Replacing Your Sales Team with AI Agents (Jason Lemkin, Founder & CEO @ SaaStr)

The Death of the Mediocre Sales Rep

I’ve been saying for months that most AI use cases being pushed on LinkedIn are hype and BS. I’ve also been saying that you should be making AI a priority anyway. That paradox has frustrated a lot of revenue leaders who want to know what’s actually working.

So when Jason Lemkin told me he’s deployed 20+ AI agents across SaaStr’s go-to-market functions—and that they’re outperforming his human reps—I knew we needed to have a real conversation. No demos. No sponsored takes. Just raw lessons from someone who’s built portfolio companies like Salesloft, Pipedrive, Talkdesk, and Intercom.

What emerged was one of the most honest discussions about AI in GTM I’ve had. The takeaways challenged some of my own assumptions and reinforced others. Here’s what I walked away with.

---

1. The Catalyst: When Human Reps Become the Bottleneck

Jason’s journey to AI agents didn’t start with a strategic initiative. It started with frustration.

“In May 2024, two of our high-paid sales reps—$150K to $200K plus—quit without notice going into SaaStr Annual,” Jason told me. “Amelia and I just looked at each other and said, ‘We’re done with this. We’re going all in on agents.’”

The decision wasn’t just about turnover. It was about chronic underperformance despite premium compensation. Jason described reps who wouldn’t follow up on LinkedIn job changes even when explicitly told to. Inbound leads sitting untouched. Basic blocking and tackling being ignored.

The lesson here isn’t about AI replacing humans. It’s about AI replacing mediocrity. When your highest-paid performers can’t execute fundamentals consistently, the calculus changes. Jason’s frustration mirrors what I hear from revenue leaders constantly—the gap between what we’re paying and what we’re getting has become untenable.

This connects to something I’ve observed at Owner: the best performers are getting exponentially better, while the middle of the pack struggles to keep up. AI is accelerating that divergence.

---

2. The Uncomfortable Truth: Mid-Pack Jobs Are in Terminal Decline

Jason didn’t mince words on this one: “Mid-pack jobs in GTM are in terminal decline.”

Here’s what he means. SaaStr now sends 70,000 hyper-personalized emails—compared to 7,000 that humans sent before. The AI-generated emails are better than what the humans produced. Not dramatically better, but consistently better. And they generated 15% of SaaStr London’s revenue.

“The agents are better than a mid-pack AE or SDR or BDR,” Jason explained. “They’re not better than your best performers. But that middle tier? They can’t compete.”

This has profound implications for compensation and career trajectory. Jason believes elite performers should earn 2-3X more—similar to how some OpenAI engineers earn millions. Traditional sales had no leverage mechanism; now it does.

The profile of who thrives in this new world looks different. It’s not the BDR who becomes an agent manager. It’s more like a growth PM or data scientist who becomes a GTM engineer. Jason recommends looking internally for “the nerdy SDR with a math degree who loves data.” External hires for “Senior GTM Engineer” roles are likely to disappoint because the role is too new to have an established talent pool.

At Owner, we’re seeing this play out. We’ve achieved 3X productivity improvement per AE and are hiring 25 people in 75 days—a perfect illustration of Jevon’s paradox. Higher productivity doesn’t mean fewer jobs; it means more investment in what’s working.

---

3. The Implementation Playbook: You Are the Agency (For Now)

Here’s where Jason’s advice diverged sharply from what most vendors are selling. When I asked about how to get started, he was emphatic: “Do it yourself. You are the agency for now.”

He means this literally. The CRO, CMO, or VP Sales needs to personally deploy the first agent. No delegating. No bringing in agencies. Twenty to thirty hours over 30 days of hands-on work.

“If you don’t roll up your sleeves in the age of AI GTM, you will become obsolete,” Jason warned.

The training process he described is surprisingly manual:

1. Point the agent at your database (or just your website initially)

2. Upload your perspectives and documentation

3. Agent generates 10-30 sample questions or emails

4. Review every single email for 1-2 hours daily

5. Correct hallucinations immediately (e.g., “Owner is for 100-chain restaurants” → “Owner scales but core is 1-5 locations with significant to-go business”)

6. Agents remember corrections via hive mind—they improve daily

The key insight: deployment failures are training failures, not product failures. Jason shared a story about Momentum (one of my favorite tools) that initially seemed broken. Turns out, a sales rep had never linked their Google account. When Jason showed them the link on a live Zoom, the rep reluctantly connected—revealing 30 days of zero activity. The rep quit that day.

