Is AI really killing SaaS?

Every few months somebody declares the SaaS business model dead. The argument runs that if a competent person can prompt an app into existence over a weekend, nobody will keep paying per seat per month for software doing roughly the same job. Someone always calls it the SaaSpocalypse.

I do not buy it. The demos are real enough, and I have built enough of these to know you can get something working in an afternoon that would have taken a small team a quarter back in 2019. Shipping the app was never the expensive part of enterprise software.

Watch how software gets bought inside a company with more than about two hundred staff. Someone wants a tool. They have no budget line, so they raise it with a manager, who raises it with finance. Before any of that resolves the request lands in front of a security reviewer who wants a SOC 2 Type II report, a penetration test summary, a data processing agreement, a sub-processor list, and a straight answer about which country the data sits in.

You cannot prompt your way to a SOC 2 Type II. An auditor watches your controls over a window of months and then writes up what they actually saw, exceptions included. The evidence trail is the thing being sold. Access reviews, timestamped. Change management records showing who approved which deploy and when. Offboarding logs proving the contractor lost access the day their contract ended, rather than eight weeks later when somebody noticed.

A generated app carries none of that. It also has nobody on the hook.

The second one is what usually settles the purchase, and it gets almost no airtime in these arguments. When the system holding your customer records falls over at two in the morning, a large part of what you pay a vendor for is that somebody else’s phone rings. There is a support contract. There is an SLA with actual teeth in it, and in a bad enough scenario there is professional indemnity insurance and a legal entity you can pursue. The weekend app has a Slack thread and whoever wrote the prompt, assuming they still work there.

So what is AI genuinely doing to this market?

Squeezing the middle out of it. Products that were a thin layer over a database, sold on the convenience of not building it yourself, are in real trouble. If your moat was that reproducing your tool would be annoying, that moat has drained. Plenty of companies are about to feel this and some of them will not survive it.

What survives is software carrying obligations. Payroll. Anything touching health records. The general ledger. Customer data in a regulated market. Those are rarely the hardest engineering problems in the building, and that is the point: the code is the smaller half of what the buyer is paying for. They are buying an audited process, a contract, and somebody to blame.

There is a version of this argument I would accept. If regulators decide that AI-generated audit evidence counts, or if the compliance frameworks themselves get automated to the floor, the calculus shifts and I would want to rethink it. I have seen no sign of either.

My position, held loosely: AI has raised the floor on what software can look like and left the ceiling roughly where it was. Your prospect can now generate a passable internal tool over a weekend. They still cannot generate two years of access review logs.