Entry No. 10 · Sell and close
SaaStr: three people, 21 agents, and a rebuilt sponsor business
SaaStr lost almost its entire sponsor base, cut to three humans, and put 21 AI agents into sales and marketing work, one agent per job, each trained on a proven human playbook. Read this file for exactly what it is: agents running go-to-market in substantial part. They do not run the whole company.
The story
Between 2023 and 2025, SaaStr lost almost its entire sponsor base and cut to three humans. The rebuild put 21 AI agents into sales and marketing, one agent per job, each trained on a proven human playbook rather than handed a vague instruction to go sell. By May 2026, SaaStr had 78 top-tier sponsors, 63 of them new.
The honest frame matters more here than in any other file in this directory. This is not an agent-run company. It is an agent-run function, go-to-market, inside a company with 13 years of content and a large list already creating demand. The agents worked, on the reported numbers, because the audience existed before they did. That precondition is part of the story, not a footnote to it.
How it was built
The article documents the operating design, not the order the 21 agents were deployed in, so no deployment sequence is stated here. The design: one agent per job, coordinated by an internal AI VP of Marketing called 10K, built on Replit. The playbooks cover inbound response, warm outbound, net-new account work, and re-engaging ghosted prospects. Humans train each agent for about 30 days.
A separate piece sits beside the sales and marketing stack: Digital Jason, a founder-clone product built on Delphi from more than 20 million words of SaaStr content. It is what happens when a founder's published thinking becomes training material with a name on it.
How it runs
The work splits by playbook. Qualified handles inbound. Artisan runs warm outbound. Monaco works net-new accounts. Salesforce Agentforce covers website visitors and ghosted prospects. 10K, the internal coordinator built on Replit, sits over the set. Humans handle negotiations, and humans review output daily, because the agents hallucinate.
The reported operating results: the inbound agent handled 442,000 chats and booked 614 meetings in the 2026 Annual cycle. One agent closed a $70,000 sponsorship on its own about six months before September 2026. The management cost is printed in the same article: each of the three humans spends 15 to 20 hours a week managing the stack. The agents run the function. The humans run the agents.
Tools in this entry
Tools named in the source, each tied to the job its agent does. Links in this box are affiliate links and are tagged as such.
The money and the proof
By February 2026: $4.8M in agent-sourced pipeline and $2.4M in closed-won revenue. The best leads are worth about $500,000 a year. One agent closed a $70,000 sponsorship on its own. SaaStr also says it put more than $500,000 into AI infrastructure. Every one of those figures is sourced to Lemkin or SaaStr, and the article flags them as not independently audited.
Filed as reported, with specifics. This is the largest money in the current directory, and it is not independently audited. The file holds both facts at once. The sponsor count behind it, 78 top-tier sponsors by May 2026 with 63 of them new, is reported the same way.
Where the seams are
The seams are staffed, scheduled, and expensive. Humans handle negotiations. Humans train each agent for about 30 days. Each human spends 15 to 20 hours a week managing the stack, and output gets reviewed daily because agents hallucinate. Add the infrastructure bill, more than $500,000 by SaaStr's own count, and the picture is complete: the function runs with three humans doing substantial management.
Steal this part
One agent per job, trained on a playbook a human already proved, and publish the operating results as the marketing. The precondition is part of the steal: SaaStr's agents worked because 13 years of content and a large list already created the demand. Copy the structure and you still have to bring your own audience.
Build your own
First move, this week: pick one go-to-market job that a human has already done well, and write its playbook down before you automate anything.
- Choose one job with a proven human playbook: inbound response, warm outbound, net-new accounts, or ghost follow-up. One job, one agent.
- Train the agent on that playbook for about 30 days, and review its output daily while it learns the edges.
- Split the next jobs the same way instead of widening the first agent. Coordination comes after separation, not before.
- Keep humans on negotiations, and budget the management time in the open: here it runs 15 to 20 hours a week per person.
- Publish the operating results with their grade attached. If the numbers are yours and unaudited, they run as reported.
Source
SaaStr's GTM Playbook: How It Rebuilt Its Customer Base With 3 People and 21 AI Agents, published by Origin Notes on Substack on October 1, 2026. SaaStr was founded by Jason Lemkin, the source of the figures graded above. Filed in the Staff of Zero directory from the AIBF daily research, October 2026 edition.