Jason Lemkin: AI Won't Kill SaaS, But It Will Kill the Vendor Who Stopped Shipping

Jason Lemkin’s claim is precise, not alarmist: the old 90/10 buy-vs-build rule — buy 90% of the time — still holds for complex, integrated platforms with data moats and workflow lock-in, but it no longer protects simple, single-purpose tools, per his SaaStr post. His proof is his own P&L. SaaStr replaced a newsletter vendor it paid $4,000 a year for — one that had “shipped nothing new material in 5+ years” — with a custom AI-built alternative assembled in about two hours, auto-populating content, summarizing posts, and pulling social feeds via API, saving their Chief AI Officer over 90 minutes a week. He calls the pattern “stealth churn”: usage metrics stay flat right up until the customer cancels at renewal, because they quietly built the replacement in-house months earlier.

Lemkin made the complementary argument himself two weeks earlier, and it’s worth reading the two pieces together. In “3 Humans, 20 AI Agents” he describes SaaStr’s own operations — 3 human employees, 20+ agents in production, 47% year-over-year revenue growth, over $1 million in agent-attributed revenue — and argues traditional B2B software still isn’t obsolete: humans need human-friendly interfaces to intervene, multiple agents need shared logic and rules, and compliance needs a canonical system of record. Point solutions don’t get that protection. A single-purpose newsletter tool has no compliance moat and no system-of-record role — which is exactly why it was the one that got replaced in two hours.

a16z’s Moses Sternstein found the same split showing up in public markets the week before: software multiples are being repriced not on current revenue growth — he found essentially no correlation between growth rate and recent stock performance — but on perceived long-term durability, with the bottom quartile of the software index dragging down an otherwise-resilient market. The market is pricing exactly the distinction Lemkin is describing at the vendor level.

The Ceiling Method read: for sellers working point solutions, “stable and reliable” is now a liability pitch, not an asset one — Lemkin says it outright. If your roadmap hasn’t shipped anything material in years, the deal you’re most likely to lose isn’t to a competitor, it’s to your own customer’s AI engineer, and you won’t see it coming until the renewal doesn’t happen.