The seat is dying: SaaS pricing when software gets cheap to build

Opinion: the following is an argument, not a build log. We build FastYoke, so we are not a neutral party — but we've tried to cite the parts of this that aren't just our own take, and to be honest about the parts of the story that are messier than the pull-quotes suggest.

Renting someone else's CRUD database

Here is what per-seat SaaS actually is, stripped of the marketing: you pay a monthly fee, per human, for access to somebody else's database schema plus the business logic wrapped around it. The CRM is a contacts table and a pipeline-stage state machine. The ticketing tool is a tickets table and an SLA timer. The expense system is a line-items table and an approval chain. None of that is magic — it's data modeling and workflow logic, packaged and rented back to you by the seat, because writing it yourself used to cost more than renting it did.

That calculus held for two decades because building custom software was genuinely expensive: engineers, time, maintenance. Renting was rational. But the thing that made building expensive — writing and maintaining the CRUD-plus-logic layer — is exactly the layer AI is getting good at generating. When the expensive part gets cheap, the rational choice changes, and the pricing model built on top of the old calculus starts to crack.

That's the thesis. Three questions worth asking before you believe it: is software actually getting cheaper to build, what are companies doing about it in practice, and — since this is our blog — what does FastYoke think the answer looks like.

Is enterprise software actually shifting cheaper?

The most useful data point here isn't a startup's pitch deck — it's the CEO of the company that sells more enterprise SaaS than almost anyone else on earth saying the model underneath his own products is exposed.

Satya Nadella has said publicly that SaaS applications, in their current form, "will collapse" in an agent-driven era, because most of them are, in his words, CRUD databases wrapped in a layer of business logic — and that logic is precisely what an AI agent tier can now absorb (Office Chai, Windows Central). That's a striking thing for the head of Microsoft — a company with an enormous SaaS and productivity-suite business of its own — to say out loud. It's not a competitor taking a shot; it's an insider naming the mechanism.

If the logic tier is what's collapsing, the pricing tier attached to it should be moving too, and the analyst data backs that up. Gartner's research (as summarized by GetMonetizely's 2026 pricing guide) and Deloitte's own technology predictions for 2026 both describe the same trend from different angles: seat-based pricing is losing share to usage-based, outcome-based, and agent-based pricing models, as vendors and buyers alike look for a unit of value that still means something once an "agent" — not a logged-in human — is the one doing the work. A per-seat price implicitly assumes a human is sitting in the seat. Once the work is done by an agent that runs continuously and never logs in, the seat stops being a meaningful unit to charge against.

You can already see the outcome-based model shipping in production, not just in slide decks. Zendesk prices its AI agent by the automated resolution — you pay when the bot actually closes a ticket, not for a seat that might sit idle. Intercom's Fin support agent runs on the same logic: a resolution fee only when the conversation actually resolves, nothing if it doesn't. Both are quiet admissions that the seat was never the thing customers wanted to buy — the outcome was, and now that outcome can be priced directly.

What are companies actually doing about it?

The build-vs-buy pendulum swinging back toward "build" is the other half of this story, and the most-cited example is Klarna's very public move away from Salesforce and Workday, replacing large parts of that stack with AI-assisted, more custom tooling (Inc.).

But the honest version of the Klarna story is messier than the headline, and it's worth saying so plainly: Klarna did not simply fire its SaaS vendors and stand up an in-house replacement built entirely by its own AI agents. Reporting on what actually happened describes something closer to consolidation — moving to lighter-weight or alternative SaaS tools alongside internally-built and AI-assisted pieces, not a wholesale build-everything-ourselves rebuild (CX Today). The industry was, reasonably, skeptical of the cleaner narrative. The real lesson isn't "one company proved AI replaces all of SaaS overnight." It's that a company with real leverage and real engineering capacity looked at its rent bill, decided some of that logic was cheap enough to own instead, and made the trade for the pieces where it made sense — while still buying software where buying still made sense.

That's a more useful signal than the maximalist version, because it's also what we'd expect a rational buyer to do next: not abandon SaaS wholesale, but get more selective about which logic is worth owning. Meanwhile the vendors on the other side of that trade are visibly repricing — moving from flat per-seat toward usage- and outcome-based models — because a pure seat price is a hard sell to a customer who just watched a competitor cut its logic-layer costs to build cost plus a fraction of the old license fee.

Where FastYoke fits: ownership beats rent, when building gets cheap

We didn't build FastYoke on the bet that "AI kills SaaS." We think that framing is mostly hype, and a lot of the loudest version of this argument overstates how fast or how completely it happens. Software categories with genuine defensibility — deep vertical expertise, network effects, regulatory moats — don't evaporate because generation got cheaper.

What we do believe, and what we built toward, is Nadella's narrower claim: the CRUD-plus-logic layer that sits underneath most operational software is becoming cheap to generate. If that's true, the winning move isn't to rent that layer forever at a per-seat markup — it's to own it. That's the actual bet FastYoke is built on: a platform where you build, and increasingly AI-generate, your own operational apps — the workflows, the approval chains, the pricing rules — and keep them, rather than renting someone else's version of the same logic indefinitely.

Concretely, that shows up in a few places we've written about separately: a Rust core built for the kind of long-lived reliability an owned system needs to earn trust over years, not quarters; a WebAssembly sandbox so the logic you generate or write runs safely no matter who wrote it or where it runs; two database options so "your data" can mean a laptop, a single VM, or a distributed cluster depending on what you actually need; and a sovereign Git vault so the record of what your business did is a set of files you hold, not a row in someone else's multi-tenant database that evaporates the day you cancel.

And on pricing specifically: we don't sell pure per-seat lock-in. FastYoke is priced on tiers plus metered usage, because once agents and generated logic are doing meaningful chunks of the work, a head-count-based price is measuring the wrong thing — same diagnosis Gartner and Deloitte are describing industry-wide, applied to how we charge for our own platform.

The real shift

The honest framing isn't "AI kills SaaS." It's narrower and, we think, more durable: when the logic layer gets cheap to build, ownership starts beating rent, for the parts of your stack where owning is worth the tradeoff. Klarna didn't torch its entire vendor list overnight, and neither will most companies — but the direction of travel, from the platform vendor's own CEO down through the analyst research to the pricing pages of the tools shipping today, points the same way.

If that's the world you're planning for, take a look at FastYoke's pricing or what we're building — the platform is built for the version of this where you own the logic instead of renting it back every month.