Peter Yang, who writes the Creator Economy newsletter and builds open-source tools for AI agents, argues that software is not dead but is getting harder to sell. He opens with his own experience: “When I left my job, I thought I’d vibe code a SaaS that made money from users while I slept.” Instead, he built free tools.

His thesis has two parts: “It’s harder than ever to make money from traditional software. Money is shifting to software that’s built for agents first.”

Why Yang thinks the pressure is rising

Yang points to several signs. He says a small team used Anthropic’s Opus model to rebuild seven Adobe apps, including PhotoCraft, a Photoshop clone that passed 45,000 GitHub stars, and is now cloning Word, Excel, and PowerPoint. He also built a Chrome extension to replace a $200-a-year YouTube tool, keeping only “the 20% that matters.”

On companies, he writes that “Retool found that 35% of enterprises have replaced SaaS tools with internal apps.” SaaS is subscription software. He also says Curative canceled a $600,000-a-year Salesforce contract after building an internal CRM, the system that tracks customers.

The examples vary in strength

The direction is believable. A $200 tool replaced by a homemade extension shows that simple single-purpose products are easy to copy. A clone with many stars shows interest, not paying customers. Retool sells tools for building internal apps, so its survey comes from a vendor with a stake in the trend.

The stronger lesson is about what stays hard to copy. A clone can reproduce features. It cannot reproduce a customer’s stored data, an audience, or the trust that comes from running something reliably for years. Builders pricing a product on its feature list are the most exposed. Those with data, distribution, or a service people depend on have more room.

Agent access as a requirement

Yang’s second claim is the more useful one, and he states it as his own buying rule: “If a piece of software doesn’t have an API or MCP that my agent can use, I probably just won’t use it period.” MCP is a standard that lets an assistant such as Claude call a product’s functions directly. He says he now uses Blender entirely through Claude and Blender’s MCP.

That is one heavy user’s habit, and his claim that money follows agent-first products is a forecast, not an established pattern. Still, if more work starts with an assistant asking a product to act, a product with no way for the assistant to do so gets skipped. Exposing your product’s core actions through an API or MCP server costs little and pays off if Yang is right.

End of article