SaaS isn't dead. Software without taste is.
AI agents and vibe coding were supposed to end software you pay for. What they really end is software whose only value was that it was hard to build.
Part of our series on taste and shipping fast. The opening essay: Anyone can ship now. Taste decides who wins.
The obituary
The idea that AI would kill software-as-a-service has been building for two years.
In December 2024, Microsoft's Satya Nadella described business applications on the BG2 podcast as essentially CRUD databases with a bunch of business logic, and said that logic is moving to AI agents. Headlines turned it into "SaaS is dead".
Klarna's chief executive told investors the company had shut down Salesforce and would shut down Workday, and the story became "AI replaced SaaS". In 2025 he corrected it himself: they had not replaced SaaS with a language model. They had moved to other tools and built their own systems for bringing their data together.
In early February 2026, new AI tools for legal, finance, sales and marketing work sent software stocks down hard. The S&P 500 software and services index fell nearly 13% in five sessions, and analysts gave the sell-off a name: the SaaSpocalypse.
Add vibe coding to that picture and the argument writes itself: if anyone can describe a CRM to an agent and have one by Monday, why would anyone pay for one?
What is actually dying
Look closely and the obituary is for something narrower. Forrester's answer to the sell-off was blunt: SaaS as we know it is dead, but not SaaS. They expect global SaaS spending to keep growing, from $318 billion in 2025 to $576 billion in 2029. What they expect to stall is the thin point tool with low switching costs, and the habit of charging per seat when agents do more of the work.
In other words: software whose main value was that it was tedious to build. If your product is a form on top of a table, AI has made you easy to copy. That was always a weak position. Now it is an untenable one.
What the code never was
Here is the part the obituary misses. The code was never the product. The product is a long list of decisions someone made so the customer doesn't have to:
Which ten fields matter, and which fifty to leave out.
What happens by default, so most people never open settings.
What the software refuses to do, because it would hurt the customer later.
How it behaves on the bad day: a failed payment, a deleted record, an auditor's question, an outage at 3 am.
Who answers when it breaks, and who is accountable for the data.
An agent can generate the screens in an afternoon. It cannot generate ten years of knowing which defaults people actually need. That knowledge is taste, and it is what customers have always really been paying for.
Build it yourself? Now you own the taste too
When a company vibe-codes its own replacement for a tool it used to buy, it does not just save a subscription. It takes on every decision the vendor used to make, and every one it will need to make next year: security fixes, compliance changes, the edge case in month eight, the person who leaves and took the understanding with them. Sometimes that is the right call. Klarna's own correction is a good reminder of what it really involves: other tools plus their own systems, not one prompt.
The survivors will be the ones with taste
The software that lasts through this will not be the software with the most features. Features are now free. It will be the software with the clearest point of view: built for someone specific, deep in their real work, opinionated about what it leaves out, and trusted with what matters. That is not a technology advantage. It is a taste advantage.
So yes, a lot of SaaS will die. Not because AI writes code. Because AI exposed which products were only code.