Jason Lemkin Replaced His Sales Team with 20 AI Agents. Here's What He Learned.

Jason Lemkin didn’t set out to run an experiment. His last salesperson quit, and instead of backfilling, he decided to try something radical: replace the entire go-to-market team with AI agents. What started as necessity became conviction. SaaStr now runs with 1.2 humans and 20 AI agents doing the work that 10 full-time people used to do, at roughly the same revenue.

This conversation with Lenny Rachitsky is part war story, part manifesto. Lemkin isn’t theorizing about AI in sales. He’s living it, making mistakes, iterating daily, and watching the same pattern play out across his portfolio of SaaS investments. The result is one of the most concrete, least hype-driven pictures of what AI is actually doing to go-to-market right now.

20 Agents, 1.2 Humans

SaaStr used to have about 10 people in go-to-market roles: SDRs, AEs, support. Now they have 20 AI agents, each with a nickname printed on the desks where humans used to sit. Reply for Replit, Quali for Qualified, Arty for Artisan, and so on.

The business is performing similarly to when it had 10 humans. Lemkin is blunt about the trade-off: the agents work all night, on weekends, on Christmas. They don’t take three months to learn what SaaStr does. If he could find two great humans who wanted to join, he’d hire them tomorrow, but he’s done hiring mediocre salespeople who still don’t understand the product after their third month.

His operating model: one human manages the agents. Another half-person handles edge cases. Everyone else is an AI. The total headcount for go-to-market went from 10 to 1.2.

The SDR Extinction Timeline

Lemkin’s most pointed prediction: the classic SDR, the junior hire out of college running cadence-based email campaigns, will be “mostly extinct” within 12 months. The qualifier role, people who field inbound “contact me” requests, will follow on the same timeline.

His reasoning is straightforward. These are pattern-matching jobs. The SDR reads a lead, writes a templated email, follows up on a schedule. AI does all of this, 24/7, without the three-month ramp. For qualifying inbound leads, it’s even more clear-cut: the current human process is a bad experience for customers, and AI handles it faster.

“The classic SDR junior kid that is hired out of college to send emails, we don’t need them. Folks that qualify leads coming in, the ‘contact me’ that we see, we have no need for them today.”

What survives, and even thrives, is the high end. Field sales is untouched. Enterprise reps at companies like Salesforce are hiring more than ever. The people who can close complex, high-value deals aren’t going anywhere. But the office-based, work-from-home middle tier is getting compressed.

The $250K SDR

Lemkin sees a new archetype emerging: the $250K-a-year SDR who manages 10 agents instead of being an agent. At Vercel, he points to Janine Pelosi’s team as an early example, humans who orchestrate AI rather than doing repetitive tasks themselves.

This is the “hyper-employable” path. If you can deploy an agent today, train it, iterate on it daily, then scale to two, three, four agents, you become geometrically more productive. The pay should follow: $250K instead of $80K-90K.

The catch: you have to actually do the work. Buy an agent, deploy it yourself, train it, ingest your data, iterate every single day. Most people won’t. Lemkin estimates only about 10% of the GTM professionals he talks to are genuinely embracing AI tools rather than just talking about them.

Forward Deployed Engineers: The New Critical Role

The biggest structural shift Lemkin sees isn’t AI replacing salespeople. It’s the rise of what he calls Forward Deployed Engineers (FDEs), borrowing the concept from Palantir but adapted for the broader SaaS market.

The old model: sales closes the deal, customer success onboards, the product may or may not work. The new model: someone technical gets the product working before the customer pays. An FDE at one AI company closed a $3 million deal entirely on their own, went on site, got the deployment running, tuned everything. Sales only managed procurement.

The ideal FDE is what Lemkin calls a “mediocre engineer who’s in love with the product.” Not someone who wants to write code all day, but someone with enough technical chops to deploy and tune an AI agent until it works. The mandate is simple: when the product goes live, it works. 100% success rate, not the 5% deployment success rate of 2024.

He recommends every startup hire at least four of these people. The term is borrowed from Palantir, but the execution is different. Palantir does nine-figure deals with armies of engineers on-site for months. For most SaaS companies, it’s about having a handful of people who ensure the AI agent actually delivers value from day one.

