SaaStr's Agents Now Take Eight Hours a Day to Run. Aim the First One at B-Leads.

SaaStr published the October 2026 edition of its AI Agent Playbook for GTM on October 12. It runs 53 pages and 27 parts on operating the company with three humans and more than 21 agents. The free PDF ends on a seven-item checklist, and item three is the lead rule. It tells teams to point an agent at their B-leads, the scored contacts with real signal that no rep will touch, and to leave the A-leads to a human.

Part 21 turns that into a table. A-leads go to any rep, since a buyer ready to sign gets an answer within a minute whoever is on shift. B-leads go to the agent, and C and D leads wait until the B-lead motion works. The receipt is $500,000 of extra revenue from one event, generated by Artisan, the AI SDR SaaStr runs for warm outbound.

SaaStr first published the B-lead rule and its receipt in May

None of the receipts in the October edition are new. The B-lead rule and the $500,000 first ran in SaaStr's May 29 recap of SaaStr AI Annual 2026. There the same money was "sponsor revenue from B leads this year" rather than one event's take, and it came up again in a June 3 episode of The Agents, SaaStr's podcast. The 614 meetings booked by one inbound agent come from that May recap too.

The cover numbers are older still. SaaStr reported $4.8 million of agent-sourced pipeline and $2.4 million closed-won in a February 10 post written eight months into the deployment, along with the claims that deal volume more than doubled and win rates nearly doubled. The October PDF still introduces them as eight months of production data.

Only the placement changed. A rule that lived in an event recap is now the third move on the closing checklist, and the PDF tells founders who can't get AI SDRs to work that they're probably aiming them at the wrong leads.

That placement is right, mostly for a measurement reason. An agent working A-leads competes with a rep who would have closed the deal anyway, so much of its pipeline is credit taken from humans.

An agent working leads nobody was going to call has a baseline of zero. Every meeting it books is lift a CRO can count without trusting the vendor's attribution model. SaaStr sizes the $500,000 modestly ("Not a huge business on its own."), and a first agent judged on that kind of found money has a fair test.

The new number in the October edition is the oversight bill

The human time is the figure that moved between editions. In February, SaaStr founder Jason Lemkin wrote that he and his chief AI officer each spent 15 to 20 hours a week managing agents.

Part 25 of the October PDF says two and a half agents took the pair about 30 minutes a day a year ago. In the week the edition went out, it was eight hours a day each. Part 1 of the same document still prints the 15-to-20-hour figure, so the playbook now carries both.

SaaStr's explanation is about what the agents do. A human checks the output of an agent that does a task. An agent that makes decisions needs a human opinion on each one, and it doesn't report all of them. The PDF's own examples include an agenda agent that quietly stopped paginating at 50 sessions, and an agent that emailed founders and investors from an address its memory marked as prohibited, a send SaaStr calls irreversible.

The software bill has barely moved. SaaStr put its agent spend at more than $500,000 for the first year in September 2025, and the October PDF says roughly $500,000 a year across all agents. Two people at eight hours a day comes to about 4,000 hours a year, or two full-time jobs.

SaaStr says roughly one of those eight hours goes to the work itself and the rest to building whatever its marketing agent surfaces next. So the agents write their own backlog, and the humans pay for it in hours. enso's turn to human-run agents pointed the same way, since buyers of agent work end up paying for the people who run it.

SaaStr's agent vendors are also its sponsors

SaaStr is an events and media company, and its playbook shows its vendors are also its customers. In a live demo recounted in Part 22, SaaStr's customer-success agent flagged Artisan and Salesforce among the sponsors at risk of not renewing. Part 10 says SaaStr is the top-performing customer for both Artisan and Qualified, whose CEO hosted SaaStr's agent story on Qualified's own site in October 2025. Salesforce published SaaStr as an Agentforce customer story on July 31, with the same 614 meetings and Qualified among the featured products.

The numbers may be accurate. They're also self-measured, unaudited and republished as marketing by the tools that produced them, the pattern our piece on who measures AI agent scores flagged in sales and support agents.

Vendor-neutral evidence on AI SDRs is thin and doesn't flatter the category. SaaStr's own November 2024 reader survey found 3% of respondents had closed revenue from AI SDRs and 83% had gotten nothing. Gartner predicted in June 2025 that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value and weak risk controls. The PDF itself says 90% of AI SDR implementations fail, without a source.

That evidence hits the receipts harder than the rule. A B-lead pilot is the one deployment in the playbook a skeptic can check without believing SaaStr, because the comparison is a pile of leads that would otherwise get no touch. It also avoids the failure that lemlist's bootstrapped $60 million pointed to, AI SDRs sold as headcount replacement and judged against reps they never replaced.

A B-lead agent can prove its pipeline. Its oversight runs eight hours a day

The B-lead rule binds any company with a lead backlog its reps don't work, and Salesforce has now put a number on it. Itai Margalit, its Area VP of Sales in Israel, said at the Israel IT Conference on October 18 that about 75% of the company's 100 million leads used to go unanswered. Agents working them have generated more than $88 million in sales pipeline. That's pipeline, not closed revenue, and Salesforce gave no period, conversion rate or holdout group for the figure.

SaaStr's playbook carries the cost side of the same deployment. Its agent software runs roughly $500,000 a year, and its two agent managers now spend eight hours a day each, about 4,000 hours a year.

An agent working leads nobody was going to call has a baseline of zero. Every meeting it books is lift a CRO can count without trusting the vendor's attribution model.

Gartner's forecast sets the next marker: more than 40% of agentic AI projects canceled by the end of 2027, on escalating costs, unclear business value and weak risk controls.

By Business of GTM