LLenny's Podcast
← All episodes
Jason Lemkin (SaaStr)01 January 2026

We replaced our sales team with 20 AI agents—here’s what happened

7Frameworks
15Insights

Frameworks in this episode

Sales4 steps

Best-Rep Email Cloning

Feed the agent your single best rep's best email, then let it AB-test variants—that's how AI beats your mid-pack.

Read →
Sales4 steps

The 30-Day Agent Training Loop

Ingest your docs, then correct the agent an hour a day for 30 days until it performs like your best rep.

Read →
Entrepreneurship4 steps

The Chief Agentic GTM Officer Play

Pick one painful problem, one leading vendor, and deploy an agent yourself—50 hours later you're hyper-employable.

Read →
Sales4 steps

The Forward-Deployed Engineer Vendor Column

Don't buy AI GTM tools on features—buy on who will actually get on the phone and deploy it with you.

Read →
Sales4 steps

The GTM Agent Deployment Sequence

Roll out AI sales agents in the order of lowest effort and highest ROI: start with support, end with reactivation.

Read →
Strategy4 steps

The Incognito Cry Test

Go through your own product as an anonymous customer, find what makes you cry, and buy an agent to fix that.

Read →
Sales4 steps

The Lead-Rich Audit

You don't need SaaStr's scale—count your website visitors and untouched CRM leads to see you're already lead-rich.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 2

Myth Buster36:30

You Don't Need SaaStr-Scale Data To Make Agents Work

A common excuse Lemkin hears is 'SaaStr has scale, we're too small.' He pushes back: even a company with 300 customers has ~30,000 website visitors and tens of thousands of leads in its CRM that no human is contacting. You need a bit of scale and traffic, not millions of names. The same holds for training data — a couple of months of really good content plus a long tail beats a 12-year archive.

  • You don't need millions of names — a bit of scale and traffic is enough
  • 300 customers usually means ~30k visitors and tens of thousands of untouched CRM leads
  • Vendors sometimes tell prospects they have 'too much data' — that's wrong
  • Deep training history matters less than expected; a couple months of good content suffices

“You don't need the scale of numbers that you and I have to make these agents work.”

Jason Lemkin · 37:00

“I thought having 12 years of content made the difference. Nah.”

Jason Lemkin · 37:30
#ai-agents#data#myth-buster#small-business
Myth Buster44:30

Nobody Cares If the Email Is From AI — And Human Emails Aren't Better

After sending hundreds of thousands of emails, SaaStr found it doesn't matter whether you disclose the sender is AI — they've tried disclosing, faking, and just sending, and now they just send it. People don't care as long as the email adds value and gets an instant response. Lemkin reinforces this with the story of inheriting Adobe reps whose emails were the worst his sales leader had ever read: AI clears that bar easily.

  • They tested disclosing vs faking vs just sending — now they just send it
  • People don't mind AI as long as the email adds value and responds instantly
  • Founders often reply 'I can tell this is AI but it's pretty good, can we meet?'
  • Human reps' emails are often far worse than people assume

“These are the worst emails that I've ever read.”

Jason Lemkin · 45:00

“We We just send it and no one cares.”

Jason Lemkin · 45:30
#outbound#email#ai-agents#myth-buster

Hot Take· 5

Hot Take13:00

AI Is Displacing the Midpack and the Mediocre, Not the Best

Lemkin argues AI isn't replacing top salespeople — it's taking the jobs people don't want to do and displacing the mid-tier and below. He'd happily hire two more great humans tomorrow, but won't hire a rep who still doesn't understand the product after three months. The best humans get superpowers from AI; he's not convinced the rest do.

  • AI takes the jobs people don't want to do and displaces the midpack and mediocre
  • The reps who don't understand what the product does are the ones at risk
  • Great enterprise reps are still in huge demand — OpenAI 'can't hire enough'
  • The best get superpowers from AI; the rest may not

“AI is replacing the jobs people don't want to do today and it is displacing the media the midpack and the mediocre.”

