The conversation around AI agents displacing enterprise software keeps getting framed as a binary: either SaaS survives or agents win. That framing flattens a distinction that actually matters.
When you look at how enterprises are deploying AI agents in 2025 and 2026, they are not replacing their systems of record - they are building orchestration layers on top of them. That sentence does a lot of work. It describes what is happening in production, as opposed to what is happening in pitch decks.
Here is the position worth defending: agents are not killing SaaS. They are making visible which SaaS was never doing real work.
The disruption is not uniform. Treating it as uniform leads to genuinely bad decisions, for investors and for the people building and buying software.
The thin tools are the ones at risk.
Tools that do one specific thing - automating a particular report, scheduling a specific kind of meeting, managing a particular support queue - are the most exposed. Gartner expects 35% of point-product SaaS tools to be replaced by AI agents or absorbed within larger ecosystems by 2030. When a general-purpose AI agent can perform the same task without a dedicated subscription, the standalone business case collapses.
That is not a prediction; it is already happening in measurable ways. Publicis Sapient reports actively reducing traditional SaaS licenses by approximately 50% - including major platforms like Adobe - by substituting them with generative AI tools and chatbots.
Thoughtworks eliminated three narrow SaaS platforms in 2025 alone, replacing them with bespoke AI workflows.
None of this is surprising when you look at what those tools were doing. A dedicated subscription for social scheduling, SEO reporting, or email sequence management is, in most implementations, a thin layer of UI over an API and a database. The moment an agent can read from and write to that same data, the subscription becomes overhead.
What survives - and why.
For use cases such as underwriting, a system that provides a correct answer six out of ten times is insufficient. These processes demand 100% consistency, which current probabilistic models struggle to guarantee without extensive re-engineering. The deterministic systems are not being disrupted; the operator is.
This is the distinction that the "SaaSpocalypse" coverage largely misses. Tools with deep data moats - years of transactional history, compliance audit trails, network effects from thousands of interconnected customers - are not going anywhere. Their value is not the interface. It is the data underneath it, and the trust that it is correct. When agents are the primary users, you don't need a beautiful UI. You need clean APIs, reliable data models, and agentic orchestration layers. The companies that have those things are in a completely different position from those that have spent the last decade perfecting dashboards.
A CRM or an ERP is not under threat from agents. It is becoming the backend that agents depend on. Rather than navigating multiple specialized applications, users interact with AI agents that orchestrate workflows across systems, generating insights, taking actions, and managing processes without requiring direct software interaction. The system of record still has to exist for the orchestration to mean anything.
The steelman for the "SaaS is over" camp.
The bull case for full disruption is real and worth taking seriously. When one user equipped with AI agents can accomplish the work of five traditional employees, the per-seat pricing model that has underpinned SaaS economics for two decades begins to collapse.
IDC predicts that by 2028, pure seat-based pricing will be obsolete, with 70% of software vendors refactoring their pricing strategies around new value metrics, such as consumption, outcomes, or organizational capability.
That pricing shift is structural, not cosmetic. If you are charging per seat and agents mean you need fewer seats, the revenue model breaks even if the software is valuable. The vendors who survive this transition are the ones who can pivot to outcome-based pricing before their growth rate makes the problem undeniable. Parts of, or even entire, enterprise applications could eventually be replaced by agents. Deloitte predicts this future may ultimately come to pass for some enterprise applications, but it will likely take at least five years or more to come to fruition, even with the rapid pace of technological development.
Five years is not "never." But it is enough time to make bad decisions in both directions - cancelling software that is actually load-bearing, or paying for software that an agent could replace this quarter.
The practical test.
The question to ask about any tool in your stack is not "can an agent do this?" The question is: what happens to the data if you cancel? If the answer is "we export a CSV and move on," that tool was a form with a monthly fee. If the answer is "we lose audit history, integrations, or network access we cannot replicate," that tool has a moat.
A teammate like Beagle sits inside Slack and Teams precisely because the context it needs - decisions, updates, handoffs - already lives there. It does not replace the systems those conversations reference. It connects them.
The agents-versus-SaaS debate will keep generating breathless analysis. Most of it is answering the wrong question. The right question is older and simpler: what is this tool actually doing, and for whom?
That question was worth asking before agents existed. It is just easier to avoid now that there is a culprit to blame.