SaaStr 859: The $257 Employee: What Agents That Actually Work Look Like Right Now with Replit's CEO and Founder

SaaStr 859: The $257 Employee: What Agents That Actually Work Look Like Right Now with Replit's CEO and Founder

Everyone's talking about agents. Almost nobody is running them the way Jason Lemkin is.

At SaaStr, two AI employees - 10K, the autonomous VP of Marketing, and QB, the AI VP of Customer Success - ran a significant chunk of this event. They emailed 331 investors with individually researched, personalized outreach. They proactively contacted 100-plus sponsors and filed their own self-assessment of where they fell short. They built dashboards, ran campaigns, and tracked 10 years of attendee data without sleeping, complaining, or asking for a coffee break.

The tab: $257 a month on Replit. 

In this session, Jason sits down with Amjad Masad, Founder and CEO of Replit, to pull apart how this actually works - and what it tells us about where agents are headed next. 

You'll learn:

• Why monorepo architecture matters more than most people realize - and how putting everything in one place gives your agents global context they can't get when apps are siloed
• How Replit built a self-improving loop where an internal agent analyzes every user interaction nightly, generates prompt changes, ships them as A/B tests, and improves the product autonomously
• What "context compaction" actually means in practice, why Replit's is better than the generic version, and how to think about long-running agents that never reset
• Why the gap between agent hype and agent reality is at its widest right now - and what the next capability jump looks like by Q3/Q4
• Why Replit users were six months ahead of Silicon Valley on agents, and what that early edge looks like when you apply it to marketing, customer success, and operations

This is for you if:

• You tried an AI tool six months ago, it disappointed you, and you haven't gone back - this session will show you how much has changed
• You're a founder or operator trying to understand what a real, production-grade agent looks like versus a chatbot with a fancy name
• You're thinking about the economics of AI replacing or augmenting headcount and want a concrete data point, not a thought experiment