Vercel published two changelog entries on 21 August that each attach a price to something previously unpriced. Deployment Storage — the build artefacts kept so a team can inspect and instantly roll back to a previous deployment without rebuilding — is now billed at $0.10 per GB per month. Separately, always-on tracing of real production and preview traffic entered beta at $0.50 per 1M span units.

What the common framing gets wrong

The storage post is written as a launch — "keeps your deployments rollback-ready" — but nothing new is being kept and rollback was already instant. What changed is that a free by-product of every build became a metered resource. Reporting it as a product launch reports a price increase as a feature. The sentence doing the most work is the reassurance: "Existing teams continue with their current pricing, with no change to their bill at this time." That is a deferral with a phrase-shaped hedge in it, not an exemption — existing customers are being told the meter exists and will reach them later. Hobby teams get 10 GB included; the changelog names no included allowance for anyone else.

"Always-on" is the one thing the tracing is not

The tracing feature samples according to rules the customer configures per environment, with optional path scoping. "Always-on" describes the collector being enabled, not every request being captured — and the gap between those two is precisely the incident you did not sample. It is also beta with a price tag, an unusual pairing: per-span billing for a product carrying no availability commitment. On the free tier, retention is one hour, so a developer who notices a problem the next morning has nothing left to look at.

Why agent-written code makes this bite

Every preview branch an autonomous coding agent opens leaves build artefacts behind, and agents open branches at a rate humans do not. Metering the residue of CI lands directly on the workflow that every dev-tool vendor — Vercel included — is currently pushing customers toward. A retention policy is now a cost control, not a housekeeping setting.

Two meters, one changelog day

Taken together the pattern is clearer than either post: observability and build residue are both moving from included overhead to per-unit spend. Whether other platforms follow is the thing to watch, because the underlying storage and span volumes are being driven up by the same agentic workflows the platforms are selling.