The Meter Is Dying: A Short History of Paying Per Seat

Billing by the CPU-second

In the 1960s a computer cost more than the building it sat in, so almost nobody bought one. You rented time. Time-sharing bureaus sold the machine by the slice, and the invoice read like a utility bill: CPU-seconds consumed, connect-minutes at the terminal, kilobytes stored, pages printed.

The folklore of that era is all about the meter. Students got semester allowances of compute and blew them by Halloween. Programmers learned to run big jobs after midnight because off-peak rates were cheaper. An entire coding aesthetic — terse, batch-minded, allergic to wasted cycles — was shaped by a billing model. We can't vouch for every 3 a.m. war story, and we won't try; that's folklore doing what folklore does.

But the meter itself was real, and it deserves respect: it is the most honest meter software has ever had. The unit on the invoice was literally the scarce resource being consumed. Nobody has quite managed that since.

The license discovers the CPU

On June 23, 1969, under antitrust pressure, IBM unbundled software from hardware. Programs and services would be priced separately from the machine. Software became a product overnight — and a product needs a unit.

The unit the industry settled on was your hardware. Per-CPU licenses ruled for decades; when multicore chips arrived, per-CPU quietly became per-core, and per-core sprouted fractional multiplier tables, so licensing a database meant doing arithmetic on your silicon. VMware once tried metering allocated memory; customers immediately named it the “vTax,” and it did not survive the naming.

Notice the drift. The CPU-second had measured cost. The core count measured something else: how big your company was, approximated by how much hardware it owned. The meter had stopped tracking the vendor's expenses and started tracking the customer's ability to pay. Few complained, because the hardware was at least still visible — you could point at the thing you were being billed for.

The seat: a proxy for a proxy

SaaS made the hardware invisible, which broke the old unit. If the customer never sees a server, you can't bill them by the core. So the industry reached for the one thing the customer could still count: people.

Per-seat pricing was a perfectly reasonable idea in 2005. One login meant one human doing work in the product, and more humans meant more value delivered. The seat was a proxy for value, the way the core had been a proxy for company size, which had been a proxy for the machine time that once genuinely cost money. A proxy for a proxy for a proxy.

Then it curdled. Seats for people who glance at a dashboard once a quarter. Seats for the accountant. “Viewer” seats, priced as though reading were a workload. On a multi-tenant web app, the marginal cost of the twelfth login rounds to zero, and everyone selling seats knows it. The seat stopped being a proxy for value and became what it had quietly always been: a sales-expansion strategy wearing a pricing model's clothes.

Meters that meter nothing

Our corner of the industry — websites and CMSes — builds some of the least honest meters going. Per-locale pricing. Per-language add-ons. Item caps. A language variant of a page is a handful of rows in a database; its marginal cost is storage, measured in kilobytes. Charging per locale isn't cost recovery. It's margin, painted as necessity.

We've written before about why StoryPress doesn't count languages. The short version: when the real cost of a thing rounds to zero, a cap on it isn't a meter. It's a tollbooth.

The traditional escape from metered SaaS was self-hosting — WordPress, a VPS, freedom. That trade is real, but you pay in a different currency: plugin patching, update roulette, and the general security babysitting that comes with owning the whole stack. Swapping a dishonest meter for an unpaid part-time job is not obviously a win.

2026: the seat dies in public

Which brings us back to July 12, 2026. Netlify's announcement is worth reading in full, because it's rare to watch a company shoot its own pricing model and then publish the autopsy. Pro used to be $20 a month plus $20 for every additional seat; now it's $20 a month, unlimited seats, with the meters moved onto usage — bandwidth, compute, requests. The reasoning is the interesting part: “Seat pricing made sense for us when software was written by a handful of developers. It makes far less sense when anyone, including any AI agent, can build software.”

That second sentence is the whole 2026 story. AI agents now do work that used to require a human with a login. An agent doesn't have a chair. It doesn't attend standup. Billing it as a seat is obviously absurd — and once it's absurd for agents, it gets awkward for humans too, because suddenly everyone can see what the seat always was.

There's a second-order effect worth naming, too. When a vendor's revenue depends on seat expansion, the product grows in seat-expanding directions: approval chains, role hierarchies, features that exist mainly so more people need logins. A pricing model isn't just how a company charges — it's how the company decides what to build. A usage-priced product has to get faster and cheaper to serve, because the vendor's margin depends on it. A seat-priced product gets more people-shaped, whether or not the work is.

The analysts agree, for whatever that's worth. IDC projects that by 2028, “pure seat-based pricing will be obsolete, with 70% of software vendors refactoring their pricing strategies around new value metrics.” That's a projection, not a measurement — analyst predictions about pricing have a long history of arriving early, late, or not at all — but the direction matches what vendors are actually doing to their pricing pages.

Transparency itself is becoming a moat. PostHog publishes its pricing principles in a public handbook — it even lists its sales discounts publicly — and has built a devoted developer following partly on that honesty. When your pricing page can survive being read aloud on stage at a conference, that's a competitive advantage. Most can't.

Honest meters

Here's the distinction that survives all six decades of this story: some meters track a real marginal cost, and some track your willingness to pay.

• Honest meters: bandwidth, compute, storage. When a visitor loads your page, electrons genuinely move and somebody genuinely pays. Metering that is fair — as long as the cap is printed in plain type on the pricing page, not in footnote nine.

• Artificial meters: seats, locales, languages, “items.” The marginal cost rounds to zero. The meter exists because the vendor's growth plan requires it to exist.

StoryPress takes a side. We're $5 a month, billed $60 yearly. We meter the things that cost real money — bandwidth and storage — with plainly labeled, sensible caps. We refuse to meter the things that don't: no per-seat pricing, no per-language pricing, and our native CMS is arriving with first-class language support and no cap on language variant count. To be clear, we are not promising “unlimited everything” — nobody who has ever paid a bandwidth bill can promise that with a straight face. We're promising something rarer: every meter on your invoice corresponds to a resource that actually exists.

The same logic decides what we build. A vendor paid in real resources has one incentive: keep the site fast, small, and cheap to serve. That's the whole plan — a website you won't outgrow and never have to babysit, with no plugin roulette and no seat matrix to renegotiate at renewal.

The short version

• June 23, 1969: IBM unbundles software from hardware; software becomes a separately priced product, and the industry starts arguing about units.

• 1970s–1990s: per-CPU licensing, later per-core. The meter tracks customer size, not vendor cost.

• 2000s–2020s: SaaS standardizes per-seat pricing. The marginal cost of a seat on multi-tenant infrastructure rounds to zero.

• July 12, 2026: Netlify ends seat-based pricing — Pro becomes $20/month with unlimited seats, and usage meters carry the load.

• December 2025: IDC projects that by 2028 pure seat-based pricing will be obsolete, with 70% of vendors refactoring pricing around new value metrics. A projection, and attributed as one.

• An honest meter tracks real marginal cost (bandwidth, compute, storage). An artificial meter (seats, locales, languages) tracks willingness to pay.

• StoryPress: $5/month billed $60 yearly, meters on bandwidth and storage with plain caps, and no meters on languages or seats.

If you'd like a website priced in resources instead of proxies, get started — it takes minutes and involves no sales call. For the deeper cuts, read why we don't count languages and what self-hosting's “free” actually costs in patching.