A suggestion by Anthropic that its Claude Code tool could be used to automate the modernization of an antediluvian programming language, COBOL, still an important sideline for IBM six decades after it was first deployed, sent IBM stock tumbling on Monday.

The company suffered a 13.2%  drop in its stock price, its biggest fall since the dot com period a quarter of a century ago.

But IBM is not the only company suffering because of the AI hype. Across a swathe of SaaS companies, a state of near panic seems to be spreading as the potential of AI to automate software creation and management starts emerge. Salesforce, Atlassian, ServiceNow, and Snowflake have all seen heavy selling in an event one trading company melodramatically dubbed the “SaaSpocalypse.”

But according to a hypothetical scenario in The 2028 Global Intelligence Crisis proposed by Citrini Research and co-author Alap Shah of Lotus Technology Management, things could soon get a lot worse for SaaS companies, and beyond them, for the wider US and global economies.

A thought exercise

The report speculates how the rise of AI might look from the vantage point of 2028, barely two years after an imagined peak for the S&P 500 of 8,000 points in October this year.

“Productivity was booming. Real output per hour rose at rates not seen since the 1950s, driven by AI agents that don’t sleep, take sick days, or require health insurance,” said Citrini Research in what it termed a thought exercise. But something else was happening as well: “The owners of compute saw their wealth explode as labor costs vanished. Meanwhile, real wage growth collapsed.”

As companies chased the higher productivity of AI, they laid off workers. But as workers were laid off, demand for products and services declined, leading companies to chase even more productivity and automation. “In every way, AI was exceeding expectations, and the market was AI. The only problem…the economy was not,” imagines Citrini.

The first to feel this pain will be SaaS platforms. In future, AI agents will perform today’s SaaS tasks for enterprises and consumers alike, weakening their appeal and forcing vendors to offer significant customer discounts just to stay afloat.

Subscription model in peril

The attention being given to this downbeat projection arrives closely on the heels of OpenAI’s early February announcement of its new Frontier enterprise agentic platform, reportedly recently touted to investors as a direct rival to SaaS.

As consultancy Xpert Digital bluntly put it: “If AI agents take over the work of entire departments in the future, or if companies simply generate their own code, the foundation of the lucrative subscription model will collapse.”

Do these troubles for SaaS vendors mean that customers will be able to negotiate better prices for licenses in 2026?  And how likely is it that some will jump ship altogether and start using systems from AI companies rather than the AI-enabled software that SaaS companies will try to temp them with?

Easy for AI to write, not easy to run

Experts contacted by CIO.com remain skeptical that AI’s triumph will be as simple to achieve as Citrini Research envisions. First, AI’s ability to perform business tasks as well as SaaS platforms do is untested and unproven.  Second, even when they approximate what these systems do, none of it will be auditable. And then businesses must factor in AI agent security worries and the question of whether they will even be cost-efficient.

“AI makes it dramatically easier to write software. It does not make it easier to run enterprise software. Those are two very different problems, and most of the cost lives in the latter,” said SAVVI AI CEO, Maya Mikhailov.

“The moment you internalize building software, you also inherit security, compliance, uptime, integrations, and 24/7 support. It sounds good in theory, but costs and complexity will squarely land on the bottom line,” she added.

According to Collin Hogue-Spears, technical expert at Black Duck Software, there are also questions about AI’s reliability. “OpenClaw went from zero to 135,000 exposed instances in weeks because it executes workflows fast. It does not produce audit evidence, satisfy license obligations, or generate the compliance documentation that a regulator demands before that code ships,” he said.

“AI agents execute tasks,” he noted. “They don’t produce the evidence trail that stands between your company and a regulatory enforcement action. The pattern repeats across every vertical: AI compresses commodity features and expands governance obligations.”

That said, SaaS customers should not take any 2026 price increases lying down. The current weakness in this sector is an unusual opportunity. “SaaS vendors face the most credible pricing pressure since cloud computing displaced on-premises licensing, and the negotiation window is open now,” said Hogue-Spears.

This article originally appeared on CIO.com.