Generated by Codex with GPT-5

This summary covers The Economist’s June 13th, 2026 Business article listed in the contents as Fear the SaaSpocalypse and published under the headline The four SaaSquatches of the apocalypse.

Software-as-a-Service once appeared to be one of technology’s safest business models. A provider could develop an application once, distribute it cheaply over the internet and collect recurring fees from every employee who used it. Artificial intelligence now threatens that formula from four directions: the largest AI laboratories, AI-native startups, companies building their own software and SaaS incumbents trying to reinvent themselves. The industry is not about to disappear, but its most profitable assumptions are becoming much less secure.

AI can sit above traditional software

The largest threat comes from frontier AI companies such as Anthropic and OpenAI. Their coding tools are already capable, but their broader advantage is that AI agents can work across many programs. Traditional SaaS companies such as Salesforce, ServiceNow and Workday tend to dominate narrow business functions. An AI agent could instead become the main interface through which an employee completes a task, calling several specialised applications behind the scenes.

That would reduce valuable SaaS products to infrastructure. Customers might still need the underlying systems, but the AI provider would control the user relationship and could capture more of the value. The labs also have abundant capital and attract AI researchers whom established software companies struggle to hire.

AI-native startups are attacking from below as well. Some focus on particular industries: Harvey builds tools for lawyers, for example, challenging established legal-software providers. Others target business functions such as customer service or IT support. Serval currently works with ServiceNow but ultimately aims to replace it with a system run largely by agents. These companies do not have to protect a legacy product or pricing model, so they can design around AI from the start.

Customers may choose to build

SaaS grew by persuading companies that buying standard software was easier and cheaper than developing it internally. Generative AI is changing that calculation. Employees can now create simple tools with little programming experience, while large organisations can use their proprietary data and expertise to build more ambitious systems tailored to their needs.

The article points to Kirkland & Ellis, a leading law firm, which plans a major multiyear investment in internal AI tools rather than relying entirely on outside vendors. This does not mean every company will become a software developer. Replacing important systems remains slow, risky and operationally difficult. But even a partial shift from buying to building weakens one of SaaS providers’ central selling points.

Reinvention creates a pricing trap

The fourth threat comes from incumbents’ own AI products. Many established vendors charge per employee, or per “seat”. AI agents undermine that model because they can replace human work without needing seats of their own. They also cost more to operate as usage rises, unlike conventional software, whose marginal distribution cost is close to zero.

Vendors are responding with combinations of seat-based and consumption-based fees. Yet their new AI revenue may not grow quickly enough to replace the licences lost when customers need fewer human users. The firms must therefore sell products that can cannibalise their profitable legacy businesses before competitors do it for them.

The pressure will not fall evenly. Cyber-security providers are benefiting as companies spend more to defend against AI-enabled attacks, and Snowflake has reported strong demand for AI-assisted data tools. By contrast, the share prices of Salesforce, ServiceNow and Workday have suffered because their applications and recurring revenue depend heavily on human employees.

The “SaaSpocalypse” is therefore less a prediction of sudden extinction than a warning about control and economics. Traditional software may persist as essential plumbing, but AI agents could own the interface, specialised startups could capture individual workflows, customers could build more themselves and incumbents could erode their own licence revenue. Software once ate the world by making reusable applications extraordinarily profitable. AI may now force that industry to consume its own business model.