Dora NodaFive developers, zero deploys, no traffic yet. On one platform that team already owes $100 a month; on three others it owes $20 to $25; on machines it owns, it owes nothing. The difference is not compute — nobody has run any. It is the seat meter, and in 2026 it fires before the compute meter on every metered platform still charging per head.
Here is the receipt, worked for the same five-person team on each platform's standard team plan, before a single gigabyte of bandwidth or build minute:
| Platform (plan) | Seat model | 5-person subscription floor |
|---|---|---|
| Vercel (Pro) | $20/seat/mo | $100/mo |
| Render (Pro) | $25 flat, unlimited members | $25/mo |
| Railway (Pro) | $20/workspace, unlimited seats | $20/mo |
| Netlify (Pro) | $20 flat, unlimited seats | $20/mo |
| Self-hosted fleet | No seats | $0 |
That table is the whole argument: three of the four platforms flattened their seat pricing in 2026, and the one that did not now charges a five-person team four to five times the subscription floor of the rest. The rest of this post shows how that happened, where the crossover points sit, why the seat line is about to get stranger now that AI agents commit code — and the one honest caveat that keeps this comparison fair.
April 2026 killed the platform seat (except on Vercel)
The flattening happened fast, and two of the three moves landed within nine days of each other.
Netlify moved first, on April 14. Its Pro plan dropped per-seat billing entirely and went to a flat $20 a month with unlimited Owners, Developers, and Git Contributors, funded instead by a shared monthly pool of usage credits that bandwidth, compute, deploys, and requests all draw down. The reversal went further than seats: Netlify had previously charged every repo committer as a full seat when CMS or Identity touched a private repo — bots and humans alike — and that per-committer charge was scrapped the same day. For a five-person team the arithmetic flipped overnight from roughly $100 in seat fees to $20 total.
Render followed on April 23. Its workspace restructure replaced the legacy per-seat lineup — Professional at $19 per member per month, Organization at $29 — with flat-rate Hobby ($0), Pro ($25), Scale ($499), and custom Enterprise, with unlimited members on every paid plan. Legacy workspaces were force-migrated on August 1. Render's own announcement put a number on the direction of travel: 75% of paying customers would see their costs decrease or stay the same. A five-person team that paid $95 a month in Professional seat fees now pays $25.
Railway flattened over the summer. Its Pro plan now reads as $20 per workspace with $20 of included usage and unlimited seats — third-party mirrors as recent as July still quoted the old per-seat wording, so the change landed after Render's. Either phrasing costs the same five-person team $20, but the model matters: headcount no longer moves the subscription.
That leaves Vercel Pro at $20 per seat per month, the last per-seat holdout of the four. Each seat bundles $20 of usage credit, viewer seats are free and unlimited, and the Hobby tier remains non-commercial — so a commercial five-person team starts at $100 a month before usage. There is nothing hidden about it; it is simply the only seat meter left in this particular lineup.
Per-seat models also leak in a way flat plans cannot: stale seats. Cost-optimization audits routinely find two or three ghost seats on a ten-person Vercel team — former members and rotated roles nobody removed — a quiet $40 to $60 a month for humans who no longer work there. Unlimited-seat plans make that entire category of waste impossible; there is nothing to audit when headcount is not metered.
The sensitivity sheet: 1, 5, and 10 people
One team size is a data point; three are a pattern. Here is the monthly subscription floor — seat fees plus base plan, before metered usage — at three headcounts:
| Team size | Vercel Pro | Render Pro | Railway Pro | Netlify Pro | Self-hosted |
|---|---|---|---|---|---|
| 1 person | $20 | $25 | $20 | $20 | $0 |
| 5 people | $100 | $25 | $20 | $20 | $0 |
| 10 people | $200 | $25 | $20 | $20 | $0 |
Two conclusions fall out immediately. First, the crossover is at two seats: a two-person team already pays $40 on Vercel against $20 to $25 everywhere else. Second, the solo developer is the one case where per-seat pricing still wins — $20 on Vercel ties Railway and Netlify and beats Render's $25. The seat meter is not a bad deal in general; it is a bad deal specifically for teams, which is exactly who pays it.
