An AI Agent Just Fired a SaaS Vendor Nobody Human Would Have Dropped

A decade-long vendor relationship is exactly the kind that survives bad service, because switching costs and human inertia both favor staying. SaaStr’s Jason Lemkin describes how his team’s AI agent ended one anyway — not because a person finally got fed up, but because the agent hit a wall and asked a question no human on the team had gotten around to asking.

What Happened

Lemkin’s team moved more than ten years of data from Marketo to Salesforce Marketing Cloud after an AI agent repeatedly hit Marketo’s API limits — roughly one usable hour of API access per day — which made basic analytics work impossible. The agent surfaced the failure, and when asked for guidance, it recommended leaving and named the alternative. The migration itself took one week and cost approximately $14 in agent time. What pushed the decision wasn’t only the API ceiling: support was, in Lemkin’s words, “the worst of any vendor we work with,” pricing had climbed for five consecutive years culminating in a 12% renewal increase, and Marketo never offered to negotiate or fix the underlying problem — even though the team would have paid $20,000 to $25,000 to stay if the vendor had simply asked.

Why It Matters

Lemkin’s argument is that agents remove the human buffer that used to protect underperforming vendors. A person managing a vendor relationship weighs switching friction, relationship equity, and the hassle of migration before recommending a change — an agent doesn’t carry any of that weight. “An agent hits your limits. It surfaces the failure,” with no loyalty to the incumbent and no patience for a fix that isn’t coming. That turns a technical limitation — an API built for a handful of scheduled syncs per day rather than an agent making 30-plus queries — into a direct churn signal, compressed into what Lemkin calls “a blocked agent and one question,” with no RFP, no committee, and no multi-quarter evaluation cycle standing between the friction and the decision to leave.

What This Means for Builders

For SaaS vendors, the takeaway is structural, not cosmetic: API rate limits and support quality that were merely annoying under human-paced usage become existential under agent-paced usage, because agents don’t tolerate friction the way relationship-invested humans do. For operators evaluating their own stack, the same logic cuts the other way — an agent auditing your vendor list will surface the same blocked-and-frustrated signals Lemkin’s did, and it’s worth running that audit deliberately rather than waiting for an agent to hit the wall on its own. Lemkin’s broader framing is worth internalizing either way: “assume every customer is mentally on a one-year deal,” because AI-accelerated workflows compress the timeline on which switching costs actually get evaluated.

Source: SaaStr

Content created with AI assistance and reviewed for accuracy.