One Company's Salesforce Storage Grew 8x in 30 Days. Almost None of It Was Typed by a Human.

SaaStr’s own Salesforce instance grew from 5GB to roughly 40GB — about 21 million records — in about 30 days, according to Jason Lemkin’s latest dispatch on the 20-plus production AI agents now running across the business. Lemkin says that volume is trivial on Postgres, roughly 1,000x cheaper for the same storage, but expensive on Salesforce’s legacy per-record pricing, because nearly all the growth came from agent API writes rather than human data entry. He cites a starker example of the same dynamic playing out between vendors: ServiceTitan terminated its nine-year, roughly 1,000-customer integration with Podium after Podium’s agentic platform evolved from a lead-handoff partner into a direct competitor handling scheduling and job tracking end-to-end. On the build side, a SaaStr team member built a custom renewal agent in half a day, pulling from Salesforce, WordPress, social accounts, podcasts, and Bizzabo, that produced 20-30 sponsor renewal decks automatically — versus five manually — with stronger response rates from silver-tier sponsors than diamond-tier ones. Lemkin also flags real failure modes: vendors raising per-call API prices during 100x data growth can turn modest rate hikes into 200-2,000% cost increases, and the agent still occasionally hallucinated numbers despite explicit guardrails.

The accountability question this raises — who owns an agent now writing millions of records into the system of record — is what Aviso AI proposed answering a month earlier with its “RevOps Agent Owner” concept: a named role split between a Business Owner (RevOps or Sales, accountable for commercial outcomes) and a Technical Owner (SalesOps or IT, accountable for infrastructure and compliance). Aviso’s warning about a “governance vacuum” — no one formally responsible when an agent acts on stale context — reads less abstract next to Lemkin’s account of an agent hallucinating numbers inside a live renewal workflow.

For RevOps teams watching CRM data volumes climb without a matching rise in human data entry, the pattern is now a checklist: know which agent wrote which record, know what happens when vendor pricing doesn’t anticipate 8x growth, and know who’s accountable when the automation gets something wrong in a customer-facing deliverable.