A deep dive with Jason Lemkin (Founder & CEO, SaaStr) and Amelia Ibarra (Chief AI Officer, SaaStr), hosted by Kraig Swensrud (Founder & CEO, Qualified)
At Dreamforce 2025, we sat down in the Qualified Studios to talk about what’s actually working in AI agents—not the theory, but the real deployments, the real revenue, and the real mistakes.
Jason founded SaaStr in 2015 and has built it into an eight-figure B2B events, media and community business. He’s also the solo GP of SaaStr Fund (managing $200M across 3 funds with portfolio winners like RevenueCat, Owner, and Gorgias). Amelia now serves as Chief AI Officer this year and manages 20+ AI agents across SaaStr’s entire go-to-market stack.
This isn’t a conversation about experimenting with ChatGPT. This is about deploying agents that book meetings, sell tickets, answer support questions, and generate real revenue—often while the team sleeps.
Here’s what we learned.
Top 5 Learnings
1. Training Matters More Than Picking the “Perfect” Tool
The mistake: Spending months evaluating vendors, agonizing over Replit vs. Lovable, or which AI SDR platform is “best.”
The reality: Just pick a leading vendor and go deep on training.
“I would rather you pick a leader in any space—support, marketing, sales, coding—and just train it,” Jason said. “Train it up front for 30 days. Commit the time. Every day for the next 30 days, train it. Then every week after that.”
Most agentic products won’t work without training. The bar has completely shifted from the old SaaS model where you’d buy from a sales rep, hire an integrator, and wait a year hoping it works.
Time to value in AI has to happen before you even sign a contract.
If you deploy an agent yourself, train it, QA it, and test it, you’ll be ahead of 90% of the world.
2. Start With Layup Roles, Not Hero Purchases
The mistake: CMOs looking for the “hero purchase” they can show their CEO to prove they’re “doing AI.”
The reality: Hero purchases almost always fail because you’re trying to improve something that’s already sort of working, the bar is high, and you don’t know how to train agents yet.
“If AI is only going to make something 20% better, maybe 2025 is not the year,” Jason said. “Maybe that’s a 2026 project.”
Layup roles are the places where literally no one is doing the work:
- Support that takes a week to respond (or never responds)
- Outbound SDRs who won’t send emails
- Qualification that relies on “fill out this form and hope someone gets back to you”
- Geographies or functions that are impossible to hire for
For SaaStr, outbound was easier than inbound to start with. “It was less risky to send a better email than a mediocre human than to trust an agent with a precious lead,” Amelia explained.
Find what’s broken in your org. Deploy there first. Get a win.
3. Stairstepping Is the Path: Simple → Complicated
The approach: Start with the simplest possible use case, get confidence, then go vertical with specialized agents.
SaaStr started with Deli—a horizontal “Digital Jason” agent that ingested 20+ million words of SaaStr content. It required minimal training (just content ingestion) and gave founders instant access to Jason’s advice 24/7.
“It was the easiest possible deployment,” Jason said. “And we gained confidence.”
Once they had that early win, they deployed vertical agents:
- Artisan for outbound (SDR work)
- Qualified for inbound (BDR/qualification)
- Finn for support
- Momentum for deal intelligence and call summaries
- Agent Force for Salesforce workflows
Each vertical agent required more setup time, better data quality, and deeper training. But because they’d already learned how to train agents with Deli, they knew what to expect.
“If you fail with your first agentic thing, not only is it frustrating—you won’t know,” Jason said. “You won’t know if it’s the tool, your training, or the use case.”
4. The Cognitive Load Is Higher, But the Output Is 10x
The reality: Managing agents is more mentally demanding than managing people.
Amelia spends an hour every morning managing SaaStr’s agent stack. She checks Qualified for overnight meeting bookings (usually 5-10 for their London event while Pacific time sleeps). She reviews Artisan’s outbound sends. She QAs support conversations from Deli. She monitors Momentum’s deal summaries.
That same hour used to be spent with human SDRs and BDRs:
- 1-on-1s with reps who maybe followed up with leads half the time
- Reviewing accounts they should be working
- Explaining context about prospects
- Dealing with turnover, bad hires, politics, and feelings
“The agents don’t cry,” Jason said. “If someone’s going to cry in a meeting with you, you just have to sit there for an hour. But you don’t have to think. You don’t have to use all your brain cells. With agents, it’s so many brain cells.”
But here’s the thing: It compounds.
When a rep leaves, you lose all that training and context. When you train an agent, it never leaves. It gets better every day. It works 24/7. And it scales infinitely.
“Instead of talking to two humans for 30 minutes each, I can now do a lot more in that one hour with my five core agents,” Amelia said.
5. Everyone Is In-Market Again (Just Like 2020)
The context: Traditional SaaS budgets are frozen. CEOs are cutting 20-30% of their apps. Half of incremental IT budgets are going to price increases from incumbents.
But AI budgets are exploding.
