Why the SaaS apocalypse may be overstated

Agentic AI could expose as much as $234 billion in enterprise application spending by 2030, according to Gartner. Yet the shift is more likely to reshape software-as-a-service than erase it, executives and analysts told ZDNET.

The threat is disintermediation: AI agents could complete tasks across multiple systems, reducing the need for employees to use traditional enterprise software interfaces. Gartner estimates that AI-driven interactions will account for roughly 20% of enterprise SaaS spending by the end of the decade.

Those projections have fueled talk of a “SaaS apocalypse.” Some estimates suggest the market has lost roughly $300 billion in SaaS valuations over the past 18 months, driven by fears that AI agents will replace established tools.

Recommended reading

AI’s Biggest Barrier Is Fragmented Data

Why enterprise software may survive AI agents

Shannon Kalvar, research director at IDC, said the apocalypse is overstated, although disintermediation is real. Users could ask Claude, ChatGPT, Perplexity, or another model to write code and combine capabilities from several systems, creating what Kalvar called an “ephemeral application” that becomes a new work surface.

“The SaaS apocalypse is overrated, but your use of the word disintermediation is correct.”

That shift threatens vendors whose main value is an application interface and a collection of standard functions. Kalvar said providers need to identify the capabilities that remain difficult to reproduce, including proprietary data, specialized logic, governance, security, and industry-specific expertise.

“Those durable capabilities are what you sell.”

Gartner also sees an opportunity for companies that build platforms and services supporting workflows across multiple business domains. The challenge is that AI companies are moving quickly and could expand into services traditionally controlled by enterprise software providers.

Workday bets on trust and workflows

Workday is positioning trust, privacy, and security as defenses against disintermediation. Clare Hickie, Workday’s CTO for EMEA, told ZDNET that many business leaders remain cautious about adopting AI because they need confidence in how systems handle sensitive information.

“What’s crucial to recognize is that privacy by design and security frameworks are built into absolutely everything that we do.”

At Workday’s innovation media event at its EMEA headquarters in Dublin, executives outlined a roadmap for integrating agents through the company’s next-generation service, Sana. The goal is to create a “front door to work,” where employees can ask natural-language questions—such as why payroll has changed—and receive personalized answers from enterprise data sources.

Hickie said Workday’s lasting value will come from providing a trusted platform for cross-business workflows, particularly for risk-averse organizations that need reliable controls around automated actions.

Freshworks is taking a similar position. CTO Murali Swaminathan said its Freddy agentic technology will sit within an open customer-support platform, while operational data remains in a system of record such as Freshworks.

“These AI specialists don’t want to build everything. They want to be that layer where you engage, but the underlying layer will still be systems like ours.”

Snowflake points to the cloud era

Snowflake co-founder and president Benoît Dageville said the company plans to remain relevant by focusing on its trusted data platform. His argument is direct: “There is no AI without data.”

Dageville compared today’s concerns with earlier fears that cloud providers such as Amazon would displace Snowflake after Amazon launched its Redshift data warehouse. Instead, Snowflake built a strong relationship with Amazon Web Services, using AWS infrastructure while serving AWS customers.

He suggested AI companies may ultimately favor partnerships and platform relationships rather than rebuilding every specialized enterprise system themselves. Anthropic or OpenAI could become Snowflake competitors, he said, but their advantage in AI models does not automatically give them the expertise needed to reproduce Snowflake’s technology.