Jason Lemkin · 13:30
#ai-agents#sales#careers#hot-take
Hot Take14:00

The Plays Still Work — It's the Playbooks That Are Broken

Lemkin's frame for GTM in the AI age: outbound, webinars, podcasts, events all still work. What's broken is the old playbooks. For companies not riding AI demand, growth has decelerated so much that nothing seems to work; for the ones blowing up, the plays work but are run from a hyper-PLG posture where the job is choosing which of too many leads to answer.

  • All the plays (outbound, webinars, podcasts, events) still work
  • The playbooks are broken — old-school SaaS plays lack the ROI they had in 2021
  • Hypergrowth AI companies have so many leads that half the job is picking which to respond to
  • The market is wildly bifurcated: low-end price increases vs high-end demand explosion

“All the plays work. It's the playbooks that are kind of broken in the age of AI.”

Jason Lemkin · 14:00
#go-to-market#sales#ai#strategy
Hot Take19:30

The Email SDR Will Be 90% Displaced by AI Next Year

Lemkin's prediction for the sales profession: the classic email-cadence SDR and the human who qualifies inbound leads will be mostly extinct within a year. AI already qualifies everyone on the website invisibly and books the meeting — there's no reason to make a prospect wait two days for a 21-year-old who doesn't know the product. Door-knocking SDRs survive; the AE's job is ~70% safe next year but declines toward 40–50%.

  • Email-based cadence SDRs will be ~90% displaced by AI next year
  • Human inbound qualifiers should be 'extinct next year' — it's a bad customer experience anyway
  • Door-to-door SDRs are not displaced
  • AE jobs ~70% safe by end of next year, declining to 40–50% over time

“The emailbased cadence SDR will be 90% displaced by AI next year.”

Jason Lemkin · 19:30

“They should be extinct next year.”

Jason Lemkin · 20:00
#sdr#sales#careers#ai#predictions
Hot Take28:30

Don't Build Your Own GTM Agents — Buy Them (Unless You're Vercel)

Lemkin is a self-described top-1% Replit user who built many apps himself, but none of SaaStr's GTM agents were built in-house. His argument: it's the same reasoning as building your own Notion — you could, but you shouldn't, because these products aren't that expensive and the pace of innovation means anything you build internally goes obsolete in a couple of months unless you have a dedicated, eager engineer.

  • None of the GTM stuff was built in-house — buy, don't build
  • Same logic as building your own Notion: you could, but don't
  • The products aren't that expensive relative to a human
  • In-house builds go obsolete in months because the pace of innovation is so fast

“None of the GTM stuff we built ourselves. Don't build it yourself. You're not Versel.”

Jason Lemkin · 28:30
#build-vs-buy#ai-agents#go-to-market#hot-take
Hot Take1:15:00

AI Has Raised the Bar: Deliver ROI Before the Contract Is Signed

Lemkin frames this as maybe the biggest change of all. The old B2B model was to avoid pilots, minimize them, then roll out over years — almost rip off the customer to get the deal. AI has flipped customer expectations: the winners blowing up are the ones delivering ROI before the document is e-signed. Companies still playing the old games are the public companies growing 8%.

  • Old model: avoid pilots, then roll out slowly over years
  • AI has raised what customers expect — go live and get value before you pay
  • Salesforce's Benioff envies Palantir's ability to make customers live before paying
  • Companies clinging to the old games are the ones growing ~8%

“I wish every Salesforce customer now could go live before they pay”

Jason Lemkin · 1:15:30

“you've got to deliver the ROI before the document is e- signed today.”

