The Setup
SaaStr, the B2B SaaS media and events company, is running sales operations with roughly 3 humans and 20 AI agents in production. Founder Jason Lemkin published the breakdown after two people left the sales org and the work got replaced by automation.
This is not a proof of concept. These agents run live systems: Salesforce, billing, marketing campaigns, event logistics. They write code, fire vendors, and handle collections. They also break in ways that matter.
What the Agents Actually Do
10K (the AI VP of everything) owns the revenue number. Daily forecasting, campaign performance, newsletter to 450,000 contacts, LinkedIn and X ads end to end. When a contract closes in PandaDoc, 10K flips it to Closed Won in Salesforce, creates the invoice in bill.com, runs collections with 7-day escalation, calculates commissions.
Started as a dashboard in January 2025. Now sitting at 1,000 commits and 14,000+ lines of code.
What worked: Recommended a 15% ticket price cut that drove 40% attendance growth. Ran the Marketo to Salesforce Marketing Cloud migration for $14 in compute. Found two customers still being billed $300/month for a product SaaStr shut down six years ago. Fired vendors, including one his own research had shortlisted, after seeing premium pricing with no conversion data and multi-month minimums.
What did not: Five minutes before Lemkin went on stage, 10K sent 1,000 emails from a prohibited address that has been off-limits for years and written into core memory. When asked how, the agent said he forgot to read the memory. Human marketing managers make that mistake too. They just cannot make it 1,000 times before lunch.
Annie runs the SaaStr Annual website and attendee comms. Rebuilt on Replit in November 2025, now at 46,000 lines of code with the highest commits per day of any agent. Handles the agenda, newsletters, visitor behavior tracking.
The article cuts off mid-sentence on parking passes, but the pattern is clear: agents that start as dashboards become operational systems once humans keep showing up to work with them.
The Comp Angle Nobody Is Talking About
Two sales roles became zero sales roles. SaaStr is estimated at roughly $5M revenue (third-party data, treat cautiously). If those were standard enterprise AE roles at $150k OTE, that is $300k in comp off the books, plus recruitment, onboarding, and management overhead.
The agents cost compute, API calls, and developer time. Lemkin does not break out those numbers, but the ratio matters: this is workforce compression at the top of the funnel, and it is happening inside a company that sells to sales leaders.
What This Means for ANZ Sales Orgs
SaaStr has limited ANZ presence (no clear local headcount in public data), but the model applies. If a media company with a 450k database can run sales on 3 humans and 20 agents, the pressure is on every VP Sales with an SDR team, a marketing ops hire, and a rev ops analyst.
The agents are not replacing quota carriers. They are replacing the work around quota carriers: list hygiene, campaign ops, pipeline admin, collections follow-up, commission calc. That is the layer getting automated first.
Lemkin's team peaked near 30 agents and consolidated back to 20 because agent sprawl is worse than SaaS sprawl. Two agents answering the same question produce conflicting answers, and both sound confident. Reconciling them is harder than reconciling spreadsheets.
The Failure Mode That Matters
10K ran supervised for three deals before going autonomous on invoicing. It has produced one incorrect invoice since. That is a real error rate on real money, and the fix was running three deals with a human watching before letting go.
The mass email failure is the worse one. Irreversible actions need a hard stop that does not depend on the agent remembering to stop. The agent knew the rule. It forgot to check. A human makes that mistake once. An agent makes it at scale.
The implied comp model here is not "AI takes your job." It is "AI takes the work, and the org shrinks two headcount before the next hire." That is a different planning problem for sales leaders, and it is happening now, not in 18 months.