”People Person” Is the New “Golf”

One of Lemkin’s sharpest observations is about the death of relationship-based selling. He has a diagnostic test: ask a mediocre salesperson what they’re really good at. The answer is always the same: “I’m a people person.” They’re on text with their top 10 customers. They play golf.

His counter: ChatGPT is the best therapist on Earth right now. People spend hours talking to Lennybot. AI can be a “people person” too. If your only defense is that you’re good with people, the ground is shifting beneath you.

“If that is your best defense in sales, that you’re a people person, the sands are sinking beneath you right now.”

This matters because the buying process has changed. Customers now expect the AI agent to work during the pilot, before the big check comes. In that world, “people person” is insufficient. What matters is whether your product delivers ROI before the contract is signed. As Lemkin puts it, referencing a conversation with Marc Benioff: the dream is that every customer could go live before they pay. That flips the entire sales model.

The Incognito Mode Test

Lemkin’s most practical piece of advice: over the holidays, open an incognito browser, create a fresh Gmail address, and try everything about your own product. Sign up. Contact sales. Try support. Go through the entire customer journey as a stranger.

His prediction: you’ll cry. You’ll see how bad your support is, how slow sales responds, how broken some workflow has been since you launched it. Then pick the thing that makes you cry the most and go buy an agent to fix it.

This isn’t about publishing to production or getting CEO approval. It’s about building a proof of concept, showing your team what’s possible. The jaw drop alone is worth it.

What’s Changed, What Hasn’t

Lemkin breaks down the GTM landscape by function:

Already changed permanently: Customer support. Whatever vendor you look at, 50-80% of support is now handled by AI. This train has left the station.

Changing within 12 months: SDRs and inbound qualification. The cadence-based SDR and the lead qualifier are on a clear extinction path.

Still unclear: Field sales, cold calling, SMS outreach. Enterprise field reps are hiring more than ever. Cold calling remains a specialized craft that AI hasn’t cracked, though startups are pushing regulatory boundaries on AI-enhanced phone calls.

Net effect: more GTM professionals needed overall, because the winners are growing so fast that even with greater efficiency, they need more humans. Gartner projects next year will be the fastest acceleration of IT and software spending in a decade. But the composition of those teams is changing radically.

The Best Startup Is the One You Have

Lemkin’s closing message is unexpectedly grounded. He’s been watching founders quit companies with millions in revenue to chase hot AI startups, and he thinks most of them are making a mistake.

He cites three conversations from the past week: a founder who raised $20M and quit the next day, a CEO at $250M who left for a hot AI company, a founder crossing $100M who wanted to do a robotics startup. His advice to all of them: if you have happy customers, don’t quit. Take those 500 happy customers, build the AI product, and sell it to them. It’s almost always harder to start from scratch than you think.

“The best startup you’re ever going to have probably is the one you’re working at today. Don’t quit if you have happy customers.”

His addendum on books and playbooks captures the same spirit: the plays all work, but the playbooks are outdated. Every GTM book he’s been asked to review is too backward-looking for the AI era. Read the books, grab two or three individual plays, but don’t adopt the playbook wholesale.

A Few Observations

This conversation stands out because Lemkin isn’t speculating. He’s running an actual business with AI agents and reporting back from the front lines. A few things worth sitting with:

• The real disruption isn’t layoffs; it’s non-backfilling. Lemkin has never fired anyone. When people leave, agents take their place. This is the larger force of nature that’s harder to see in headlines.
• The 10% adoption problem is real. Most GTM professionals talk about AI but haven’t deployed a single agent. The gap between early adopters and everyone else is widening fast.
• “Deliver ROI before the contract is signed” is a fundamental shift in how enterprise software is sold. Companies that figure this out are growing explosively. Companies still playing the old game are growing 8%.
• Marketing is lagging sales in AI adoption, which Lemkin attributes partly to founder interest. Lots of founders want to build AI SDR tools. Nobody’s building the Cursor for marketing yet.
• The vibe coding thread is revealing. Lemkin can’t code but has built 12 apps on Replit in 150 days, used a million times. When agents started debating with each other about how to build better code, he “fell out of his chair.” The distance between non-technical founder and shipped product has collapsed.