SaaStr
A B2B founder community, annual event and media business run by Jason Lemkin.
SaaStr replaced 2 to 3 SDRs and up to 5 AEs with 20 AI agents and 1.2 humans, keeping about the same output. Hundreds of thousands of AI emails performed the same labeled or not, and a Agentforce agent got a 70% response rate from leads reps skipped. The lesson: buy agents, train them on your best rep, segment your database, and assign a daily reviewer (10 to 15 hours a week).
- 2 to 3 SDRs
- Up to 5 AEs
- Reps quit at the event
- 20 AI agents
- 1 full-time AE
- 20% of Amelia's time
Productivity "about the same", per Lemkin's own account
From Jason Lemkin interview
What was said about SaaStr
1Replacing the SaaStr sales team with 20 agents and 1.2 humans
Twenty AI agents and 1.2 humans now match a ten-person sales team. SaaStr went from two to three SDRs and up to five AEs to one full-time AE, about 20% of chief AI officer Amelia's time, and 20 agents. Lemkin calls the productivity "about the same, not better, not worse."
The claim rests on Lemkin's own account. No revenue, conversion or cost figures are set against the old team, and SaaStr's sales are mostly inbound and renewals.
Rep churn and a self-closing clone triggered the switch to agents. Two well-paid reps quit on site at the 10,000-person event, and Lemkin decided he could not pay a junior SDR $150,000 a year to quit again on his eighth team. A general digital-clone agent not trained for sales had already closed a $70K sponsorship on its own.
He stopped hiring humans in sales and pushed agents as far as they would go. The office now has desks labeled with agent names, including Artisan's "Arty" and Qualified's agent.
Each sales motion gets its own agent, sometimes from a different vendor. Sponsorships averaging $70 to 80K, tickets from a couple hundred dollars to about $2,000, and lapsed attendees each run on separate agents and workflows. A human might reach 2,000 people from a 1.2 million list; an agent can work the whole list.
No master agent exists to coordinate others, so a human has to segment the database to stop agents colliding across inbound, retargeting and new outreach. "You've got to segment your base otherwise it just becomes a mess."
2How SaaStr trains and oversees its agents
- 1Upload website, wiki, emailsTop rep's best emails as template
- 2Answer the agent's questions
- 3Fix mistakes daily1 to 2 hours a day, ~30 days
- 4Agent A/B tests variantsPersonalizes lightly with CRM data
- 5One owner reviews dailyAmelia: 10 to 15 hours a week
From Jason Lemkin interview
Agents need about 30 days of daily training before they perform. Training means uploading the website, wiki and best emails, answering the agent's questions, then spending an hour or two a day for about 30 days fixing its mistakes. Lemkin puts it at roughly 50 to 60 hours.
Vendors in 2024 claimed you could turn a product on and get revenue; that and weaker models caused failures. The digital Jason once gave out wrong event dates, and a public B2B company worth over $10 billion had not trained its agent at all.
Use your best rep's emails as the template, then let the agent test. SaaStr uploads the best copy from its top rep or marketer; the agent generates variants, A/B tests them and personalizes lightly with CRM data and website visitor tracking. It has sent hundreds of thousands of emails this way.
The bar is the midpack human, whose emails are often worse. Lemkin says poor AI emails usually mean a poorly trained product or a bad vendor.
Give one data-loving owner the job of reviewing agents daily. Amelia, SaaStr's chief AI officer, spends 10 to 15 hours a week reviewing output from 20 agents, which work nights, weekends and Christmas. Lemkin is not sure they can add a 21st.
He recommends promoting an internal marketer, RevOps or product person who has already built apps with AI tools. Buying an agent and disappearing gives zero ROI.
3Where SaaStr's agents paid off first
Start where no human does the work today, so any result is gain. The digital clone began answering event support questions, then an inbound agent qualified every website visitor invisibly and booked meetings around the clock. One got a sponsor at 11 p.m. on a Saturday.
Lemkin says being qualified by a 21-year-old a day after hitting "contact me" "doesn't feel good" to the customer, and most apps lack 24/7 support.
An agent working leads reps had skipped got a 70% response rate. SaaStr trained its Agentforce agent on the same prompt and sent it to inbound leads sales had judged not worth their time. "Those are people that were dying to interact with us."
Reps skip leads when commission is small or they are busy. Lemkin says a few thousand leads or website visitors is enough data to benefit.
Labeling emails as AI made no difference to responses. SaaStr tested emails signed as digital Jason or Amelia against ones posing as human, across hundreds of thousands of emails. Recipients did not care as long as the email added value and replies were instant.
Some founders reply that they can tell it is AI and ask for a meeting anyway. "We're creating issues as excuses to not do the work."
Find the worst experience by testing your product as a stranger. Lemkin's incognito test: open a private browser with a fresh Gmail address and try support, contact sales, newsletter signup and onboarding. Fix the worst one with an agent, and repeat quarterly.
He says you will find slow sales replies and bad support you had forgotten about.