Jason Lemkin · 1:16:00
#sales#roi#pilots#ai#hot-take

Explainer· 2

Explainer17:30

Why AI Startups Are Blowing Up: Everyone's In-Market At Once

Lemkin explains that the traditional B2B assumption was that only 3–5% of prospects are in-market to buy in any given year. With AI, that has flipped — in many categories more than 50% are in-market at once, because there's top-down pressure to adopt AI everywhere. It's not one law firm evaluating Harvey; it's everyone. He warns this is a bubble-like window that will eventually revert.

  • Traditionally only 3–5% of prospects are in-market in a given year
  • For high-ROI AI products, it's now north of 50% in-market at once
  • The demand is driven by top-down 'we need to adopt AI' pressure across whole categories
  • This artificial window will end and revert to more old-school patterns

“the traditional metric was in most categories, 3 to 5% of prospects would be in market a year.”

Jason Lemkin · 17:30

“In many categories, we're north of 50% in market.”

Jason Lemkin · 18:00
#ai#demand#market-timing#b2b
Explainer54:00

Who Should Orchestrate Your Agents: Promote a Data Nerd, Not a Salesperson

Lemkin doesn't believe the 'GTM engineer' hire exists yet — instead, promote someone internally who is quant, loves data, and will sit in front of it for a couple hours a day. They can come from product, marketing, or RevOps, but almost never from regular sales. The key point people misunderstand: agents operate autonomously but require constant oversight and iteration — buy one and disappear and you get zero ROI.

  • Promote internally — the role is best filled by a quant 'nerd', not hired off LinkedIn
  • Candidates come from product, marketing, or RevOps; odds from regular sales are ~zero
  • Agents operate autonomously but need constant oversight and iteration
  • Buy a product and disappear and you'll get zero ROI

“It's got to be a nerd, someone that likes marketing and sales and is quant.”

Jason Lemkin · 54:00

“Odds they come out of regular sales approach zero.”

Jason Lemkin · 54:30
#orchestration#ai-agents#hiring#revops

Story· 1

Story07:00

How SaaStr Went From 10 Sales Humans to 1.2 Humans + 20 Agents

After two paid, high-end sales reps quit on-site at SaaStr's 10,000-person event, Jason Lemkin decided to stop hiring humans in sales entirely and push the limits with AI agents. He now runs the go-to-market org with 1.2 humans (one AE plus 20% of a chief AI officer) and 20 agents, doing roughly the same business he did with ~10 people. His verdict: net productivity is about the same, but it's far more efficient because software scales.

  • Two high-paid reps quit on-site at the annual event; it was the eighth sales team he'd built
  • A horizontal 'general' agent (Deli / digital Jason) had already closed a $70k sponsorship on its own
  • Now ~1.2 humans + 20 agents does about what 10 human GTM people did
  • Net productivity is roughly the same, not better or worse — the win is efficiency and scale

“We're done with hiring humans in sales. We're done.”

Jason Lemkin · 07:30

“The pro the net productivity is about the same. It's not better. It's not worse.”

Jason Lemkin · 09:00
#ai-agents#sales#go-to-market#saastr#automation

Tool· 1

Tool1:33:30

Jason's Cheat-Code Tool: Reve for B2B Marketing Images

In the lightning round, Lemkin recommends an image app called Reve (formerly Reeve/Reevar, at app.reve.com) whose team built their own image LLM. He uses it two or three times a day for the 'boring B2B stuff' — thumbnails, article images — and finds nothing better for that use case, replacing the old workflow of firing up Canva or waiting days for a designer.

  • Reve (formerly Reeve/Reevar) is at app.reve.com; the team built their own image LLM
  • Best-in-class for boring B2B marketing images, thumbnails, and article art
  • Lemkin uses it two or three times a day
  • Replaces the old Canva-or-wait-four-days-for-a-designer workflow

“There is an app called Reeve. It used to be Reevar.”

Jason Lemkin · 1:33:30

“it's just a simple prompt. Go to it, type in whatever image you want to make.”

