Y Combinator’s CEO Garry Tan has become the poster child for AI obsession, and honestly, the internet is divided on whether that’s genius or delusion. At SXSW this past weekend, Tan told venture capitalist Bill Gurley that he’s running on basically no sleep, surviving on just four hours a night because he’s too excited about working with AI agents to actually rest. He even joked about having “cyber psychosis”, though given that AI-induced sleep deprivation and obsessive behavior is becoming a real concern for some, maybe it wasn’t entirely a joke.

Tan recently released something called gstack, an open-source Claude Code setup he shared on GitHub that bundles together six AI “skills”, essentially fancy prompts that tell Claude how to behave in different roles like CEO, engineer, code reviewer, and designer. The idea is that instead of just asking an AI to build something, you’re simulating an entire engineering org structure with different perspectives and quality checks built in. The setup went absolutely viral. Within days, it accumulated nearly 20,000 stars on GitHub and trended on Product Hunt.

But here’s where things get messy. Tan posted a tweet claiming that a CTO friend told him gstack was “god mode” and had instantly found a security flaw that the friend’s entire team had missed. This sparked immediate backlash. Critics pointed out that gstack is basically just a collection of text prompts, something most developers using Claude Code already do on their own. One founder called out Tan directly, saying if the CTO’s story was true, that CTO should be fired for missing such a basic security issue. YouTuber Mo Bitar even made a video titled “AI is making CEOs delusional”, and plenty of people on social media suggested that gstack only went viral because of Tan’s name and position, not because it was actually revolutionary.

When we checked with the AI models themselves, the verdict was surprisingly nuanced. ChatGPT called gstack “reasonably sophisticated prompts” but emphasized that the real value is the concept of structuring AI work like an organized team. Gemini agreed, calling it “less about making coding easier and more about making it correct”. Claude, unsurprisingly, gave Tan a thumbs-up, calling it “a mature, opinionated system”.

So what’s the real deal? Gstack isn’t magical, but it’s also not useless. It’s a thoughtful framework for working with AI that shows how structured prompts can improve code quality. Whether that’s worth the hype or just another product that blew up because a famous person attached their name to it remains an open question. What’s certain is that Tan’s enthusiasm, whether it’s genius or concerning, has definitely gotten everyone talking about how we should be using AI for coding.