In the first week of February, software company shares lost more than $285 billion in value in a few days. The trigger wasn't a recession or a scandal. It was an AI company releasing a set of plug-ins that could do some of the work people pay software subscriptions for. A Jefferies analyst gave it a name that stuck: the SaaSpocalypse.
Eight months on, the dust has settled a bit, and it's a good time to look at what's actually changing, what turned out to be panic, and what any of it means if you run a small business in Ireland.
What actually happened
Most business software is sold as a subscription per person: so much per user, per month. The fear was simple. If an AI assistant can draft the contract, chase the invoice and update the customer record, why pay for twenty seats of three different tools?
Some companies were hit hard. Monday.com fell about 37% in February, though weak guidance played a big part and its own boss said it wasn't yet seeing an AI impact. Atlassian lost over 45% in a month. Legal and data firms like LegalZoom and Thomson Reuters dropped sharply in a single day.
Then reality set in. Business software is more than a few screens. There's the data you've built up, the compliance, the integrations, and the fact that your accountant uses the same package. By October the main software index was back up for the year, and an analyst note from Phillip Securities was titled, simply, "End of the SaaSpocalypse". I prefer how a newsletter from AI researcher Andrew Ng's team put it: software isn't dying, it's becoming AI-native.
AI is writing a lot of the code now
Google's boss said in April that around three-quarters of the company's new code is now generated by AI and approved by engineers. In Stack Overflow's 2026 survey, 73% of developers who use AI assistants use them every day.
But the evidence on whether it makes developers faster is surprisingly muddy. A careful study by the research group METR in early 2025 found experienced developers were 19% slower with AI tools, while believing they were faster. Their 2026 follow-up suggests that's probably flipped, but METR itself calls the new data "only very weak evidence" of how big the gain is.
Where it breaks
What AI is actually good at
Boilerplate code, first drafts, small internal tools, converting data between formats, explaining unfamiliar code, and turning a clear description into a working prototype in an afternoon.
Where it goes wrong
Security, edge cases, and anything where "almost right" isn't good enough. In Stack Overflow's 2025 survey, 66% of developers said their biggest frustration was AI answers that are almost right, but not quite.
The security numbers are the ones that worry me. In a 2025 study by Veracode, 45% of AI-generated code samples contained flaws from the standard OWASP Top 10 list of common vulnerabilities, and newer models didn't do any better. Another analysis found copy-pasted, duplicated code shooting up as AI use grew.
And then there are the horror stories. In July 2025, an AI coding agent on the Replit platform deleted a company's live database during a code freeze, then gave a misleading account of what it had done. Replit's boss called it unacceptable and added safeguards, but it's a good reminder of what happens when an AI is given access to the real thing with no guard rails.
Businesses are building instead of buying
This is the part that's real, and it's what spooked the markets. In a 2026 survey by Retool, 35% of companies said they'd already replaced at least one software subscription with something custom-built, mostly workflow automation, admin tools and dashboards. Retool sells tools for building those apps, so treat the number with a pinch of salt. But the trend matches what you see everywhere: a small tool that does exactly what one business needs is now cheap enough to be worth making.
The same survey had a warning buried in it: 60% of those tools were built outside IT oversight. That's how you end up with a spreadsheet-replacement app holding customer data that nobody backs up, patches or remembers how to fix.
What about jobs?
Here the news is less comfortable, especially for people starting out. A Stanford study found employment for 22- to 25-year-olds in the jobs most exposed to AI fell about 13% compared with older workers. In Ireland, IDA figures reported in October showed entry-level hiring in AI-augmented occupations down 16.2% year on year in the second quarter, with junior software and data roles hit hardest.
Both sets of researchers are careful to say AI isn't the only cause. Interest rates and a post-pandemic hiring hangover play a big part. But if you're an Irish teenager weighing up computer science, the entry door is narrower than it was five years ago.
What it means for a small business
- Don't cancel your CRM or accounts package tomorrow. The big platforms are adding AI themselves, and your data and integrations live there.
- Do look at the small, annoying jobs. The spreadsheet that takes someone a morning every week, the quote that's typed out by hand. These are exactly where a small custom tool or an AI automation pays for itself.
- Watch out for per-seat creep. Software firms are shifting to usage and AI-credit pricing. Check what you'll actually pay before renewing.
- Treat AI-built tools like any other software. Someone needs to own them: security, backups, updates.
- If you use an AI chatbot with customers, tell them. Since 2 August 2026, the EU AI Act requires people to be told when they're talking to an AI.
Where that leaves you
The shock to the software industry this year was real, but the "software is dead" panic was overdone and the markets have mostly recovered. What has changed is the cost of building software. AI writes a big share of the code at the large companies now, and small businesses are swapping subscriptions for tools built just for them. Quality and security still need a person who knows what they're doing. So use AI for the boring, repetitive jobs, keep your core systems where they are, and make sure anything AI-built has someone looking after it.
Sources: The Batch: Cowork plug-ins and the sell-off (Feb 2026) · Nasdaq: Monday.com in February · FNN: Atlassian · Phillip Securities: End of the SaaSpocalypse (Oct 2026) · The Decoder: Google's AI code share (Apr 2026) · Stack Overflow 2026 survey · Stack Overflow 2025 survey · METR productivity update (Feb 2026) · Veracode (Jul 2025) · GitClear code quality · eWeek: Replit incident · Retool Build vs Buy 2026 · Stanford Digital Economy Lab · Irish Times: entry-level hiring (Oct 2026) · iubenda: AI Act August 2026 obligations
