Coding in God Mode: The Truth Behind Garry Tan’s AI Setup
Y Combinator CEO Garry Tan is currently running on four hours of sleep and pure adrenaline. He calls it cyber psychosis. It is a state of mind where he is so obsessed with building with AI agents that he does not want to close his eyes. During a recent talk at SXSW, he joked that he is not the only one. He thinks a third of the CEOs he knows are feeling the same buzz. This is not about the old school startup hustle powered by drugs like modafinil. This time, the excitement is natural because AI is changing the math of how startups are built.
Tan says that working with AI agents today feels like having a ten person team and ten million dollars in capital at your fingertips. He is using these tools to revive old projects like Posterous and Posthaven while also managing his political site, GarrysList. To help others get on his level, he released his personal Claude Code setup on GitHub. He calls it gstack. It is an open source collection of “opinionated” skills that tell the AI exactly how to act during different parts of the coding process.
The setup works by simulating a real engineering team. Instead of just asking the AI to write a feature, Tan uses specific prompts to give the AI different roles. One skill makes Claude act like a CEO to vet a startup idea. Another makes it an engineer to write the code. A third skill turns the AI into a cynical code reviewer to hunt for security flaws. This modular approach is designed to catch mistakes that a single prompt might miss.
When Tan first shared gstack, the internet went wild. It trended on Product Hunt and racked up nearly 20,000 stars on GitHub. The love was immediate from people who wanted to replicate his workflow. However, the hate arrived just as fast. The drama peaked when Tan tweeted that a CTO friend used gstack to find a hidden security flaw in his company’s code. Critics pounced, calling the story delusional. Some argued that if a simple set of prompts found a major bug, the CTO should be fired. Others pointed out that gstack is really just a text file full of prompts that many developers already use.
Is it magical or just hype? If you ask the AI models themselves, they lean toward the former. ChatGPT noted that the real power of gstack is not in the words themselves, but in how they simulate a professional organizational structure. Gemini called it a “Pro” configuration that focuses on getting things right rather than just getting them done. Even Claude weighed in, calling it a mature system built by someone who actually uses the tool heavily.
Tan is not backing down from the critics. He recently posted that he loves coding with AI because it removes the friction between an idea and the finished product. To him, the experience of speaking and seeing his vision built instantly is the most powerful thing he has ever felt. Whether you think gstack is a revolutionary toolkit or just a collection of fancy prompts, it has clearly started a massive conversation about how we will build software in the very near future.