable for it when it behaves strangely during the quarter close, and they know they are not the expert who can promise it will hold. So the tool stays on one desk, and the company gets one person's productivity instead of a department's.
People are happy to generate software. They are not happy to be accountable for decisions in a field where they are not the expert, at least not where there are stakes. That is not a technology problem, and no model release will fix it.
What accountability is made of
The word sounds soft until you break it into the things it actually requires.
It requires that someone read the logs on a day nobody asked them to. It requires a person who decides what ships and who can explain, afterwards, why they decided it. It requires that a change arrive with the proof it works attached, rather than with a reassurance. It requires that when production is down, a phone gets picked up by somebody who knows the system rather than somebody reading a runbook for the first time.
None of that is typing. All of it is judgment, and judgment is what stays expensive.
This is why our own delivery runs the way it does. Agents do most of the building, in their own working copies, against a record of everything the project has ever said. An independent verifier that never wrote a line of the work goes at it and tries to break it. Then an engineer with a name reviews what survived, and a human gate stands between that and production. Every stage but one is machinery. The one in the middle is a person, standing at the only place a person still belongs.
What this means for the price
Pricing the stamp rather than the code changes the shape of a quote. The first number is bigger than a demo-seller's, because it includes the migration that cannot lose a row, the permission model that knows an ex-employee from a current one, and the failure that arrives as a readable message instead of a blank screen. Every number after it is smaller and quieter, because the surprises were engineered out while they were still cheap.
It also changes what we can honestly promise after delivery. The maintenance subscription is not an insurance policy against our own shortcuts. It is the cost of keeping a name attached to a system that is still moving: hosting it, watching it, shipping the changes your business asks for, and having an engineer who knows it when something goes wrong.
We think the gap between the price of building and the price of answering only widens from here. Generation becomes free. Accountability does not, because it is made of human attention, and human attention is the one input that agents make more valuable rather than less.
So when a company asks us what exactly they are paying for, we try not to reach for the feature list. They are paying for software they own, built fast because agents built most of it, and a name at the bottom of it that stays after the handover. The first two are increasingly ordinary. The third is the whole product.