Buy a suit off the rack and it fits no one exactly. You learn to stand so it hangs right, and you stop noticing you are doing it.
Companies have worn their software the same way for thirty years.
The software almost fits. Close enough. You change how your team works so the tool makes sense. You rename your own steps to match its buttons. You keep a spreadsheet on the side for the parts it never quite handled. After a while you stop noticing the adjustments, the way you stop noticing a jacket that pulls a little at the shoulder. The fit was never the deal. The price was the deal.
That kind of software has a name. The industry calls it SaaS, software as a service: one program, built once, rented to everyone, usually billed by the person per month. A seat. It is the model behind almost every business tool you have ever used, and for thirty years it was the obvious way to buy software. For a reason worth being honest about.
Software cut to one company always fit that company better. That was never in question. It lost on price. To build software around your company, someone first has to understand your company: every team, every exception, every word you use differently from the firm down the street. Then they have to keep maintaining it, because the day your business changes, the software starts to rot. Renting the same tool as everyone else spread those two costs, understanding and upkeep, across thousands of buyers. You gave up the fit, and in exchange you stopped paying to be understood and stopped paying to be maintained. For thirty years that was the right trade.
Here is what changed. The two things that made tailored software expensive, understanding a company and maintaining the result, are the two things AI turned out to be best at.
Reading ten years of a company's own documents and finding the pattern inside them is close to free for these systems now. Writing software to that pattern, and rewriting it when the company changes, gets cheaper every few months. The expensive part of bespoke was never the code. It was the understanding, and understanding just stopped being scarce. So the old trade is inverting. When software built to fit one company costs about what off-the-rack costs, there is no reason left to keep wearing off-the-rack. The tool can finally be cut to the company, instead of the company quietly altering itself to fit the tool.
You can see the market bracing for this already. Early this year, investors knocked about a fifth off the value of the big software companies in a matter of weeks, on a single worry: if one AI agent can do the work of ten people, who keeps paying for ten seats? The panic did not hold at the level of the sector. The software index clawed back much of the drop within months, though the companies most exposed to the fear stayed well down, and at least one big vendor still sells its AI agents by the seat. The stock swing is not the point. The fear underneath it was roughly right and its reason slightly wrong. The threat to renting by the seat is not only that agents replace the seats, but that the same technology makes the tailored alternative affordable for the first time.
I should be careful here, because the honest version of this is quieter than the headline. Software companies are not about to disappear. Spending on business technology is still climbing, not falling. The firms that sell these tools will adapt, change how they charge, and many of them will be fine. I rent plenty of their software myself, and I will again tomorrow. What is ending is not the industry but the bargain at the center of it: you, quietly altering your company to fit a tool built for everyone else.
Here is what the fitted version looks like in practice. A global brand came to us trying to decide whether to enter a market it had never operated in. The whole decision came down to one question. Would the numbers hold once you accounted for an unfamiliar supply chain, an unfamiliar way of reaching customers, and an unfamiliar buyer, each in a different part of the world. Their existing financial tools could not answer it. Not because the tools were bad, but because they were built for the company the brand already was, in the markets it already knew. They had no way to hold a question shaped like this one.
So we built one that did. A model cut to the brand's own situation: supply, distribution, and demand from three regions in a single place, the entry scenarios run under different assumptions, and one honest view of where the answer was solid and where it was not. Eight billion dollars of possible market, weighed down to one decision. Enter, or hold. The off-the-shelf tool had not failed because it was bad software. It failed because it was everyone's software, asked to hold one company's question.
What happened to that brand is the good version. The common one is failure. Companies are buying AI in bulk and quietly scrapping it. S&P Global found that 42% of them had scrapped most of their AI projects, up from 17% a year earlier. A widely-read MIT study of companies investing in generative AI found something sharper still: about 95% were getting no measurable return on it. The reason was not that the models were too weak. It was that the tools never learned the business they were dropped into.
This is the part most companies have not realized yet. The fix is not a better generic tool but something they already own, something no vendor can sell them: their own record. The years of documents, decisions, exceptions, and hard-won judgment a company's people pile up just doing the work. A vendor builds for the average customer, and there is no average customer, so the things that make your company itself were never in any product. They were in your own files the whole time, waiting to be read. Two companies can buy the same software. No two companies have the same ten years.
And most companies cannot read their own files yet. A decade of their best work sits in a document store nobody can really search. The work is to turn that pile back into something the company can actually ask a question of, and get its own answer back.
What you get when software is built from your own material is not only a better fit. You own and operate it, instead of renting it and hoping it stays the same. It does not raise its price on you overnight or quietly retire the one feature you depended on. It is yours, the way your own records are yours.
For thirty years the question a company asked its software was: how do we fit ourselves to this? That question is ending. The next one runs the other way: what would software look like if it were cut to us?
Most companies already own the answer. It is sitting in their own files, in their own words, waiting for someone to read it back to them. For thirty years your software almost fit, and you learned to stand so it hung right. That is the part that is ending. What comes next is cut to your measurements, from your own cloth, and it is yours to keep.