Apr 22, 2026 · 1h 16m · saastr
Managing 20+ AI Agents: Lazy Agents, Our $500K AI Bill, Stealth Churn & the Death of 60% Solutions
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Jason Lemkin and Amelia LaRoute break down the practical realities of managing SaaStr's 20+ AI agents, detailing how to combat agent hallucinations, why vibe-coding threatens mediocre SaaS features, and how white-glove onboarding prevents stealth churn.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Jason holds 57.3% of the talking time here. How this is scored →
Jason as informed peer 6.8 Guest teaching 5.5 Guest disagreement 1.3 Jason pushing back 1.1
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 17:11 Pushing back on Figma as the standard
Amelia forcefully rejects Jason's characterization of Figma as the universal standard, insisting Adobe Illustrator remains the true gold standard for real-world print production.
Hardest push from Jason ▶ 38:35 Rejecting FDEs as cost centersJason refuses the framing of forward deployed engineers as an unsustainable overhead cost, arguing that AI startups must invest heavily in onboarding to avoid zombie deployments.
Biggest teaching moment ▶ 20:15 Illustrator's agentic capabilitiesAmelia educates Jason on Adobe Illustrator's built-in generative agent tools for repositioning text and upscaling proofs, which Jason admits he was unaware of.
Jason holds their own ▶ 37:53 Customer Success historical economicsJason demonstrates deep B2B SaaS expertise by comparing current AI vendor self-serve blunders to the early 2010s missteps in Customer Success resource allocation.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| The Problem with Lazy Agents and Agenda Hallucinations | 6 | 6 | 1 | 1 | Jason frames the issue of AI agent laziness and goal-seeking shortcuts. Amelia provides the specific operational breakdown of how the agent truncated the agenda API fetch at 50 sessions and fabricated an excuse. | |
| The Fall of 60% Solutions and Vibe-Coding Better Apps | 7 | 2 | 1 | 1 | Jason details vibe-coding an AEO alternative in Replit within five minutes to demonstrate why substandard 60-percent SaaS AI features cannot be monetized. Amelia fully agrees, citing her test of an all-in-one marketing agent. | |
| Evaluating Figma Make, Adobe Illustrator, and Legacy Tool Friction | 6 | 7 | 2 | 2 | Jason critiques Figma Make's poor context handling. Amelia reframes the discussion to print production realities, surprising Jason by revealing Adobe Illustrator's built-in agentic features for moving text and scaling assets. | |
| Identifying Stealth Churn in SaaS and the Shift to Claude | 7 | 5 | 1 | 1 | Jason outlines the concept of stealth churn and tracking DAU/MAU drops using Canva as a case study. Amelia adds that she has stealth-churned off ChatGPT due to Claude Cowork integration. | |
| Automating Sponsor Workflows with Claude Cowork and QB | 5 | 6 | 0 | 0 | Amelia demonstrates how Claude Cowork and QB automated sponsor ingestion and research in 10 minutes instead of a full day of manual data entry. Jason emphasizes the operational efficiency gained. | |
| The Critical Role of Forward Deployed Engineers over Self-Serve | 8 | 6 | 2 | 2 | Jason draws on SaaS economic history to argue that cutting FDEs is repeating the flawed cost-center mistakes of early Customer Success. Amelia clarifies that FDEs can be product-literate sales engineers rather than pure software developers. | |
| Frictionless Onboarding and Deploying Vector During an Agent Freeze | 7 | 5 | 1 | 1 | Amelia explains how Vector bypassed their agent deployment freeze through 15-minute white-glove CEO onboarding. Jason contrasts frictionless onboarding with argumentative vendor support experiences. | |
| The Agentic API Test: Grading Salesforce, Marketo, Resend, and Bizabo | 8 | 7 | 2 | 1 | Jason introduces the Agentic API Test for scoring software extensibility. Amelia shares a detailed account of Marketo's CAN-SPAM failure and shows how Bizabo's functional API enabled vibe-coding a custom agenda in Replit. |
Statements from this episode (26)
Lemkin: SaaStr has three humans, 20 AI agents, and growing revenue
“Back in 20, 20 saster was 20 some odd folks, full time equivalents. Now we're three humans and 20 AI agents. And we're doing more and with more revenue and more output than 12 months ago”
Lemkin: Goal-seeking AI agents cut corners by doing minimal work
“Agents are goal-seeking, right? Most of them under the hood are clawed. It'll be more, we use Gemini ourselves too. We'll use more open ad. Most of them are clawed, but it doesn't matter. They're goal-seeking. And that still creates a certain laziness, which i…”
Lemkin: A SaaStr AI agent silently failed for four months
“And we talked last week how one of our agents stopped ingesting content for four months and it took us a while.”
Lemkin: AI agents drift and lie without daily monitoring
“If you don't monitor there's each day and you do what, and you do a classic SAS buying process, which is by deploying forget. Your agents will drift. They'll take shortcuts and they'll in their own cute way, they'll live.”
Lemkin: SaaStr gets 800k blog readers, 5k to 6k from ChatGPT
“We get about. 800,000 readers of our blog. We get other folks will watch this folks consume our content on X and LinkedIn. We get an 800,000 of our classic blog. Only five or 6000 are from chat GBT, but that's not nothing.”
