The Agent That Ate the App
The agents came for the app layer this week, and the people who build software said it out loud.
The clearest tell came from SaaStr. Jason Lemkin and AI chief Amelia Todorova now run the business with three humans and 21 agents, revenue up 47% year over year, and they spent the week narrating what that actually looks like. Claude wired into Replit as an "AI VP of product." A ten-year Marketo-to-Salesforce migration, 200 campaigns, that an agent ran for about $14 in model cost against a vendor quote of a year and $100,000. A $10,000-a-year app killed and rebuilt in an hour. Their point was not the savings. It was the threat. "The risk isn't that you'll vibe code it yourself. The risk is the agent says I can build it." When they needed a data vendor, the Replit agent picked one and they never evaluated a competitor. Whatever that vendor's rival was, it lost the deal without knowing there was one.
That is the quiet part. Distribution is moving to the agent. The buyer stops shopping.
Glean's Arvind Jain, on 20VC, drew the enterprise version of the same line. Big customers are "terrified" of the frontier model providers, not because the models are bad but because handing your workflows to someone else's agent hands over your institutional learning too. His read: enterprises are pushing toward open source to keep control, and the consumption pricing AI forces breaks the seat-based bundling SaaS was built on. You do ten times the work to earn the same revenue.
Put the two together and the week has a shape. The build layer is nearly free, so the moat moves. It is no longer the app. It is the data, the context, and the customer relationship the agent has not captured yet.
If you sell software, the operators worth hearing this week are not debating whether agents are real. They are pricing in what happens after.
Listen:
- SaaStr — How Agents Will Steal Your Customers (Jul 17). The three-humans, 21-agents build diary. Start here.
- The Twenty Minute VC — Why OpenAI and Anthropic Won't Win the App Layer, Glean's Arvind Jain (Jul 11). The enterprise control-and-pricing argument.
As AI agents increasingly replace software workflows and control buyer decisions, SaaS companies must compete on data and customer relationships rather than app features.