We just did a deep dive with Lenny Rachitsky on what we’ve learned deploying 20 AI agents at SaaStr. And I want to share what I think is the single most important piece of advice for any sales, marketing, or GTM leader heading into 2025:
Pick one AI agent. Deploy it yourself. Do the training yourself. Do the ingestion yourself. Hands on keyboard.
Don’t hire an agency. Don’t tell someone on your team to “figure it out.” Don’t wait for a vendor to magically make it work.
Do it yourself.
Why This Matters More Than Any Other Resolution You’ll Make
Here’s what we’ve learned talking to hundreds of GTM leaders over the past year: about 80% of them are still panicking about AI. They’ve read everything. They use ChatGPT. They’ve seen the demos. They’ve maybe even bought a tool or two.
But they haven’t actually deployed anything themselves.
We literally just did a consulting call with a public B2B company worth well over $10 billion — a company you would think is an AI leader. We asked their team of 20 people on the call: “How much of this have you done yourself?”
Crickets.
They thought they could take an untrained agent with no training and just magically give it to a bunch of young 20-year-old SDRs and it would sell on its own.
That’s not how any of this works.
The 20% Who Will Thrive
The folks who will win in 2025 and beyond? They’re doing the work themselves. They’re not waiting for permission. They’re not delegating to agencies. They’re learning by doing.
Here’s the thing about AI agents that nobody tells you: the jargon is intimidating, but the actual work isn’t that hard. It’s just different.
Ingestion? That’s just uploading your stuff. Your website URL. Your wiki. Your training docs. Maybe your prospectus. The agent processes it.
Training? That’s just answering questions and correcting mistakes. Every day, when the AI sends emails or handles support tickets, it will say some dumb things. Maybe hallucinations. Maybe wrong dates. You correct it. Each day you spend an hour or two fixing those mistakes. By day 30? It’s pretty damn good.
Orchestration? That’s just managing which agents talk to which segments of your customer base.
Anyone who has been in B2B or SaaS can do this. It is not that different from other things we’ve done. It’s just sequenced differently.
The Incognito Mode Test
Here’s my challenge for you this month:
When you have a quiet moment — maybe over your morning coffee or during a slow afternoon — fire up your browser. Do it in incognito mode. Create a fresh Gmail address. And then:
- Try your own support
- Hit “Contact Me” on your website
- Sign up for your newsletter
- Try to buy something
- Go through your onboarding flow
Do everything a customer would do.
I promise you: you will cry about some of the things you see. You’ll cry about how bad your support response is. You’ll cry about how long it takes sales to get back to you. You’ll cry about some broken workflow you forgot existed.
Pick the thing that makes you cry the most. Go buy that agent and fix it.
That’s it. That’s your January assignment.
What We Learned Deploying 20 Agents at SaaStr
Let me share what actually happened at SaaStr, because I think it’s instructive.
We used to have about 8-10 people on our GTM team. SDRs, AEs, the whole thing. Two of our top people quit at our annual event this year. Just quit on site.
I turned to Amelia, our Chief AI Officer, and said: “We’re done hiring humans in sales. We’re done.”
Not because I hate humans. If I had two more great humans that wanted to join tomorrow, I would hire them. But I just couldn’t pay another junior SDR $150,000 a year to quit after three months without ever learning what SaaStr actually does.
So we pushed the limits with agents. And here’s where we are today:
1.2 humans + 20 AI agents doing what 8-10 humans used to do.
The productivity is about the same. It’s not dramatically better. It’s not worse. But it’s so much more efficient and it scales.
The key insight: agents work all night. They work weekends. They work on Christmas. They never quit because they got a better offer. They never show up to your standup claiming they’re “working on that deal with Vercel” when they haven’t done anything in 30 days.
The Training Reality Nobody Tells You About
Here’s what nobody tells you: training these agents takes real time. When I started with Deli (our digital Jason agent), I spent almost an hour a day training it in the beginning.
It started telling people the wrong dates for our events. I had to fix it. Every morning I would fire up the dashboard, review the issues, and answer the questions it got wrong.
I don’t have to do that anymore — it’s well trained now. But that first 30 days? It was work.
And here’s the thing: Amelia now spends 10-15 hours a week reviewing outputs from our 20 agents. It’s exhausting. Because agents never sleep.
This is not a good job for lazy people.
But if you’re willing to put in that initial effort? The compound returns are extraordinary.
How to Pick Your First Agent
Here’s my actual advice for picking your first AI agent:
1. Pick the problem that makes you cry the most (from your incognito mode test)
2. Pick a leading vendor — I don’t care which one. Artisan, Qualified, Agent Force, whatever. Just pick one that’s actually shipping product.
3. Pick the vendor that will help you the most — This is the critical insight everyone misses. These agents are all running on Claude or similar LLMs under the hood. They’re more similar than different. The difference is who will actually help you get it working.
