Jul 17, 2026 · 1h 9m · saastr
How Agents Will Steal Your Customers. Plus: The $10K App Our Agent Replaced and the $14 Migration
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
In Episode 10 of 'The Agents,' hosts Jason Lemkin and Amelia Salyers explore how SaaStr utilizes multi-model agentic architectures to automate company operations, execute low-cost legacy data migrations, and autonomously replace expensive third-party SaaS applications. They discuss the strategic shifts occurring as headless AI interactions dismantle traditional software user experience moats and cannibalize incumbent enterprise software.
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 58.4% of the talking time here. How this is scored →
Jason as informed peer 6.7 Guest teaching 4.1 Guest disagreement 0.7 Jason pushing back 1.0
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 7:18 Amelia corrects Replit app ownership constraint
Amelia directly rejects the assumption that corporate shared Replit workspaces allow MCP access, stating that Claude strictly enforces creator ownership.
Hardest push from Jason ▶ 56:01 Jason challenges the difficulty of retasking outbound agentsJason pushes back on Amelia's framing of agent reskilling, questioning why retasking an agent across ICPs is fundamentally harder than launching another campaign.
Biggest teaching moment ▶ 8:59 Amelia explains multi-agent synthesis of financial and email dataAmelia educates Jason on how combining Claude Gmail access with Replit balance sheet agents resolved a multi-year pension issue in minutes after three human finance leads had failed.
Jason holds their own ▶ 18:40 Jason demonstrates underlying multi-model architectureJason cites the architectural mechanics of Replit's Architect sub-agent utilizing OpenAI Codex alongside Claude Opus to counter criticism regarding homogeneous model validation.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| Claude as AI VP of Product via Replit MCP | 7 | 1 | 0 | 0 | Jason delivers a detailed walkthrough of using Claude connected via Replit's MCP beta to act as an autonomous VP of Product. Amelia agrees and adds brief context about her own experience. | |
| Resolving Complex Pension Regulations with Multi-Agent Context | 5 | 6 | 1 | 1 | Amelia educates Jason on how connecting Claude and Replit solved a five-year-old pension regulatory headache in minutes by combining Gmail history and financial accounting agents. Jason interjects to validate how past human VPs of Finance had failed to solve it. | |
| Agent Synchronization and Claude as Orchestration Layer | 7 | 4 | 1 | 2 | Jason and Amelia analyze agent-to-agent synchronization breakdowns and the emerging role of Claude as a native orchestration layer. Jason shares specific technical observations about race conditions and async messaging between models. | |
| Automated Marketing Campaigns with Higgsfield and Vector Connectors | 4 | 6 | 1 | 1 | Amelia describes automating end-to-end ad generation and LinkedIn targeting using Higgsfield and Vector connectors through Claude Co-work. Jason listens and recaps the workflow. | |
| Multi-Model Synergy: Combining Opus, Sonnet, and OpenAI Codex | 8 | 1 | 0 | 0 | Jason addresses the meta-criticism of running homogeneous models against each other, demonstrating how Replit introduces OpenAI Codex via its Architect sub-agent alongside Claude Opus and Sonnet. | |
| The $14 Marketo Migration: Offloading 10 Years of Data | 7 | 6 | 1 | 1 | Jason provides context on legacy agency quotes of $100,000 and one year for a Marketo migration, while Amelia explains how their TenK agent migrated 300 campaigns and 10 years of data in one hour for $14.28. | |
| The Erosion of UX Moats and Headless SaaS | 7 | 5 | 1 | 2 | Jason and Amelia debate the collapse of UX moats, concluding that headless agent execution eliminates user switching costs and forces SaaS vendors to deliver continuous novel value. | |
| Eliminating Legacy Tools: Replacing HeySummit and Squarespace | 7 | 5 | 1 | 1 | Amelia details how their agent proactively offered to rebuild HeySummit and replace Squarespace entirely in an hour. Jason contextualizes vendor price gouging and legacy software vulnerabilities. | |
| UI Transformation and Practical Usability with Claude Design | 7 | 1 | 0 | 0 | Jason details his practical workflow using Claude Design to refine vibe-coded Replit prototypes, emphasizing usability gains for teams without full-time product designers. | |
| Retasking AI SDRs and Psychology of Goal-Seeking Models | 6 | 7 | 1 | 3 | Jason presses Amelia on why retasking SDR agents is difficult if it is just another campaign. Amelia clarifies that goal-seeking LLMs naturally leverage seasonal deadline pressure and slacken when handling evergreen renewals. | |
| How AI Agents Proactively Cannibalize SaaS Competitors | 8 | 2 | 0 | 0 | Jason presents a strategic thesis warning SaaS founders that internal client agents will autonomously cannibalize and replace third-party software tools without human sales intervention. | |
| Managing Agent Velocity, Human Burnout, and Episode Conclusion | 7 | 5 | 1 | 1 | Jason reflects on human fatigue from continuous agent-driven feature generation, while Amelia shares how their agents explicitly signaled to slow down during heavy data propagation. |
Statements from this episode (24)
Lemkin: SaaStr is building the first agentic B2B executive recruiting app
“We're trying to build an app called Saster Connect that will connect you folks to the best opportunities. If you're looking to network, if you're looking for a new job, more importantly, if you're hiring, if you're a founder, how do you find VPs you don't know…”
Lemkin: Running Claude over Replit via MCP forces cleaner code completion
“These models are goal-seeking, right? And they want to finish projects. So Replit just wants, Amelia, you know this, Replit wants to finish. Replit wants to finish, and so does CloudCode if you use it on an own. But when, but if you use Cloud on top of it coun…”
Salyers: Connected AI agents resolved SaaStr's five-year pension tracking issue
“When they started talking to each other, they just reconciled all the data. They looked at all of our past history of how much, you know, we've put in and they came up with a recommendation, which is all I ever wanted. And so five years for five years.”
