Forty gigabytes spread across roughly 21 million records is about 1.9KB a row [17], which is a fair description of what agents actually deposit into a CRM: task records, email sends, opens, clicks, call metadata, enrichment [24]. It's exhaust, not a decision. On Neon or Supabase or Databricks, in the author's framing, 40GB is nothing [25], and when SaaStr asked Claude what 21 million records would cost on Postgres the estimate came back around a thousand times cheaper, which the author is careful to label an agent's estimate rather than a quote [9].
That gap measures price, not value, and the firm's own reasoning explains why it hasn't moved on it despite that: ten years of hard data that does not drift and does not hallucinate, with Artisan and Qualified integrating natively, and an explicit willingness to pay a few nickels more for that [16]. The engineering hours went somewhere else entirely, into an agent called 10K that the author now describes as headless Salesforce, with classic CRM, Qualified, Momentum, Agentforce, Marketing Cloud Next and a custom quote-to-cash app reaching PandaDoc, Bill.com and QuickBooks all running through it [14]. Its builder had never logged into Salesforce in a decade before building it [15]. The real commercial question here is where the next 100GB lands, since moving ten years of history was never really on the table [13].
The platform side of the same trade has sharper arithmetic. Thirty days of notice across about 1,000 shared customers is roughly 33 joint integrations a day to re-plumb [20], and 30 days is about 0.9% of the nine-year relationship being ended [21].
The counter-thesis, which I think is the stronger one on a two-year view, runs like this. First, a thousand-fold gap on a storage line that is still small against a decade of accumulated switching cost is a rounding error dressed up as an argument, and a vendor that reprices machine writes into a cold tier makes it disappear. Second, ServiceTitan cut off a partner whose agents had crossed nine figures of revenue and begun holding the job record themselves [6], while smaller competitors stayed on the platform [8]. That reads less like a doctrine about categories than a doctrine about scale: one that only binds partners large enough to be worth enforcing against.
What would prove the storage thesis wrong is simple enough to check. If the next 100GB lands in Salesforce anyway, because native integrations with the agent vendors keep winning [16], then this was three humans and 20-plus production agents [1] discovering that a meter exists, and nothing more.
