FEAR OF THE SAASPOCALYPSE IS TORMENTING TECHLAND / THE ECONOMIST

Artificial intelligence

Fear of the SaaSpocalypse is tormenting techland

Software once ate the world. Now it is in danger of eating itself

A stylized illustration of Bigfoot walking near a city skyline and bridge. It contrasts a mythical creature with a modern urban setting, highlighting the legend's larger-than-life presence. / Illustration: Mike Haddad

While TYPING away at a WeWork in San Francisco recently, your correspondent spied a plane flying across the skyline trailing a banner. 

“SAAS IS DEAD”, it declared in huge letters. 

The software developers with whom he was sharing the co-working space also noticed. 

“Thanks for reminding us,” one groaned.

The stunt, paid for by an artificial-intelligence startup, reflects the increasingly widespread belief in techland that AI presents an existential threat to the Software-as-a-Service (SaaS) industry, which only a few years ago looked unstoppable. 

Since the brutal sell-off in their stock in February, its bosses have been desperately trying to persuade investors that fears of a so-called “SaaSpocalypse” are wildly exaggerated. 

Over the past month or so the share prices of listed American software companies have, on the whole, begun to recover some lost ground (see chart). 

Yet many investors remain cautious.

Their worries are unlikely to dissipate soon, as AI agents that can operate other computer programs become cleverer and more capable. 

It doesn’t help that the threat to SaaS incumbents is coming from four directions: large AI labs; AI-native startups; DIY software development; and the industry’s own disruptive efforts to reinvent itself. 

Call them the four SaaSquatches of the apocalypse.

It is the AI labs that loom largest over the landscape. 

Led by Anthropic and OpenAI, they build models with the most advanced capabilities, are raising mountains of capital and employ the leading AI boffins, whom middle-aged SaaS companies find hard to attract.

So far their coding efforts—Claude Code and Codex, respectively—have been most notable. 

But that is only the start. 

The labs are well placed to take advantage of one of SaaS companies’ big vulnerabilities: their siloed nature. 

Despite numerous acquisitions over the years, businesses such as Salesforce, ServiceNow and Workday have rarely succeeded in breaking out of their specialisations, or “verticals”. 

AI labs, by contrast, threaten to move horizontally, producing agents that operate at a level above the SaaS products, stitching together various programs through plug-ins and a single chat-based interface. 

Some SaaS applications that were once considered snazzy business tools risk becoming more like plumbing.

Meanwhile, AI-native startups are already attacking the vertical dominions of the SaaS giants head on. 

One type is the industry specialist, such as Harvey, most recently valued at $11bn, which makes AI tools for lawyers and is causing consternation for legal-software stalwarts such as Thomson Reuters. 

The AI labs are increasingly offering specialised versions of their bots that perform similar functions.

Other newcomers offer AI tools for particular business functions. 

Numerous AI customer-service startups have cropped up, including Sierra, co-founded by Bret Taylor, the chairman of OpenAI and former co-chief of Salesforce. 

Serval, a two-year-old startup valued at $1bn in December, is taking aim at ServiceNow, which makes software for IT help desks. 

For the moment Serval’s agents work with ServiceNow’s platform. 

But the firm’s ultimate goal is to create a fully agentic alternative.

Some enterprises will be wary of swapping dependence on one vendor for another. 

Instead they may opt to harness AI to create their own DIY software systems. 

The SaaS industry built its success on persuading customers that it is cheaper and easier to buy standardised software programs than to develop custom ones in-house. 

Software companies ploughed a fortune into building their products upfront, then profited handsomely by distributing them across the corporate landscape at almost no marginal cost. 

Now, however, enterprises are embracing a “build versus buy” spirit, says Tim Tully of Menlo Ventures, a venture-capital firm.

This has progressed well beyond “vibe-coding” simple tools to automate tedious tasks. 

Kirkland & Ellis, the world’s highest-grossing law firm, recently said that it planned to invest $500m over the next few years to develop AI tools of its own that draw on the expertise of its staff, rather than relying on third-party ones.

None of this makes the collapse of the SaaS industry imminent. 

Adopting new systems is fiddly and slow. 

And the leading firms still have legions of salesmen to woo customers at lavish events (try finding a hotel room in San Francisco when Salesforce’s “Dreamforce” is on).

Indeed, some corners of the SaaS industry are booming thanks to AI. 

The share prices of America’s three biggest providers of cyber-security software—CrowdStrike, Fortinet and Palo Alto Networks—are up by an average of 52% since the start of the year, as enterprises spend more to defend themselves from AI-powered hackers. 

The value of Snowflake, a data manager, jumped by 36% on a single day last month when it reported surging revenue on the back of strong demand for its AI-powered data-querying tool, among other things.

Others, however, look more exposed. 

The share prices of Salesforce, ServiceNow and Workday are down by an average of 34% since the start of the year. 

All three businesses are built on selling applications to human employees, and rely on a recurring-revenue model in which customers are charged per user, or “seat”. 

That model, however, is beginning to make less sense as these providers offer their own AI tools to keep up with the disrupters. 

One problem is that, unlike traditional software, AI agents cost a supplier more as their customers increase their usage. 

Another problem is that, by replacing human workers, these new tools eat into the revenue that comes from their legacy products. 

AI agents “don’t need seats”, notes Manny Medina, founder of Paid.ai, the startup behind the flying “SAAS IS DEAD” banner, which sells tools to help developers monetise their agents.

Where companies such as Salesforce offer their own AI agents, they typically charge a combination of seat-based and consumption-based pricing, based on how many tokens—the chunks of text processed by AI models—a customer uses. 

Yet although the revenues SaaS companies are generating from these new products are growing rapidly, they are nowhere near large enough to offset the sales they stand to lose if customers buy fewer of their traditional licences, argues Tal Liani, an analyst at Bank of America. 

“The risk is cannibalisation,” he says. 

There lies an irony. 

Not long ago, software was said to be “eating the world”. 

Now the danger is that it eats itself.