When the hype outlives the headlines: Rethinking the AI threat to SaaS
The narrative of AI wiping out the SaaS industry makes for compelling reading. The reality is considerably more nuanced.
There's a certain irony in the fact that some of the most breathless commentary predicting the death of software-as-a-service has itself been machine-generated.
This is the so-called "SaaS-pocalypse", the idea that AI will make entire categories of business software obsolete overnight. It makes a punchy headline. Whether it reflects where we're actually headed is another matter.
The market is reacting. But to what, exactly?
Software stocks have taken a battering. Some well-known names are down significantly from their peaks, and the selling has been broad, hitting strong businesses alongside weak ones. Yet underneath the noise, most software companies are actually performing well. Growth is healthy. Customers are staying. Some are accelerating.
What markets are really wrestling with isn't the present, it's the future. Specifically, the fear that AI doesn't just change how software is built, but fundamentally changes the nature of the work that software supports. That's not an irrational concern. But it's also not the same as saying the industry is about to collapse.
The problem is that many investors, even professional ones, have never actually used the products they own. Enterprise software is a Rorschach test: it means different things to different investors, because most have never used it, they just project their fears or hopes onto it rather than understanding it properly.
Much of that fear centres on the core idea that AI can now build software so quickly and cheaply that the case for buying it off the shelf simply disappears. It sounds compelling. But it misses how software actually gets built, and what organisations are really paying for.
There's more to software than writing it
AI coding tools are genuinely impressive. Building software is faster and cheaper than it was two years ago, and that gap will keep widening. But organisations were never just paying for code. They were paying for something that works reliably at scale, meets compliance requirements, integrates with everything else they run, and that someone is responsible for when it breaks. None of that has changed.
Before a single line of code is written, someone still needs to have mapped the process, defined the requirements, and understood what the software actually needs to do. AI doesn't generate that knowledge, it requires it as input. Add to that a testing burden that becomes more demanding with AI-generated output, plus security, regulatory compliance, audit trails, and 24/7 availability, and the proposition of swapping out battle-tested platforms for freshly built alternatives starts to look considerably less appealing.
Something that looks impressive in a demo can fall apart the moment real users, real data volumes, and real operational pressures are applied to it. Established SaaS platforms have already been through that process, sometimes painfully, and what customers are paying for, in large part, is the result of that experience.
What AI actually disrupts
None of this means SaaS vendors should feel comfortable. The disruption is real, it just isn't a replacement. It's a shift in what customers are willing to pay for.
For decades, software was priced around users and storage. That made sense when the job of software was to hold information and let people access it. AI changes the expectation. Organisations now want software that does things, not just stores things, automation, insight, fewer manual steps, faster decisions.
That puts real pressure on vendors whose instinct has been to bolt AI features onto platforms built twenty years ago rather than rethink them from the ground up. Customers can tell the difference, and their renewal conversations are starting to reflect it.
One of the more underappreciated shifts is what AI has done to the cost and speed of experimentation. Building a proof of concept has never been easier. Non-technical people are no longer limited to mockups and wireframes, they are shipping zip-tie-and-duct-tape tools that actually work, good enough to test a real idea, onboard early clients, and survive contact with reality. This coin has two sides. On one, competition is growing: the barrier to building internal tools or even nascent SaaS products has dropped dramatically. On the other, it has handed a genuine edge to people who combine product instinct with execution speed. Jakub Sadowski at Surfer is a case in point — a product manager who built a working prototype that has since been folded into a mature, shipping product. Prototype instead of spec. That's a different game entirely, and established vendors are not necessarily the ones best positioned to play it.
Not every product faces the same risk. Software directly tied to headcount, customer support tools, for example, faces genuine pricing pressure if AI reduces the number of people needed. Software embedded deep in how a business runs is considerably stickier, because the cost and risk of replacing it outweighs almost any alternative.
The harder question, which few are willing to ask plainly, is whether the teams who built and maintained CRM platforms, ERP systems, and content management suites are the right teams to build AI-native successors to them. The skill profiles are genuinely different, and institutional expertise in one domain is not automatically transferable to the other.
Adaptation over apocalypse
The most realistic outlook isn't a sudden industry wipeout, it's a period of change in which some vendors adapt successfully and others don't. Point solutions with limited switching costs and clear AI-native alternatives face the sharpest pressure. Platforms embedded deeply in operational workflows, with years of configuration and integration work behind them, are considerably stickier.
The organisations that navigate this well, vendors and enterprises alike, will be those that resist treating this as a binary choice. Replacing everything is as misguided as replacing nothing. The work is in developing the judgment to distinguish between what AI genuinely improves, what it disrupts economically but not practically, and what endures regardless.
The SaaS-pocalypse makes for compelling content, and extinction makes for better headlines than evolution. But evolution is almost certainly what's coming.
