The Storage Bill Nobody Saw Coming
SaaStr went from 5GB to 40GB in Salesforce in roughly 30 days. That is 21 million records: task logs, email metadata, enrichment data, call tracking. Jason Lemkin and his team had not logged into the CRM in a week when the overage notice arrived.
The humans were not writing. The agents were.
Lemkin runs SaaStr on a skeleton crew: three people, 20+ AI agents in production, eight-figure revenue. The agents live in the CRM: writing as they go, updating records, tracking activity. What used to be a few dozen human clicks per day is now constant API writes. Storage costs built for human users do not scale when agents never stop.
Claude estimated the same 40GB would cost about 1,000x less on Postgres. Salesforce is not gouging anyone: the pricing model assumes humans, not agents writing millions of rows. The question is what happens when that assumption breaks.
When the Partner Becomes the System
ServiceTitan just cut off Podium after nine years and roughly 1,000 shared customers. Podium used to feed leads into ServiceTitan's trades management platform. Then Podium went agentic, crossed $100m in AI agent revenue, and the agent started holding the customer record.
ServiceTitan's position: you can compete with us, you can access records via API, but you cannot use the partnership to displace us. Podium got 30 days' notice. Smaller competitors are still integrated.
This is the real systems-of-record question for 2025. If an AI sales agent talks to your customer, schedules the meeting, tracks the deal, and writes into the CRM, who owns the record? The platform or the agent layer?
What This Means for Sales Ops
If you are running AI SDR agents, renewal automation, or AI-assisted outbound through Clay or similar tools, your CRM storage is about to move. Fast.
API pricing models built for occasional human data pulls do not work when agents query constantly. Raising API fees while data volume goes 10x is not a 20% price increase: it is 200% or more, and it will push teams to headless architectures or alternate storage.
Lemkin's setup is instructive: Salesforce still holds ten years of clean data, but the agent (10K, their AI VP of Revenue) sits on top of six systems at once: Salesforce, Qualified, Momentum, Agentforce, Marketing Cloud, and a custom quote-to-cash stack. The same architecture that makes the CRM 10x more useful makes it 10x easier to leave.
For sales leaders: budget for CRM storage growth if you are deploying AI agents for lead generation, data entry automation, or workflow tasks. For CRM vendors: the per-seat model is dead, but metering API calls while agent-generated data explodes will not survive either. Customers want pricing tied to outcomes, not a tax on reaching their own records.
Worth noting: ServiceTitan has roughly $961m in revenue and significant platform leverage. Podium is smaller. When systems-of-record vendors start policing AI agent usage at the API level, the power dynamic favours incumbents. For now.