4Buying agents and choosing vendors
Pick the vendor whose engineers will help you deploy. SaaStr chose Artisan and Qualified because they offered the most hands-on help, and both deployments worked quickly. Another vendor wanted $100K up front; others feared bad PR.
Lemkin says leading AI GTM tools are "more similar than different" under the hood, so add a column to the vendor matrix for the forward deployed engineer or solutions architect and talk to them before signing.
Buy GTM agents rather than building them. Lemkin builds apps in Replit, including a startup calculator used 800,000 times in 90 days and a pitch deck reviewer that has reviewed almost 3,000 decks, but none of SaaStr's GTM agents. "Don't build it yourself."
He compares it to building your own Notion: vendor pace makes an internal build obsolete in months. Only a company like Vercel with an eager full-time engineer should build.
5What the experiment means for sales jobs
Email SDRs and lead qualifiers go first, then the midpack AE. Lemkin predicts cadence-email SDRs will be 90% displaced within a year and inbound qualifiers mostly extinct. He expects the safe share of AE work to fall from about 70% to 40 or 50%.
Agents beat the midpack, "not the best." These figures are his guesses, and he offers no fix for disappearing entry-level paths into sales.
Keep humans on the top enterprise targets and use AI for volume. For a whiteboard of 50 sponsors worth half a million dollars over two years, hand-written outreach is worth it and AI only helps with research and editing. At 5,000 prospects, humans cannot reach enough people.
Enterprise, in-person and high-value sales still favor humans, and fast-growing AI companies like OpenAI say they cannot hire enough enterprise reps.
People who deploy an agent themselves become far more employable. Lemkin's advice: pick the most painful problem, pick a vendor that treats you well, and do the ingesting, training and QA yourself. "If you can go do this and get it live into production, you're hyper employable."
Teams that hand the work to an agency or juniors fail. He imagines $250,000 SDRs managing ten agents each, and says only about 20% of today's GTM leaders fit the AI era.
6The SaaStr business behind the experiment
SaaStr grew from a founder blog into an eight-figure sponsorship business. Lemkin started the blog in 2012 after selling his startup to Adobe; it became a community, a 10,000-person annual event and an investing pipeline. Revenue comes mostly from about a hundred sponsors averaging $70 to 80K plus $4 to 5M a year in tickets.
He has invested almost $200 million, returning almost 10x, only in founders from the community. Events carry heavy costs while the media side has almost none.
We replaced our sales team with 20 AI agents—here’s what happened next | Jason Lemkin (SaaStr)
Replaced a team of 2 to 3 SDRs and up to 5 AEs with 20 AI agents, 1.2 humans and about the same output. Its eight-figure revenue comes mostly from sponsorships averaging $70 to 80K and $4 to 5M a year in tickets. It has sent hundreds of thousands of AI emails with no difference whether labeled or not.
Twenty AI agents and 1.2 humans now match a ten-person sales team. SaaStr went from two to three SDRs and up to five AEs to one full-time AE, about 20% of chief AI officer Amelia's time, and 20 agents. Lemkin calls the productivity "about the same, not better, not worse."
The claim rests on Lemkin's own account. No revenue, conversion or cost figures are set against the old team, and SaaStr's sales are mostly inbound and renewals.
Rep churn and a self-closing clone triggered the switch to agents. Two well-paid reps quit on site at the 10,000-person event, and Lemkin decided he could not pay a junior SDR $150,000 a year to quit again on his eighth team. A general digital-clone agent not trained for sales had already closed a $70K sponsorship on its own.
He stopped hiring humans in sales and pushed agents as far as they would go. The office now has desks labeled with agent names, including Artisan's "Arty" and Qualified's agent.
Each sales motion gets its own agent, sometimes from a different vendor. Sponsorships averaging $70 to 80K, tickets from a couple hundred dollars to about $2,000, and lapsed attendees each run on separate agents and workflows. A human might reach 2,000 people from a 1.2 million list; an agent can work the whole list.
No master agent exists to coordinate others, so a human has to segment the database to stop agents colliding across inbound, retargeting and new outreach. "You've got to segment your base otherwise it just becomes a mess."
Agents need about 30 days of daily training before they perform. Training means uploading the website, wiki and best emails, answering the agent's questions, then spending an hour or two a day for about 30 days fixing its mistakes. Lemkin puts it at roughly 50 to 60 hours.
Vendors in 2024 claimed you could turn a product on and get revenue; that and weaker models caused failures. The digital Jason once gave out wrong event dates, and a public B2B company worth over $10 billion had not trained its agent at all.
Use your best rep's emails as the template, then let the agent test. SaaStr uploads the best copy from its top rep or marketer; the agent generates variants, A/B tests them and personalizes lightly with CRM data and website visitor tracking. It has sent hundreds of thousands of emails this way.
The bar is the midpack human, whose emails are often worse. Lemkin says poor AI emails usually mean a poorly trained product or a bad vendor.