Jason Lemkin · 1:34:00
#tools#image-generation#marketing#recommendation

Takeaway· 4

Takeaway23:30

Want To Be Hyper-Employable? Pick One Agent and Train It Yourself

Lemkin's advice for anyone worried about their job: pick one agentic tool for your most painful problem, pick a leading vendor, and personally do the ingestion, training, QA, and daily corrections for ~30 days. Almost nobody does this — most people stay panicked — so the ones who actually get an agent live into production become hyper-employable and could be hired as a 'chief agentic GTM officer.'

  • Pick one tool for your most acute problem and a leading vendor that treats you well
  • Do it yourself: ingest the data, train it, QA it, correct mistakes for ~30 days
  • By day 30 of an hour or two of daily correction, the agent is 'pretty good'
  • Almost nobody does the work — doing it makes you hyper-employable

“Pick a tool, an agent, an agentic tool to solve one of your problems.”

Jason Lemkin · 23:30

“If you can go do this and get it live into production, you're hyper employable.”

Jason Lemkin · 27:00
#careers#ai-agents#upskilling#advice
Takeaway33:00

Pick the Vendor That Will Actually Help You Deploy, Not the Best Features

SaaStr's success with Artisan and Qualified came down to one thing: those vendors offered to help the most. Because these agents take a month to train, the world's best software with no deployment help is not what most buyers should choose. Lemkin says to add a new column to your evaluation matrix — the forward-deployed engineer — and get on the phone before writing a check to confirm they'll really do the deployment.

  • The deciding factor for the first two agents was 'they offered to help the most'
  • Best software with no training help is wrong for 99% of buyers
  • Add a 'who will help me' column (forward-deployed engineer / SE) to your vendor matrix
  • The best vendors even turn away business they don't think they can make successful

“But the key to the first two and and if you're going to pick an agent is they they offered to help the most.”

Jason Lemkin · 35:00

“before you write a check get on the phone with Lenny and see if Lenny's really going to do deployment”

Jason Lemkin · 36:00
#vendor-selection#ai-agents#forward-deployed-engineer#sales
Takeaway42:30

The Trick To Good AI Outbound: Train It On Your Best Person's Emails

Lemkin's two biggest learnings on email quality: take your best salesperson or marketer's email copy and use it as the template for the AI, then let the agent A/B test and generate variants from it. The 2024 failures came from launching before Claude 4 and from vendors lying that you could just 'turn it on.' If you're getting terrible AI emails, that's a poorly trained product from a bad vendor.

  • Use your best person's email copy as the AI's template, then let it A/B test variants
  • Agents are good at creating variants — ask Claude for three versions of your best email
  • 2024 failures were pre-Claude-4 and from vendors claiming no training was needed
  • Terrible AI emails = poorly trained product from a bad vendor

“take your best person on your sales team, the best marketer you have, take their email copy and use that as a template for your…”

Jason Lemkin · 42:30

“If you're getting terrible emails, it's a poorly trained product from a bad vendor.”

Jason Lemkin · 44:00
#outbound#email#ai-agents#training#sales
Takeaway1:25:30

The Incognito Test: Buy Your Own Product With a Fresh Gmail

Lemkin's actionable advice for anyone who doesn't feel the excitement: over the holidays, open an incognito browser, use a fresh Gmail, and go through your own product end to end — try support, contact sales, sign up for the newsletter. You'll cry at how bad your support is and how long sales takes to respond. Pick the thing that makes you cry the most and go buy an agent to fix it.

  • Use incognito + a fresh Gmail to experience your own product as a new customer
  • Try support, contact sales, and signup flows end to end
  • You'll be shocked by bad support and slow sales response
  • Pick the worst experience and buy an agent to fix it — do this quarterly

“Do it in incognito. Go to your app and do everything with a fresh Gmail address.”

Jason Lemkin · 1:25:30

“Pick the thing that makes you cry the most over your mold wine and go buy that agent and fix it.”

Jason Lemkin · 1:26:00
#customer-experience#advice#ai-agents#support