What do customers do when a vendor meters agent API calls?

Per Lemkin, they route around it. Agents read far more than they write, so the customer syncs the records to its own database, reads from the copy, and writes back to the vendor only when something changes. He calls it a weekend project, and says a bad answer on agent pricing is now disqualifying when SaaStr picks a new tool.

What should the cap look like?

Lemkin's better example is Atlassian's Rovo credits: overage starts on a published date (2026-12-03) at a published rate ($0.01 a credit, a basic action at 10 credits), every paid plan carries a pooled allowance of 25, 70, or 150 credits per user per month by tier, reads draw no credits today, and an admin can cap extra usage. Published rate, published date, allowance, switch.

Why is pricing the outcome the hardest of the three fixes?

Because the unit has to be defined, disputed, and defended before it can be invoiced. In the per-seat sunset I ran, the unit definition produced more than 200 contested cases a week at scale and disputes came in at 600 against a projection of 50. A per-outcome contract needs the unit definition, a dispute window, arbitration, a committed minimum, and a price ceiling in writing first.

What is the margin trough in a pricing migration?

Moving from seats to outcomes drops blended gross margin to roughly 58 to 65% around month 12, with recovery to 70% and above in months 18 to 30. In the sunset I ran, outcome revenue reached 91% and margin recovered to 71% by month 22. The board approves that curve before month one or the migration fails politically first.

Should a SaaS Vendor Charge per API Call for Agent Access

Should a SaaS vendor charge per API call for agent access?

THE SHORT ANSWER

Not as a meter stacked on top of the seat. Jason Lemkin's 2026-09-28 post puts Salesforce's agent meter at $5,000 to $100,000 per million calls against about $83 per million for the same call from an integration, and his customers' response is to sync the records to their own database and read from the copy. His three fixes for vendors, cap the rate, let a registered agent replace a seat, or price the outcome, are one migration in that order, not three options. The cap ships this quarter. The seat swap is the six-quarter migration with a margin trough near month 12. The outcome price is month 18, and in the sunset I ran it produced disputes at 12x projection. The stacked meter takes that trough, agents leave first and seats follow a renewal later, and nothing gets repriced.

Jason Lemkin's post is the buyer's side: almost every pre-AI vendor SaaStr uses now charges for agent access, Salesforce meters third-party agents per successful MCP or API call, HubSpot meters its own agents and leaves its MCP server free, and the customer's answer is a synced copy of the data. The vendor's side is in Kill the Per-Call Agent Meter: his three fixes are one migration in order, and the meter stacked on the seat is a substitute for that migration that takes the same trough.

The cost of doing it properly is in Field Report: What Broke When We Killed Our Per-Seat Tier, and the reason pricing the reads spends retention rather than defending it is in Kill the System-of-Record Slide. Both belong to the SaaS to AI Business Models argument.

Last reviewed 2026-09-29 · 1 min read

Should a SaaS vendor charge per API call for agent access?

SOURCES

THE LONG VERSION

RELATED ANSWERS