SaaStr Churned Notion After 7 Years — Not Because It Failed, But Because Its AI Agents Never Reached For It

Jason Lemkin describes SaaStr canceling Notion after seven years as a customer — a subscription with zero support tickets, zero feature requests, and zero escalations the entire time, strong enough that a SaaStr teammate had once been invited to present at Notion’s own user conference. His point isn’t that Notion did anything wrong: SaaStr’s AI agents, which now run the company’s staff meetings, simply have no use for a wiki or document tool built for humans to read. Lemkin’s sharper claim, made in We Churned Notion After 7 Years, is that no customer health-scoring system on the market would have caught this churn coming, because the signals of disengagement look nothing like a traditional usage drop-off — the product didn’t get worse, it just stopped having a job to do. He frames it as a new churn category: happy, non-complaining customers who can silently vanish once an agent absorbs the task the software used to perform.

This isn’t a one-off anecdote from Lemkin — it’s the same pattern he described a month earlier running his own company. In Claude Became Our AI VP of Product, Lemkin detailed SaaStr operating as an 8-figure business with three humans and 21+ agents in production, revenue up 140% year over year. The headline example: Replit quietly shipped an unannounced MCP beta that let Claude Opus operate as an “AI VP of Product” on top of Replit — retaining its own context, negotiating directly with Replit’s build agent over MCP, and catching it when it prematurely declared a broken feature “working as specced.” The same post’s anecdotes — moving “10 years off Marketo” for $14, an agent “killing” a $10K app in an hour — are the same cost-collapse logic that just took out Notion.

For B2B software vendors and sales leaders, the two posts together are a warning that a customer’s health score can look perfect right up until an internal agent absorbs the job the product used to do — and that risk isn’t hypothetical at companies already running agents this deep into daily operations.