SaaS isn’t dying. It’s becoming infrastructure for AI agents

The loudest narrative says AI will kill SaaS. The way enterprise platforms are actually moving suggests something different: AI is moving on top of them.

For the past two years, one of the loudest narratives in technology has been simple: AI is going to kill SaaS.

The logic is easy to follow. If an AI agent can write code on the fly, why pay for dozens of separate applications? Why not have AI build exactly what you need, when you need it? Public markets certainly took the idea seriously, and software valuations felt the pressure.

Then we looked closely at the recent partnership between Salesforce and Anthropic, and a different picture came into focus. AI isn’t replacing enterprise software. It is moving on top of it.

How the workflow is changing

Consider how work actually flows through an organisation.

The old way: employee → SaaS application → action.

The new way: employee → AI agent → SaaS platform → action.

In the new model, a platform like Salesforce no longer needs to be the screen a salesperson clicks around in for four hours a day. Instead, its data, permissions, workflows and business logic become the trusted infrastructure that AI agents rely on to get work done.

That distinction matters. An agent that updates a forecast, qualifies a lead or files a support case still needs a reliable system of record. It needs to know who is allowed to see what. It needs the approval rules, the audit trail and the integrations that enterprises have spent years building and refining. Those things don’t disappear because the interface changes.

The real divide

This changes how we should think about the software market. The dividing line isn’t AI versus SaaS. It is between two kinds of SaaS:

  • SaaS that is merely an interface. If a product’s main value is a set of screens over someone else’s data, an agent can increasingly do that job instead. These products are at real risk.
  • SaaS that owns the underlying data, workflow and business logic. These platforms become more valuable as agents arrive, because every agent needs somewhere authoritative to read from and act on. They are highly defensible.

Possessed, not replaced

We may well see more SaaS platforms effectively “possessed” by Claude and similar models, with agents doing the clicking, querying and updating that people used to do. That may make the AI labs even bigger. But rather than calling an exorcist, we think it gives those platforms a new lease of life. The platform stops competing on its interface and starts competing on the quality of its data, its rules and its integrations.

For the organisations that run these platforms, the practical questions change too. Are our permissions and data models clean enough for an agent to use safely? Are our workflows exposed through APIs an agent can call? Is our business logic encoded in the platform, or does it live in people’s heads? This is where much of the engineering work in enterprise platforms is heading: making existing systems ready to be operated by agents as well as by people.

Act II

The last two years felt gloomy for anyone building or investing in software. But watching how enterprise platforms are actually responding, this feels less like the end of SaaS and much more like its second act.

We’re curious to hear from founders and builders: are AI agents bypassing traditional software stacks in your world, or are they increasing your dependence on core platforms?

Have something worth engineering?

Whether you’re planning a product, modernising a platform or exploring a partnership, it starts with a conversation.

Let’s Talk