Can SaaS stocks survive AI?

For most of the last 15 years, the software business had a reliable trick.

Find a messy process that ran on spreadsheets and email. Turn it into a an SaaS app. Charge per person who logs in. Grow as the customer hires more people.

AI is breaking that trick.

This week’s software stock selloff wasn’t just “market drama.” It was investors reacting to a simple idea that’s getting harder to ignore. If AI can do the work, fewer humans need to touch the software. And if fewer humans touch the software, “pay per seat” stops compounding the way Wall Street expects.

Why the Anthropic moment rattled everyone?

The latest wave of anxiety was sparked by what companies like Anthropic are shipping, not just what they’re promising.

A chatbot that helps you write is an add-on. It makes existing tools nicer, but you still need the tool and the human.

What’s new is agent-style behavior. That means the AI doesn’t just answer questions. It can take steps. It can open a browser, click around, copy data from one place, paste it somewhere else, fill out forms, and finish a task that normally takes a person hopping between apps.

That difference sounds small. It isn’t.

Once an AI can operate software, it can also route around software. Many SaaS products are, at their core, a structured interface for moving information between systems and people. If an agent can do that movement, the interface becomes less important.

That’s the reason the market response got sharp. Investors weren’t pricing in “AI features.” They were pricing in the possibility that parts of today’s software catalog become optional.

Stephen McBride predicted this will happen back in 2024.

The simple math Wall Street is worried about.

Public SaaS companies are valued on a story about predictable growth. Two assumptions sit under that story.

One, pricing rises with headcount. If a customer has 1,000 employees using your tool and grows to 1,500, revenue grows with it.

Two, workflows get sticky. Once your tool becomes the place work happens, switching is painful. That creates long customer lifetimes.

AI puts pressure on both.

Start with headcount pricing. If an agent can do the work of several people, the customer’s rational move is to buy automation and cut seats. Even if the company keeps the same number of employees, fewer of them may need to log into a given tool every day. That hits the “expand forever” model right where it’s most sensitive.

Now stickiness. Agents can act like translators between systems. If the AI can pull data from System A, update System B, then summarize the result in System C, the customer becomes less locked into any single interface. They can keep the underlying data sources but swap out the front-end workflow tool more easily.

That’s why the fear language has been extreme. It’s not that software disappears. It’s that the unit economics of a big chunk of SaaS start to look different, fast.

What survives when AI is doing the clicking?

SaaS that’s mostly a UI for routine steps is most exposed.

Think tools where the user spends their day copying, pasting, updating fields, assigning tickets, or chasing approvals. An agent can do a lot of that.

SaaS that owns something harder to replace has a better shot. There are three defensible positions.

First, the system of record. This is where trusted, important data lives. Even if an AI is doing the work, it still needs an authoritative place to read from and write to.

Second, the rails. These are the pipes that move data and trigger actions reliably. If an agent is going to run real business processes, it needs integrations, permissions, and dependable execution.

Third, the governed execution layer. This is the “do it safely” stack. Access control, audit trails, policy enforcement, compliance reporting, and the ability to explain what happened after the fact. Enterprises won’t let a model run wild in sensitive workflows. They’ll demand guardrails, logs, and accountability.

This is where incumbents can win, but only if they accept a hard truth. The product isn’t the login screen anymore. The product is trusted data plus controlled outcomes.

It’s tempting to make this binary. “AI kills SaaS” or “this is overhyped.”

Neither is right.

AI adoption in big companies is still slower than the headlines. Models can make mistakes. Regulated workflows need approvals and traceability. Procurement takes time. That means there’s runway left for many SaaS businesses.

But the direction is clear. Customers are going to buy more results and fewer seats.

So the question to ask about any SaaS company isn’t “Do they have AI features?” It’s “What do they sell that still matters when the AI can do the steps?”

If the answer is mainly “a nicer way to move information around,” it’s in the danger zone. If the answer is “the trusted data, the rails, or the controls,” it has leverage.

One of the 5 Signs an Industry Is Ripe for Disruption is dissatisfied customers. We're seeing that principle in action today in SaaS.

If you run a SaaS business, start shifting pricing away from seats and toward measurable outcomes that you can verify and govern.