Now the honest caveat, because "before compute" needs it: not every dollar in that table evaporates. Vercel's $100 in seat fees arrives with $100 of usage credit the team can spend on bandwidth, builds, and functions. Railway's $20 includes $20 of usage. Netlify's $20 includes its monthly credit pool, and Render's $25 bundles 25 GB of bandwidth and 1,000 pipeline minutes. A team burning serious usage converts much of the floor into compute it would have bought anyway.
So the precise claim is narrower and still damning: the seat meter sets a headcount-scaling floor, not a pure tax. Add a sixth teammate who deploys twice a month and Vercel charges another $20 for the privilege of their existence; the other three platforms charge nothing for the headcount and meter only what that person actually runs.
When usage is lumpy or the team is growing faster than its traffic, the floor — not the meter — dominates the bill. That is the normal shape of an early-stage startup: hiring ahead of revenue, deploying often, serving little traffic. The seat meter taxes exactly that phase the hardest.
AI agents multiply the seat line
Here is where the seat model goes from expensive to incoherent. The industry spent 2026 discovering that per-seat pricing was built for a world where humans operate software — and AI agents do not sit in seats.
The numbers tell the story from both sides. Pure per-seat pricing dropped from 21% of SaaS companies to 15% in a single year, per Bessemer and Deloitte's TMT survey, and Bloomberg now forecasts subscription pricing falling from 60% to 30% of SaaS models within a decade. From the buyer side, Gartner reported AI coding bills leaping from $20–$100 to $2,000–$5,000 per developer per month as agent vendors shifted from seat-based licensing to consumption pricing, with extreme cases hitting $20,000 in token charges. The vendors already know the seat cannot hold the value of an agent; the platforms are catching up at different speeds.
Netlify's reversed per-committer charge is the concrete exhibit. When every identity that touched a private repo — including bots — counted as a billable seat, the arrival of an AI agent that commits daily would have multiplied the seat line by headcount-plus-agents. Netlify killed that logic in April. But the general question is still open on every platform that counts human-shaped identities: does an agent that opens pull requests, triggers previews, and deploys to production count as a seat? A team running one agent per developer has, for billing purposes, either five members or ten — and the platform's answer can double the subscription floor overnight.
This is the per-seat paradox in its sharpest form: automation that removes seats gets punished by pricing that counts them. A team that hires two agents instead of two engineers saves salary and grows its seat bill. Consumption pricing has its own pathologies — nobody enjoys a $5,000 token surprise — but at least it meters what the agent does rather than charging rent for the agent's existence. Expect every remaining seat meter in hosting to face this question within a year: the agent is already a daily committer, and the invoice has not decided what it is yet.
What to do with this if you are choosing this week
The decision rule is simple enough to fit on a sticky note:
- Solo commercial project: the seat meter is harmless. Vercel's $20 ties the flat plans and the usage credit softens it further. Pick on features, not seats.
- Two or more humans: flat and unlimited wins. The $75–$80 monthly gap at five people is $900–$960 a year of pure headcount rent — before compute, before overages, before growth.
- Any AI agents committing code: avoid anything that counts identities. Prefer platforms where the bill scales with usage, or a fleet where seats do not exist as a concept.
- Self-hosted floor: seats cost nothing on machines you own. The tradeoff is operational labor instead of subscription rent, which is a real cost — but it is a cost that does not multiply when you hire.
The deeper lesson is about leverage. A subscription floor that scales with headcount is a bill that grows when the team grows, regardless of whether the product does. 2026 was the year three platforms admitted that was a bad bargain and flattened it. The fourth has not — yet.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, with no seats to count and no per-head meter. Star the repo on GitHub or deploy your first app today.