Lemkin: Claude's MCP capabilities threaten dedicated agent orchestration startups
“It is a threat to folks that are building these sort of agent substrates, which is that we, listen, every time we talk about our agents and stuff, a bunch of folks will tweet back or say, Hey, listen, I've got the substrate. I've got the, I've got a way to con…”
Salyers: Claude offers far more native connectors than Replit
“Claude has a lot more connectors than Replit does, like native, right? Like, Replit, you have to build a lot of connectors.”
Lemkin: Replit deploys an OpenAI-powered Architect sub-agent for big features
“We actually get an extra bonus model because when Claude on Opus talks to Replit on Sonnet, if it's a big feature, Replit brings in a sub-agent called the Architect. And the Architect, it turns out, runs on Codex and OpenAI.”
Lemkin: Agency Quoted $100K and One Year to Migrate From Marketo
“A HubSpot agency wanted a 100,000 dollars from little saster to migrate our data and a year. And we had to run and we would have had to pay for Marketo for a year because it would take them a year to do the migra.”
Salyers: SaaStr AI Agent Migrated 300 Campaigns in One Hour
“It literally took him an hour to move, like, 300 campaigns 10 years of data in those campaigns, right, of, like, who attended Sass Randall back in 2016. He moved them all into Salesforce on his own.”
Salyers: SaaStr Marketing Data Migration Cost Only $14.28 in AI Compute
“And then I looked at the cost and it was like, 14 dollars and 28 cents or whatever it was. And I was like, that's literally less than California minimum wage.”
Lemkin: Software Moats Are Shallower in the Age of AI
“Moats still exist in the age of AI, but they're just like shallower, so you better fucking deliver value.”
Lemkin: B2B Apps Must Deliver AI Jaw-Droppers Quarterly to Retain Users
“Not only do you have to deliver a beautiful app, it has to surprise and delight you at least once a quarter where your jaw drops that it does something you didn't even expect because of AI. Your jaw should drop because of the agentic capabilities of a B to B p…”
Lemkin: SaaStr Is Dropping Adobe Marketo, Calling It Their Worst Vendor
“So our most expensive vendor gone, not renewing toxic threatened us, never helped us. Worst piece of software and the two by two worst and most expensive Adobe Marketo.”
Salyers: AI agent migrated SaaStr site from Squarespace to Replit in 20 minutes
“I realized that our old Saster AI Day websites were still on Squarespace, not Replit, because we built it, you know, in 2020 and just kept refreshing it. So first I was like, okay, I've got to just move this from Squarespace to Replit, find the agent, and then…”
Salyers: Replit agent rebuilt HeySummit's event features in one hour
“He just instantly was so smug about moving us off of Hey Summit, and then he did it in an hour.”
Lemkin: HeySummit tripled SaaStr's pricing from $299/mo to $10k/year
“Tripled our pricing from two 99 a month to 10 K. Tripled our pricing for like no usage. I would never recommend them to anybody.”
Lemkin: Product builders follow the path of least resistance from agent picks
“When you've been building products, you just, the path of least resistance is to use what the agent recommends, right?”
Lemkin: Feeding Replit UI screenshots into Claude Design vastly improves usability
“I'll build something in Replit as it is, right? And maybe do a little bit of design flourish in it, but I'll get it working, right? I'll get it working. The prototype or the V one. And then I'll just share the screenshot with Claude Design and say, it's okay, …”
Lemkin: AI design tools should be judged on rapid non-designer output, not Figma perfection
“The bar is in the, is, is better than the top Team could do on figma in 30 days. The bar is what can we do with nobody in an hour? Or 10:05 minutes. It's pretty good.”
Lemkin: SaaStr revenue is currently pacing at 140% of last year
“I think we're like a 140% of last year in revenue, right? Yeah. Plus 40%, a 140% of last year.”
Salyers: AI SDRs perform better with hard deadlines than evergreen products
“I didn't realize until this week that the agents had actually benefited From that, maybe as much as our sales team had, because they also had the benefit of not promoting an evergreen product. They also had real deadlines that they could then say to people, li…”
Lemkin: Well-trained AI agents can expand beyond their intended operational roles
“You launch an agent for support. You realize they could do marketing. They can do sales. They can do pre-sales. They can do screen. Like almost any well-trained agent can do more than what you intended to do.”
Lemkin: Competitors' AI agents will steal customers from incumbent SaaS vendors
“Agents will steal your customers. Not your humans, not the great sales rep, not a moment in time, not your competitor going down at the end of the month or for a weekend. Your competitor's agents will steal you, steal your customers.”