“Traditionally, you’d hope a prospect would be in market every 5-6 years for a big buy,” Jason said. “Now everything’s in market again. This has never happened since maybe when the web started.”
The only incremental budget in most companies is AI. No one is putting more budget into old SaaS software.
If you’re not tapping into AI budget, you’re dead. But if you are? Business software is growing faster than it has ever before.
“Budget for e-signatures and contact centers in 2020 just went 10x because everyone was at home,” Jason said. “We’re not like that today, but there’s so much money in ‘I need to go AI today.'”
For buyers: Everyone’s looking for new vendors. Leaders are emerging but haven’t hit $1B ARR yet. Time to value expectations have completely changed. Just get going.
For sellers: If you don’t have an answer to “What can you do in AI that’s 10x better than before?”—you get an F.
The Halloween Deadline
Here’s where Jason drew a hard line:
If you’re a startup and you don’t have a disruptive AI agent in production by Q4 2025, you’ve already lost.
“Every VC I know—every investor I know—where the startup hasn’t made the jump yet, they’ve given up,” Jason said. “They’ve all lost hope.”
You could get away with being skeptical in 2023. You could have a principled position in 2024 (“This stuff doesn’t work yet”).
But in 2025? Everything is great now.
“Even mediocre software just got 5x better when you put Claude 4 in,” Jason said. “Replit took 10 years to get to $1M ARR. They add Cloud 4, now they’re at almost $200M in one year. Same IDE. Same software.”
If you’re a startup and you haven’t shipped something by Halloween, you need a brand new team. “Maybe 80% of them have to go. Give them a nice Thanksgiving bonus and a turkey and tell them they have to go. Because you’ve had 10 months.”
For enterprises and larger companies? It depends.
“If literally your business is fine—if your growth is as good or better than 12 months ago, you’re in a traditional industry, everything’s working—you can be a late adopter,” Jason said.
But if you’re feeling any external pressure? If competitors are moving? If your board is asking about AI?
Rip the band-aid off.
Real Results at SaaStr
Let’s talk numbers:
- Qualified booked 7 meetings autonomously in its first week of deployment
- 100+ tickets sold to SaaStr London in 6-8 weeks (out of 2,000 total attendees)
- 50%+ of inbound conversations happen overnight while the Pacific time team sleeps
- One rep quit the day Momentum was deployed (it exposed zero activity via automated Slack summaries)
- Zero contact form abandonments (instant AI response vs. week-long human wait times)
“Half the time when I wake up, most of the conversations for London have already happened because when we wake up in Pacific time, it’s their afternoon or evening,” Amelia said.
The agent isn’t just helping. It’s selling tickets. It’s booking meetings. It’s closing deals.
And critically: People trust it now.
“People will walk up to me at events and say, ‘You’re that Amelia AI,'” Amelia said. “I’m like, ‘Yeah, I’m Amelia. That’s my AI you talked to.'”
The line between human and agent is blurring for customers—and they’re okay with it, as long as they get value.
The New Job Description
Here’s the uncomfortable truth for go-to-market executives:
If you haven’t deployed an agent yourself by now, you’re unemployable in 2026.
Jason shared two stories that illustrate this:
Story 1: The CRO Who Panicked
“I recently had a CRO I’ve worked with for years—the deepest respect—reach out to me in an almost panicked email. ‘I need to learn AI. I’ll do anything. I’ll intern for you. I’ll hang out in Amelia’s office.'”
Jason’s response: “No, you need to DO AI. Don’t learn it. Pick an agent, pick the simplest possible use case, deploy it yourself, train it, QA it, test it. If you do that, you’ll be ahead of 90% of the world.”
Story 2: The CMO Who Got Nothing
“Another CMO who I love—I’ve gotten her her last two jobs—just reached out after many years. She’s ready for her third job. And I’m like, I got nothing for you. I was really direct. ‘Can I be honest with you? I got nothing for you. You don’t know this stuff yet.'”
Jason told her: “Go out and deploy an agent. Tell me how it worked. Tell me how the training went. You come back to me, I’ll get you two jobs. But I got nothing for you now.”
And then he added this:
“There is much more demand for CMOs than they think. I know 100 CEOs today that for a great marketer, they are desperate. They’ll do anything for a great marketer. But they do not want someone running the Marketo 2019 playbook. There’s just no interest in hiring that person.”
The skills that matter now:
- Deploying agents hands-on (not delegating to your team)
- Training LLMs and agentic workflows
- Understanding time to value in AI tools
- Managing cognitive load across multiple agents
- Knowing when to stair-step vs. when to go deep
“Buying a subscription to Claude does not count as learning AI,” Jason said. “Own the deployment. You don’t know what training is? It’s okay. Take your time. Do something simple.”
What About the Sales Team?
One of the most powerful moments in the conversation was when Amelia shared advice she gave a founder at Dreamforce:
“She was asking about our agents. She’s like, ‘I have some agents, but I’m worried about rolling this out to our sales team. I’m worried about sales specifically. I don’t know how our 50 sellers are going to react. I think they might revolt or some of them might quit.'”
