AI Agents Are Putting the SaaS Model Under Pressure: What Gartner Sees and What to Do

In early July, Gartner published an analysis that landed hard in the software world: AI agents threaten up to $234 billion in SaaS spending through 2030, roughly 20 percent of the enterprise SaaS market. The cause now has a name: agentic arbitrage. Agents complete tasks across systems while bypassing the interfaces that vendors monetize.

We build both custom software and integrations with SaaS products, so we regularly sit right in this tension with our clients. In this article we explain what Gartner is actually saying, how much of it holds up in practice, and what it means for companies facing a software decision this year.

The core of the analysis, covered in the Netherlands by Dutch IT Channel, is that agentic systems deliver results directly. Where an employee used to log into a dashboard, set filters and export a report, they now ask an agent for the answer. The application still runs, but nobody looks at it anymore. Gartner analyst George Brocklehurst sums it up: the link between user growth and revenue growth is being broken.

That is where the redefinition of the term Saaspocalypse comes from: according to Gartner, less an apocalypse and more a metamorphosis. Software vendors that derive value from their interface and their per-seat price must switch to value based on outcomes, or watch agents reduce their product to a data layer.

The same Gartner research line predicts that by the end of 2026 about 40 percent of business applications will contain a task-specific AI agent, up from less than 5 percent in 2025. We see this reflected in the requests coming in: where clients asked for a dashboard a year ago, they increasingly ask for a system that performs actions itself, from preparing quotes to pushing stock mutations into the accounting system.

At the same time there is reason for sobriety, and Gartner provides it too: more than 40 percent of agent projects are expected to be cancelled before the end of 2027 due to rising costs, unclear value or poor risk controls. The technology is real, and so is the hype around it. The difference between the two almost always comes down to whether there is a concrete, measurable process underneath.

The practical translation for an SME is this: the value of your software stack is shifting from the screens to the data and the integrations. An agent that prepares your quotes has no use for a pretty CRM screen, but every use for a CRM with a good API and clean data. Systems that lock their data behind an interface without a decent API become the bottleneck of every automation you will want in the coming years.

For us as builders this is not a new insight. We have long worked from the principle that logic and data are central and the interface is a layer on top. It is also why we always push for abstraction layers in integrations: if your processes are exposed through APIs, you can connect every new generation of tooling without rebuilding. That was true for mobile apps, and it is true now for agents.

If you pay for ten, twenty or fifty seats of a SaaS product today, this is the moment to look critically at how many of those licenses exist because people occasionally need to look something up in a screen. Those are exactly the licenses an agent makes redundant. The question for your vendor is simple: do you have a full-featured API, and what is your pricing model when my agents do the work instead of my employees?

Vendors without a good answer will either price themselves out of the market in the coming years or overhaul their model. Both scenarios hit your budget. In our comparison between custom software and SaaS, user count was always a tipping point in the calculation; agents sharpen that calculation further, because custom software has no per-seat price.

Gartner points out that AI-native service providers can act as the agentic layer on top of existing business systems: they deliver measurable results instead of features. That is essentially what we do when we build a custom solution for a client that combines agents with existing systems: the accounting stays in Exact, the CRM stays where it is, but the repetitive work in between disappears.

A concrete example from our own practice: a client with a rental platform had requests reviewed manually, averaging twenty minutes per request. An agent workflow that reads the request, verifies data against external sources and stages a proposal brought that down to two minutes of human review. The surrounding SaaS systems stayed exactly the same; the value sat in the layer between them.

Gartner's analysis is no reason to cancel your SaaS subscriptions tomorrow, but it is a reason to look at your stack differently. Value is shifting from interfaces to data, APIs and outcomes. Systems without a decent API become dead weight, and per-user pricing models get shaky once agents take over the work.

Want to know where an agent layer would actually pay off in your processes, and where it is hype? Book a conversation. We are happy to think along soberly, even if the conclusion is that you should do nothing for now.