SaaStr Runs 20 AI Agents and 3 Humans — and Still Needs 'Real' B2B Software

SaaStr now runs on 3 human employees and more than 20 AI agents in production — and revenue still grew 47% year-over-year, with over $1 million in agent-generated revenue directly attributable to AI-driven workflows.

Despite that extreme ratio, Jason Lemkin argues traditional B2B and CRM software isn’t obsolete. He gives four reasons: humans still need human-friendly interfaces to review and intervene in agent work; multiple agents operating on the same business need shared logic and rules rather than each reinventing its own; organizations need a canonical system of record that external integrations and partners can reliably connect to; and compliance, auditing, and guardrail requirements are difficult to replicate in an ad hoc, self-built database compared to mature enterprise software. It’s a direct rebuttal to the “software is dead, agents replace SaaS” narrative circulating in venture and founder circles — agents and B2B infrastructure are complementary, not substitutes.

Lemkin’s own numbers from six days earlier back the ratio with real unit economics: SaaStr’s top AI agent, an AI VP of Marketing named “10K”, ran a one-hour, one-minute session logging 125 actions for $13.42 in compute — a figure no legal US wage undercuts, against a human VP of Marketing’s fully loaded cost of roughly $145 an hour. That’s the economics underneath a 20:3 ratio holding at 47% revenue growth. 11x’s playbook for scaling past initial deployment — volume, complexity, consistency, and agent-performance tuning — describes the same trajectory from the vendor side: teams that plateau after deploying agents versus ones that keep compounding gains, illustrated by Cofenster running 12-15 parallel campaigns against 86,000-plus prospects with a single operator.

The reps and orgs actually pulling ahead aren’t asking whether agents replace software or headcount. They’re running the ratio Lemkin just published and still keeping the system of record, the audit trail, and the human review step agents can’t replace. That combination — not agent count alone — is the real test of who’s actually ahead.