Amelia’s response: “That’s okay. Let them.”
“It’s almost Halloween. It’s okay. It’s nearing the end of 2025. If your team can’t get over the hurdle, some of them may quietly leave, and that is okay.”
She continued: “You can’t forsake the future of your company and how much faster it could grow with AI booking meetings, doing outbound for you, just to not hurt people’s feelings. You have to rip the band-aid. You got to go for it.”
The right framing for sales teams:
“Show them that they can be empowered with AI. They can have more use cases. They should actually be excited because they can crush their quota with AI. They can book more meetings. They can have better conversations. They can 10x themselves as AEs once they get over that fear.”
But if they hide in the corner? “Give them a turkey and let them weed themselves out.”
The best sellers will want AI because it means more money. More quota attainment. More commission. More wins.
The ones who don’t? They were already hiding from the work anyway.
SaaStr’s Top 5 Mistakes (So You Don’t Make Them)
1. Not Starting With the Simplest Possible Use Case
What we did wrong: We almost started with something too complex right away because we were excited about the technology.
What we learned: Our horizontal Deli agent—which just ingested content with minimal training—gave us the confidence to go vertical. If we’d started with a high-stakes, complex deployment, we might have failed and given up.
Your move: Pick the absolute simplest agent deployment you can. Support chatbot. Outbound emails. Something where the bar is low and you can get a win fast.
2. Expecting Agents to Work Without Training
What we did wrong: In the early days, we thought “AI-powered” meant plug-and-play.
What we learned: “Training is more important than picking the perfect vendor,” Jason said. Every single agent required 30+ days of intensive training up front, then weekly maintenance after that.
Even the “self-ingesting” tools like Deli required us to upload our prospectus, clean our data, and teach it when to escalate.
Your move: Block out 30 days on your calendar. Every single day. That’s what it takes to train an agent properly. If you’re not willing to do that, don’t buy the tool.
3. Deploying Agents Before Fixing Our Data
What we did wrong: We assumed our Salesforce data was “okay.”
What we learned: “I thought we had okay data quality in Salesforce, and then I was like, no, it’s terrible,” Amelia said.
Agents expose bad data instantly. When we hooked up Momentum (which Slack-notifies summaries after every sales call), we discovered one rep had zero activity. They quit that day.
Your move: Audit your CRM data before deploying agents. Clean up duplicates, fill in missing fields, standardize naming conventions. The agent will surface every flaw—better to fix it proactively.
4. Trying to Boil the Ocean Instead of Stair-Stepping
What we did wrong: We wanted to deploy everything at once after our early wins.
What we learned: “Without another human, this is about the most we can do,” Jason said, referring to their ~12 core agents. “The bar is so high for us to add a 13th core tool now.”
Each agent requires daily management. Amelia spends an hour every morning across all agents. More agents = more cognitive load.
Your move: Go from 0 → 1 agent (horizontal). Then 1 → 3 agents (vertical specialization). Then 3 → 5. Stair-step your way to a full stack. Don’t try to deploy 20 agents in month one.
5. Looking for the Hero Purchase
What we did wrong: Early on, we almost fell into the trap of wanting to buy the “coolest” AI tool to show off to our community.
What we learned: “Hero purchases almost always fail,” Jason said. “You’re solving medium-hanging fruit because it’s already sort of working. The bar is high. And you don’t know how to train it because it’s your first agent.”
Instead, we focused on layup roles:
- Outbound (broken: our SDRs wouldn’t send emails)
- Inbound qualification (broken: contact forms with week-long response times)
- Support (broken: ghosting customers half the time)
Your move: Find what’s literally not getting done in your org. Don’t try to make something that’s working 10% better. Make something that’s failing 100% better.
The Bottom Line
Classic SaaS is dead. Or at least geriatric.
But B2B software? It’s exploding.
“AI is putting more CIO budget and probably more SMB budget into business software than ever before,” Jason said. “Traditional budgets are frozen. But all the energy is in AI-first solutions.”
If you’re selling the way you sold in 2021, you’re done. But if you’re tapping into AI budget, building agents, training workflows, and moving fast?
You’ll win everything.
The companies that move fastest in this window are going to dominate the next decade. The ones that wait—that “learn” instead of doing—are already gone.
So here’s the challenge:
By the end of this month, deploy one agent.
Pick a layup role. Something broken. Support. Outbound. Qualification.
Pick a leading vendor. Don’t overthink it.
Spend 30 days training it. Personally. Hands-on-keyboard.
And then see what happens.
Because that’s how you’ll actually understand this. Not by reading. Not by “learning.”
By doing.
Want to see how we’re actually deploying these agents at SaaStr? Come to SaaStr AI London on December 1-2, 2025. We’ll have speakers from OpenAI, Intercom, Finn, Clay, and more sharing real playbooks—not theory. Plus, you’ll get to talk to the actual Amelia (and her AI). Grab your tickets at SaaStr AI London before sell out.