When we were evaluating vendors:
- One argued with us and demanded $100K upfront before they’d help
- One was scared of SaaStr — didn’t want bad PR if it failed
- Artisan said “We’ll do it” and showed up to help
Guess which one we picked?
4. Get on the phone with their Forward Deployed Engineer — Before you write a check, talk to the person who’s actually going to help you implement. If they rock and the other vendor is “better” but won’t help you? Don’t do it.
The Specific Agents We Use (Go Copy Us)
Go to saastr.ai/agents — we’ve listed everything we built. Copy us. I mean it.
Here’s the quick rundown:
- Delphi — Our digital Jason/general agent. Started as a clone, evolved into support + sales
- Artisan — Outbound SDR agent. 60,000 emails sent. High response rates.
- Qualified — Inbound qualification. 24/7 response. Closed deals at 11pm on Saturday.
- Agent Force — Reactivation of leads that sales decided weren’t “worth their time.” 70% response rate on those “dead” leads.
- Momentum/Attention — RevOps. Tracks everything automatically in CRM. (One rep quit the day we deployed this — he hadn’t done anything in 30 days.)
The meta-learning: each one was hard the first time. But we got it down to a single prompt. Then we gave that prompt to Agent Force and it worked in a day.
The first agent is brutal. The second is easier. By the fourth, you’re the master of the universe.
For Junior Sales & GTM Folks: Embrace It
If you’re a junior salesperson or SDR reading this and thinking “oh no, my job is in trouble” — here’s your path:
Be the person on your team who embraces every bit of output from AI agents.
Is it annoying that the agent set up four calls for you when you only wanted two? Embrace it. You’ll be twice as productive.
Is it uncomfortable that every single thing you do is now automatically tracked in the CRM? Embrace it. (The people who fight this are the ones who haven’t been doing anything anyway.)
The days of working 20 hours a week and phoning it in for a few deals? Those are over.
At Owner.com, Kyle is targeting $3-5 million in revenue per rep with AI. A few years ago, that same role would have been $300-500K per rep. That’s an order of magnitude more efficiency.
The reps who can work with these tools? They become more valuable than ever.
The reps who fight it? They’re becoming obsolete.
The “People Person” Problem
Here’s how you can tell a mediocre salesperson in the AI age:
Ask them what they’re really good at.
“I’m a people person, Lenny. You know how good I am? I’m on text with 10 of my best customers.”
Okay, what are the toughest technical objections you face with your product?
…silence…
Being a “people person” is insufficient now. ChatGPT is arguably the best therapist on planet Earth. It’s a people person too.
If your only defense in sales is that you’re good with people, AI can do that. What AI can’t do (yet) is truly understand the product, handle complex negotiations, navigate enterprise procurement, and build genuine long-term relationships.
The bar has been raised. Technical competence + relationship skills is the new minimum.
What Will Actually Change in 2026 (and 2027)
Let me be direct about what’s changing:
Will be mostly extinct by end of 2026:
- Email-based cadence SDRs (the junior kid hired to send emails and respond to inbound)
- Human lead qualifiers (the BDR who makes you wait two days to schedule a call)
Still safe for now:
- Field sales / in-person sales (no idea how AI handles knocking on doors)
- Complex enterprise AEs (70% safe by end of 2025, but declining to 40-50% over time)
- Leadership roles (we haven’t produced an autonomous CEO yet)
The real threat isn’t layoffs — it’s not being backfilled.
Most AI-related job losses aren’t happening through layoffs. They’re happening because when people quit, companies are replacing them with agents instead of humans.
That’s what we did. We didn’t fire anyone. They quit. We just didn’t hire replacements.
It’s Time to Pick a Path
This is the most exciting time of our lifetimes to be in software.
I can’t even code, and I’ve built 12 apps on Replit in the last 150 days that have been used over a million times. I built a whole app where you can practice selling Harvey, Cursor, Replit, and ChatGPT Enterprise — and it works. This wasn’t even possible at the start of the year.
We’re running an eight-figure business with 3 humans and 20 agents.
But here’s the thing: it’s not less work. It’s more work. The agents are so productive you have to keep up. I’m working the hardest I’ve ever worked.
So pick a path:
Path 1: Embrace it. Deploy an agent this month. Do it yourself. Learn the skills. Become hyper-employable.
Path 2: If that’s not you, if you don’t want to be on this journey — no judgment. Go join a slower-growing company. They still need people. Just not as many.
But don’t pretend there’s a middle path. That doesn’t exist in GTM anymore.
Your Assignment for January
- Do the incognito mode test on your own product
- Pick the thing that makes you cry the most
- Pick a leading AI agent vendor for that problem
- Get on the phone with their implementation person — make sure they’ll actually help you
- Deploy it yourself — hands on keyboard
- Spend 30 days training it — an hour or two a day correcting mistakes
By February, you’ll have learned more about AI in GTM than 80% of your peers.
And you’ll be ready for whatever comes next.
The future’s coming anyway. We might as well embrace it.
