YC President Garry Tan: Build Your Own AI Agent Instead of Using Packaged AI

Garry Tan — president of Y Combinator and one of Silicon Valley's most influential voices — made a pointed argument this week on X: stop using packaged AI tools and build your own agent instead.

In a series of posts, Tan pushed the concept of personal AI software — open-source frameworks that give you full control over your agent, its memory, and its behavior — as a superior alternative to out-of-the-box products like ChatGPT or Claude.

Tan is promoting GBrain, an open-source MIT-licensed agent framework designed to work with OpenClaw or Hermes agent platforms. The code is available at github.com/garrytan/gbrain.

One of its key architectural ideas: separate your knowledge wiki repository from your core agent repository. According to Tan, this lets multiple agents use git to update a shared memory base — so your agent's knowledge lives independently from its operating system.

Tan also offered a concrete workflow: open your agent's own repository in Claude Code and manually code-review its commits. He calls this trust but verify — let the agent build itself autonomously, but stay in the loop on exactly what it changed and why.

This gives developers a way to audit agent behavior without micromanaging it, and stay connected to what the agent is building on their behalf.

The personal AI movement is a direct response to growing frustration with packaged AI tools: limited customization, data privacy concerns, and vendor lock-in.

For developers and businesses in Lithuania and across Europe — where data sovereignty is increasingly a concern — running your own agent on your own infrastructure is more than a technical preference. It's a compliance and ownership choice.

The GBrain approach represents a practical entry point: an opinionated, ready-to-use setup that doesn't require building from scratch.

As Tan put it: "If you are consistently hitting the limits of what the packaged AI software world can give you, then you owe it to yourself to try the personal AI software world." @garrytan, X

This isn't just a technical argument — it's a philosophical shift toward AI as a personal tool rather than a service. WebEdge.dev tracks these developments to help businesses make informed AI choices.