SaaStr replaced sales team with AI agents, then learned monitoring matters
Jason Lemkin's SaaStr deployed around 20 AI agents to handle sales outreach, supervised by a small team including Chief AI Officer Amelia Lerutte. Then a PR firm, acting on behalf of a hot AI startup, sent SaaStr a blunt email: never contact their client again. The executive had supposedly had a terrible experience.
Lemkin and Lerutte were confused. They thought the engagement went well. Weeks later, the executive reached out directly. He had no idea what his PR firm had sent. He loved working with SaaStr and wanted to do more. The PR firm had been fired.
The incident highlights a gap most sales orgs are ignoring: who is actually monitoring what your agents say? Not the template you approved. What they are saying right now.
SaaStr has seen it on both sides. One of their AI SDRs invited a prospect to meet "next week" at an event happening that week. Another vendor's AI agent pitched SaaStr a product they already paid for. A human would catch both mistakes in 30 seconds. No human was watching.
The monitoring problem applies to AI SDRs, human BDRs, and third-party agencies. Every outbound email, every call, every LinkedIn message is your brand. The prospect does not distinguish between a rogue agent and your intentional strategy.
Lemkin's point: if an agent, human or AI, is speaking for your company, you are responsible for what they say. That means auditing actual emails going out, listening to calls, checking sequences. Not once. Daily.
SaaStr's broader AI experiment required daily training, quality control, and a "truth layer" anchored in Salesforce to keep the agents accurate. The lesson applies to any team deploying AI for outbound, prospecting, or lead response: automation at scale requires monitoring at scale.
For sales leaders evaluating AI SDR tools or agent-assisted outbound, the implication is clear: budget for the monitoring layer. The tooling exists: sales performance management platforms, CRM dashboards, conversation intelligence tools that flag when an agent goes off-script. The question is whether leadership is actually using them.
Most sales orgs are not. They deploy the agent, approve the sequence, then assume it runs clean. SaaStr got lucky: the executive came back. Most of the time, they do not. The relationship just dies, and you never know why.
Worth noting: this applies to human reps too. If your SDR team is running sequences without manager review, or your agency is speaking for you without approval, the risk is identical. The agent type does not matter. The oversight does.