This Week in Agents: The SaaSpocalypse, Decoded

A name went around Wall Street this year: the SaaSpocalypse. The pitch behind it is that AI agents are about to gut the software business, because why pay for ten apps when one agent can do the work across all of them? Software stocks fell hard on the idea. So here’s the honest read, because the headline gets it half right: the software isn’t dying. The way it’s sold is. And the agents doing the killing still need a person watching the parts that matter.

So what set off the panic?

The term surfaced in February 2026, reportedly from a trader at the bank Jefferies, and it stuck. Software stocks shed something on the order of a trillion dollars in market value over the first stretch of the year (that number is market chatter, not an audited tally, so treat it as directional). The fear underneath it is specific. Most business software is billed per seat: you pay a monthly fee for every human who logs in. Move the work to an agent and the human seat count drops. When the seat count drops, so does the bill. The whole model assumes people opening the app, and agents don’t open apps.

Where did the idea even come from?

Not from this year’s selloff. It came from Satya Nadella, who runs Microsoft, on a podcast back in December 2024. He said something that sounded technical and turned out to be the whole thesis. Most business apps, he argued, are “CRUD databases with a bunch of business logic.”

Unpack that. CRUD is just the boring plumbing every app does: create a record, read it, change it, delete it. Store a customer, update their address, close their ticket. The “business logic” on top is the rules about what happens when, and that’s the part you were really paying for. Nadella’s call: that logic “is all going to these agents.” The agent holds the rules, reaches into whatever database it needs, and the app you used to open every morning quietly turns into plumbing the agent talks to. You stop paying for the front door because you stopped using it.

Is per-seat pricing actually breaking? Watch what Salesforce did.

Forget the predictions and look at what the biggest sales-software company on earth did to its own price tag. When Salesforce shipped its agent product, Agentforce, it didn’t charge per seat. It charged $2 per conversation. Then, in May 2025, it moved again, to credits you burn by the action, around 10 cents an action. Update a record, resolve a case: each one draws down a balance, the way a utility meter ticks.

That’s the tell. A company that built a $300-billion empire on per-seat licenses just hedged the model on its newest product, because an agent doesn’t buy a seat. It does a thousand small actions. You can’t bill a piece of software the way you bill an employee, so the meter is replacing the seat. When the incumbent starts pricing against its own moat, the SaaSpocalypse crowd has a point.

Then why isn’t SaaS just dead?

Because the loudest experiment in this whole story already ran, and it cut both ways. Klarna, the buy-now-pay-later company, put an AI assistant on customer support in early 2024. The numbers were real and they were big: in its first month it handled two-thirds of all service chats, about 2.3 million conversations, work the company pegged at 700 full-time agents. Resolution time dropped from 11 minutes to under 2. Klarna put the profit impact near $40 million for the year. If you wanted proof that an agent eats a software-and-staffing line item whole, this was it.

Then came the part the SaaSpocalypse headlines skip. By May 2025 Klarna’s CEO, Sebastian Siemiatkowski, admitted they’d cut too far. His words: pushing that hard on cost gave you “lower quality.” Customers hit a wall on anything the bot couldn’t handle and hated it. So Klarna started rebuilding, this time around a rule that a customer can always reach a human. The same company that became the poster child for replacement quietly became the poster child for the limit.

The cheap part was the volume. The expensive part was the judgment. Klarna automated the first and learned, in public, that the second still needs a person on it.

So what do you actually do with this?

If you run a department, here’s the decode. The thing dying is per-seat pricing for software that doesn’t do much beyond store and fetch your data. Bet on that going away. The thing not dying is the work, the data, and the human sign-off on anything you can’t take back. An agent that handles the two-thirds Klarna handled is a genuine bargain. An agent you trust with the last third, alone, is how you end up explaining to customers why the bot told them something wrong with total confidence.

That last point is the whole reason Natively builds the way it does. Our use cases do the volume; a person approves the calls that bite. The companies winning right now aren’t the ones who fired everyone and bought an agent, and they aren’t the ones still paying per seat for a glorified spreadsheet. They’re the ones who figured out which two-thirds to hand over and which third to keep a human on. For the longer version of that argument, see why the best agents come with clear limits, and why most AI rollouts fail for people reasons, not technical ones.

The SaaSpocalypse is real in the way most market panics are real: it’s right about the direction and wrong about the speed and the body count. Software didn’t end. It got a new price tag and a human still standing at the door.

This Week in Agents is Natively’s running read on the AI-business news that actually matters, with the numbers checked.