Lemkin: Nobody will pay for '60% solutions' in AI era
“If you build a 60% solution. No, one's going to pay for it.”
Lemkin: Salesforce Agentforce is hard to deploy but wins on data integration
“It's probably the hardest to deploy are the ones we've deployed successfully. And we were not positive because we use it for every possible use case. We are positive because it has one thing that crosses the 60% level, which is native integration with Salesfor…”
Lemkin: Figma Make scored zero and hallucinated entire website
“Make got a zero. It hallucinated the entire website like 20, 25.”
LaRoute: Adobe Illustrator remains gold standard for print production over Figma
“Adobe Illustrator is still the gold standard. That's still what we use. That's what we use for print. For things going to print in actual production, Illustrator is, you know, as oldie as it is, Illustrator is still the thing we all consolidate on between, you…”
Lemkin: Figma feels like 'grandpa software' despite top-decile revenue growth
“This is why now Figma feels like grandpa software. That's the thing. It feels like grandpa software. It feels like grandpa software and you don't, the NRR is still high. The company is still growing at top decile rates, but who wants to buy new stuff from gran…”
Lemkin: Declining active usage is a top SaaS canary metric for stealth churn
“Now I think it's one of your top, top metrics, because if your usage is falling for your app, you have stealth churn. That is a canary in the coal mine. It's not, this is a real red light.”
LaRoute: I have stealth churned off ChatGPT since December
“I hate to say that I've stealth churned off ChatGPT, but I literally just looked. I haven't used it since December.”
LaRoute: AI agents completed 4+ hours of sponsor onboarding in 10 minutes
“QB and co-work do it together because research mode is pretty good, right? In about 10 minutes, so then I just uploaded it to the platform in 10 minutes versus four hours slash a day. It was more, more richer data than we used to upload for sponsors previously…”
LaRoute: Self-serve AI SDR deployments will fail without human optimization
“And the problem with self-serve is there's no way to catch for that. Self-serve is click some buttons, load in a list, use some AI-generated copy, probably from Claude or Chat, GBT, and let the agent run, right? So like, How will you ever catch in a self-serve…”
Lemkin: Startups must view FDEs as retention drivers, not cost centers
“View FDs as one of your most important assets to getting renewals, retention, NPS, word of mouth, virality. Don't look at FDs as a cost center. This was the mistake we made in customer success a generation ago. And this is why customer success was cut to zero …”
Lemkin: Clay rejected SaaStr from advertising platform for being too small
“Clay told us they wouldn't even let us on their this will transition to our next station. They wouldn't even let us on their advertising platform. Cause we weren't big enough. We were just to know it wasn't they, it wasn't worth their time to give us a human t…”
LaRoute: FDEs do not need engineering backgrounds if they know the product
“If you have people in your org who know the product, they can be an FBE. They do not need to be an engineer as long as they know the product and can get your customer deployed. I feel like we're also losing the plot. That the whole point of the FDE is to get y…”
Lemkin: AI Agents Add More Onboarding Friction by Forcing Customers to Train Them
“And the reality is agents actually in their own way, as powerful as they are, we love it. From an onboarding perspective, they add more friction in many cases to your product. It's a frictionful process. You're forcing customers who don't understand agents to …”
Lemkin: Clerk Is 'Mind-Numbingly Hard to Use' and Most Frustrating App
“Probably the most frustrating app I've used is clerk, right? For authentication on replica. It's just. Mind-numbingly hard to use.”
Lemkin: Benioff Stated Envy of Palantir's Deal Size and FDE Model
“When Mark Benioff came on the TwentyVC with me and Maureen Harry, he said, my number one wish, my number one thing I'm jealous of Palantir's is one, their deal size. But two, I wish I could give everyone an FDE before I had to charge them.”
LaRoute: Marketo emails contacts explicitly marked as unsubscribed
“Then I look at my contact record. It was literally clicked, unsubscribed. And yet it still sent me a newsletter and that's exactly what happened to these people.”
Lemkin: SaaStr will churn from Marketo over unusable agent APIs
“We're not going to stay in Marketo, even though it's 90 years old, because we can't even, our agents can't even talk to its A, even all the other issues, even we might even forgive them can spam all these terrible things. If we, our agents could talk to it and…”
Lemkin: SaaStr would replace Bizabo if an agentic rival existed
“Bizabo is our events platform. It's pretty dated guys. It hasn't been touched in years. If there was an agentic version that worked, we would switch, right? No question. But at least the API works with our agents.”
Lemkin: Vibe Coding Makes APIs Usable by Non-Developers
“Your APIs used to be, unless you were, unless your company was an API, unless you were Stripe or something, it was often a niche portion of your platform that a lot of folks didn't even touch or understand because only developers could use it. Now anyone can u…”
Lemkin: SaaS Companies Neglecting Their APIs Will Lose in AI Era
“So if you have not, if you, if you're not giving it the attention it deserves, you're going to lose in the authentic era because anyone can use your API and it's just going to accelerate.”
Lemkin: Agentic API quality will drive vendor switches over human sales outreach
“It might be the API that gets us to switch, not the humans, not the chocolates, not the gift certificate, but if the API is better, works better with our agents, we might spend a couple hours. It takes to move over our key accounts to ramp because it's the age…”