Give one data-loving owner the job of reviewing agents daily. Amelia, SaaStr's chief AI officer, spends 10 to 15 hours a week reviewing output from 20 agents, which work nights, weekends and Christmas. Lemkin is not sure they can add a 21st.
He recommends promoting an internal marketer, RevOps or product person who has already built apps with AI tools. Buying an agent and disappearing gives zero ROI.
Start where no human does the work today, so any result is gain. The digital clone began answering event support questions, then an inbound agent qualified every website visitor invisibly and booked meetings around the clock. One got a sponsor at 11 p.m. on a Saturday.
Lemkin says being qualified by a 21-year-old a day after hitting "contact me" "doesn't feel good" to the customer, and most apps lack 24/7 support.
An agent working leads reps had skipped got a 70% response rate. SaaStr trained its Agentforce agent on the same prompt and sent it to inbound leads sales had judged not worth their time. "Those are people that were dying to interact with us."
Reps skip leads when commission is small or they are busy. Lemkin says a few thousand leads or website visitors is enough data to benefit.
Labeling emails as AI made no difference to responses. SaaStr tested emails signed as digital Jason or Amelia against ones posing as human, across hundreds of thousands of emails. Recipients did not care as long as the email added value and replies were instant.
Some founders reply that they can tell it is AI and ask for a meeting anyway. "We're creating issues as excuses to not do the work."
Find the worst experience by testing your product as a stranger. Lemkin's incognito test: open a private browser with a fresh Gmail address and try support, contact sales, newsletter signup and onboarding. Fix the worst one with an agent, and repeat quarterly.
He says you will find slow sales replies and bad support you had forgotten about.
Pick the vendor whose engineers will help you deploy. SaaStr chose Artisan and Qualified because they offered the most hands-on help, and both deployments worked quickly. Another vendor wanted $100K up front; others feared bad PR.
Lemkin says leading AI GTM tools are "more similar than different" under the hood, so add a column to the vendor matrix for the forward deployed engineer or solutions architect and talk to them before signing.
Buy GTM agents rather than building them. Lemkin builds apps in Replit, including a startup calculator used 800,000 times in 90 days and a pitch deck reviewer that has reviewed almost 3,000 decks, but none of SaaStr's GTM agents. "Don't build it yourself."
He compares it to building your own Notion: vendor pace makes an internal build obsolete in months. Only a company like Vercel with an eager full-time engineer should build.
Email SDRs and lead qualifiers go first, then the midpack AE. Lemkin predicts cadence-email SDRs will be 90% displaced within a year and inbound qualifiers mostly extinct. He expects the safe share of AE work to fall from about 70% to 40 or 50%.
Agents beat the midpack, "not the best." These figures are his guesses, and he offers no fix for disappearing entry-level paths into sales.
Keep humans on the top enterprise targets and use AI for volume. For a whiteboard of 50 sponsors worth half a million dollars over two years, hand-written outreach is worth it and AI only helps with research and editing. At 5,000 prospects, humans cannot reach enough people.
Enterprise, in-person and high-value sales still favor humans, and fast-growing AI companies like OpenAI say they cannot hire enough enterprise reps.
People who deploy an agent themselves become far more employable. Lemkin's advice: pick the most painful problem, pick a vendor that treats you well, and do the ingesting, training and QA yourself. "If you can go do this and get it live into production, you're hyper employable."
Teams that hand the work to an agency or juniors fail. He imagines $250,000 SDRs managing ten agents each, and says only about 20% of today's GTM leaders fit the AI era.
SaaStr grew from a founder blog into an eight-figure sponsorship business. Lemkin started the blog in 2012 after selling his startup to Adobe; it became a community, a 10,000-person annual event and an investing pipeline. Revenue comes mostly from about a hundred sponsors averaging $70 to 80K plus $4 to 5M a year in tickets.
He has invested almost $200 million, returning almost 10x, only in founders from the community. Events carry heavy costs while the media side has almost none.
More in Business software
Business softwareMicrosoft
A software company founded by Bill Gates and Paul Allen that made BASIC, MS-DOS and Windows.
40 sources · 100 tips
Business softwareFigma
A collaborative design platform with design, whiteboarding, slides, dev handoff and AI prototyping tools.
20 sources · 81 tips
Business softwareShopify
A commerce platform that lets merchants build online stores and sell to customers.
23 sources · 55 tips
Business softwareAtlassian
A software company behind Jira, Confluence, Trello and other team collaboration tools.
13 sources · 40 tips
Business softwareSlack
A workplace messaging platform for teams, acquired by Salesforce.
25 sources · 46 tips
Business softwareAirtable
A no-code platform for building collaborative business apps on a real-time database.
9 sources · 43 tips
Business softwareMiro
An online visual collaboration whiteboard for teams.
7 sources · 36 tips
Business software37signals
A small, independent, profitable software company behind Basecamp, HEY and Highrise, run remotely with about 60 people.
5 sources · 41 tips