SaaS isn’t dead. The eulogy is just easier to write than the harder question underneath it.
In February, the S&P North American Software Index dropped 15% in four days. Analysts called it the AI Substitution Effect. The argument was simple: agents execute work that humans used to do inside software interfaces. No humans clicking means no seats. No seats means no revenue. Ergo, SaaS is over.
That logic is half right. The seat-based model is in trouble. The category is not.
The mistake is treating this as a binary. SaaS dies or SaaS survives. The more useful frame is positional. Where does software sit relative to the agent? That question has a specific answer, and most companies are not asking it yet.
The Build Line Is Moving
Engineering teams at well-resourced companies are building internal agents that handle work third-party tools used to own. RCA. Ticket triage. Code review. Deployment workflows. The build-versus-buy line moves every quarter, and it is moving toward build.
This is not a fringe pattern. It is the leading edge of a structural shift. As agentic capabilities become more accessible, the cost of replicating surface-layer SaaS functionality inside a custom agent approaches zero. A product that lives at the execution layer has an existential problem. Agents execute. Execution is becoming commoditized.
The question for every SaaS company is not whether customers can build a version of your product. It is whether what you offer sits at a layer agents cannot easily replicate.
Where the Gap Actually Lives
Agents are exceptional at executing once they have a signal. The trigger, though, still waits for a human. Someone pastes an error into Slack. Someone flags a customer escalation. Someone notices the metric moved.
The agent wakes up reactive.
That gap is structural, not temporary. Practitioners on the floor at TDX said it plainly: agents can act quickly, code can be generated quickly, integrations can be deployed quickly. Organizational clarity, ownership, and data definitions do not evolve at the same pace. That is not a system failure. It is a confidence failure. And confidence failures are expensive.
It exists because correlation across systems, organizational memory, and structured context that predates any single workflow are not things agents generate on their own. They need that layer to exist before they can act on it.
Salesforce named it directly at TDX. Headless 360 is not a new interface. It is a decomposed system of context and work, built so agents can reach into the platform the same way a developer calls an endpoint. The platform vendor is building for the layer agents depend on. That signal is worth paying attention to.
This is where software still wins. Not at execution. At the starting line.
The companies that survive the agentic shift will be operating upstream of the agent. Providing the signal that tells it what matters, when it matters, and why. Context that arrives before the human has to notice something is wrong.
That is not a feature. That is a different layer of the stack entirely.
What This Means for Product Strategy
Three responses to the agentic floor are wrong. Competing on surface functionality agents can replicate. Retreating to a smaller feature wedge. Pretending the shift is overstated.
The right move is structural. Ask what your product does that an agent cannot generate for itself. Not what it displays. Not what it automates. What context, correlation, or organizational knowledge does it hold that the agent needs to do its job well?
IDC forecasts that by 2028, 70% of software vendors will have refactored their pricing away from seats toward consumption, outcomes, or organizational capability. That pricing shift reflects something deeper. The value has moved. It is no longer in the interface. It is in the signal the interface was always sitting on top of.
The companies that recognize this early will not build around what agents replace. They will build around what agents require.
The Seat Is Upstream
The agentic floor will keep rising. More workflows will be automated. More internal tools will be built. More of what used to require a SaaS subscription will be handled inside a single prompt.
That is not the end of software. It is the end of passive software.
What survives is the layer that feeds the system. Signal correlation. Structured handoffs between systems that were never designed to talk to each other. Context that compounds over time and cannot be reconstructed on demand.
The agent needs a starting line. That starting line is the product.
SaaS is not dead. The companies still selling seats to humans who click through dashboards are in serious trouble. The ones operating at the layer agents depend on are not.
Find the seat upstream of the agent. That is where this category is going.