The Numbers
SaaStr's two production AI agents cost $257 monthly to run. Combined.
One handles marketing operations: refreshes ticket data, updates dashboards, drafts newsletters, compares year-over-year performance, logs financials. The other manages 100+ event sponsors, sends personalized check-in emails, and runs an autonomous chatbot for contractors.
Founder Jason Lemkin initially thought the $257 was a daily burn rate. It is not.
Why Agent Economics Changed
95% of LLM calls now run on GPT-4o mini, which costs under a penny per call. Force-ranking data, running comparisons, and drafting updates do not require expensive models. The hallucination rate is manageable with basic cleanup.
The expensive work happens in connected apps, not the AI layer. API calls to Salesforce, marketing automation, and event platforms are mostly free at the margin. Database storage: $0.10 to $0.30 monthly.
Real cost stack: $257 for LLM calls, $22k annually for Salesforce and connected apps, $30 monthly for auth. That is it.
What the Agents Actually Replace
SaaStr's Director of Demand Gen used to do these tasks manually: compile weekly numbers, schedule emails, write newsletters, draft social posts pre-event. That role now runs through an agent.
The agent cannot do strategy, PR, hiring, cross-functional negotiation, brand judgment, or crisis management. Those remain human work. But the bottom half of a marketing team, the analyst layer, the ops coordinator, parts of junior content work: automated.
One agent sends 83 personalized sponsor emails at 12:20am. Another built an applicant tracking system in 10 minutes at midnight for $2 in LLM costs.
What This Means for GTM Teams
Cost is no longer the constraint. You can run multiple autonomous agents for less than one SDR's monthly phone bill. The question shifts from "can we afford this" to "what should we automate first."
For lean ANZ SaaS teams, this changes headcount planning. If marketing ops, sponsor management, and personalized outreach run for $250 monthly, the math on hiring changes fast.
Lemkin's take: if a vendor shipped a better AI CMO tomorrow, he would switch in 60 seconds. He has not seen one yet. Most "AI CMO" tools are ad managers with better branding.
Historical context: 11 months ago, 80 to 90% of agent code was throwaway. You burned real money on the agent fixing its own mistakes. That waste is mostly gone. Reliability improved, costs dropped, and suddenly the unit economics make sense for operational work that does not require judgment.
The agents are not replacing VPs. They are replacing the work that used to justify three headcount: the analyst, the coordinator, the junior marketer. That is enough to matter.