“Shame on us,” Jason said. “Pre-Claude 4, products genuinely didn’t work. Post-Claude 4, all products are above the line. Failures are implementation failures.”

---

4. The Vendor Selection Framework: FDEs Over Brand Names

When I asked how to choose between the explosion of AI vendors, Jason offered a framework I haven’t heard elsewhere. He evaluates on a 2x2 matrix:

1. Quality of the vendor

2. Quality of the forward-deployed engineer (FDE) support

“Don’t sign a contract until you talk to your forward-deployed engineer,” he advised. “Agents require weeks of training before going live. Better to have a worse vendor with a great FDE than the best vendor with no support.”

His first agent was Artisan—not because they were “best,” but because they offered full deployment support. A competitor demanded $100K upfront with no deployment help. Jason walked.

For companies just starting, Jason recommends inbound AI as the lowest-hanging fruit. “There is no excuse on planet Earth for some 21-year-old SDR to qualify me whether I’m worth their time.” Tools like Qualified (bought yesterday by Salesforce!) handle instant answers, automatic scheduling, ICP filtering—24/7, any language.

The selection process:

1. Pick ONE tool solving a medium/high priority problem

2. Demo and research

3. Critical: Say “I want to talk to my FDE”—bypass the sales rep

4. Talk to the forward-deployed engineer before signing

5. Budget $50K-$100K for the first tool

Don’t be intimidated by the jargon. “Training” means uploading data and answering questions. “Ingestion” means data upload. “These are all SaaS apps,” Jason reminded me. “Just with an LLM hookup.”

---

5. Where We Actually Are: Bottom of the First Inning

Despite all the progress, Jason was remarkably humble about the current state. “We’re in the bottom of the first inning for innovation in AI and GTM. I don’t even think we’re in the second inning.”

The 70,000 emails SaaStr sends are better than human emails—but only a little better. True hyper-personalization still uses just 1-3 data points. Token costs remain high for comprehensive data synthesis.

Jason’s vision of what’s possible: “AI should send emails as good as Adam’s investor cold email”—the one that got Jason to invest in Owner. That email represented top 0.1% IQ with several hours of crafting. Future AI should pull every competitor, adjacent product, website visit, and 10-year interaction history, then synthesize the entire database in real time.

“AI is smarter than us now,” Jason said. “It should leverage the full hive mind.”

The products that achieve this vision will be massive winners. But we’re not there yet.

---

The Hard Truth About What This Means for You

I asked Jason what advice he’d give to sales professionals navigating this transition. His response was characteristically direct:

Be self-aware about your growth appetite. Want venture scale with triple-triple-double-double growth? Prepare for the hardest work of your life. Want lifestyle? Join a company growing 10-20% and burning $0. “Be fine with less upside.”

Step up or find a new role. Mid-pack performance equals obsolete. You must be 5-10X more productive than you were a year ago. Elite performers will earn 2-3X more.

Roll up your sleeves. Personally deploy at least one agent. This isn’t an agency game yet—maybe in 24 months. You can’t bring the 11 agencies from your last job.

When I reflected on my own experience, I realized I’m working as hard as I ever have—despite all the tailwinds at Owner. Jason echoed this: “I’m working the hardest I ever have.” The magical middle ground of 2020-2021 is gone. You could have had lifestyle, growth, and low hours. Now it’s either high intensity or slow-growth lifestyle companies.

---

What I’m Taking Away

This conversation reinforced something I’ve been thinking about for months: the gap between AI hype and AI reality is closing fast, but only for leaders willing to do the work.

The companies winning with AI aren’t the ones buying the flashiest tools or hiring the most agencies. They’re the ones where revenue leaders personally deployed the first agent, spent 30 days training it, and built institutional knowledge about what actually works.

At Owner, we’ve built 9 high-impact AI production use cases and achieved 3X productivity improvement per AE. But I’ll be honest—we’re still in the early innings too. The infrastructure we built in 3 weeks (that others said would take months) is just the foundation.

The question isn’t whether AI will transform GTM. It’s whether you’ll be the one doing the transforming—or the one being transformed.

---

If this resonated, I’d love a 5-star review on Apple Podcasts or Spotify. It helps us reach more revenue leaders navigating this same transition.