Y Combinator CEO Garry Tan told a SXSW audience that he has "cyber psychosis" and is barely sleeping because he's so excited working with AI agents. Speaking alongside fellow venture capitalist Bill Gurley during an onstage interview on Saturday, Tan said he's sleeping about four hours a night and joked that a third of the CEOs he knows are experiencing the same AI obsession.
Tan compared his current AI-powered productivity to his earlier startup days, saying he was able to recreate work that previously took $10 million in venture capital, 10 people, and two years of effort. He recalled using the sleep-preventing drug modafinil during those years but said the current AI revolution keeps him naturally awake with excitement.
What Is gstack?
On March 12, just two days before his SXSW appearance, Tan publicly shared his Claude Code setup on GitHub under an open-source MIT license. He called it "gstack."
The setup includes a collection of "opinionated" Claude Code skills — reusable prompts stored in special skill files that instruct AI how to behave in specific roles or tasks. For example, one skill has Claude act as a CEO to evaluate whether a startup idea is worth pursuing. Another makes it write code like an engineer, while a third reviews the code for bugs and security issues. Additional skills cover design, documentation, and other roles.
The repository currently lists 13 skills, with Tan frequently adding new ones and posting updates on X.
Viral Love
The response was immediate and massive. Tan's tweet went viral on X and trended on Product Hunt. The gstack repository has accumulated nearly 20,000 stars on GitHub with 2,200 forks — people who copied the files to modify for themselves.
The appeal is clear: Tan essentially packaged a workflow that treats Claude Code like a virtual engineering organization rather than a single assistant. Instead of asking AI to simply build a feature, gstack breaks the process into structured roles — ideation, development, review, and documentation — simulating how a real tech team operates.
The Backlash
But shortly after releasing gstack, Tan posted a tweet that caused a heap of criticism. He shared that a CTO friend had called gstack "god mode," claiming it instantly found a security flaw in his company's code and predicting that over 90% of new repositories would use it going forward.
The reaction from critics was swift and harsh. One founder responded on X saying Tan should be embarrassed for the tweet and that if the claim was true, the CTO should be fired immediately for missing such a basic vulnerability.
A vlogger published a take titled "AI is making CEOs delusional," pointing out that the project was essentially a collection of prompts in text files. The common complaint from developers was that anyone who already uses Claude Code has their own version of this kind of setup.
One Product Hunt commenter wrote that if Tan weren't the CEO of YC, gstack wouldn't even be featured on the platform.
What Do the AI Models Think?
In an interesting twist, TechCrunch asked AI models themselves to evaluate gstack. ChatGPT called it "reasonably sophisticated prompt workflows" that aren't magical, but noted that the real insight is simulating an engineering org structure rather than just asking AI to build a feature.
Gemini called the setup "sophisticated" and described it as essentially a professional-grade configuration focused more on making coding correct than making it easier.
Claude described gstack as a mature system built by someone who actually uses it heavily, calling it one of the better examples of Claude Code skill design available.
The Bigger Debate
The gstack controversy reflects a broader tension in the AI coding world. On one side, enthusiasts believe AI agents are fundamentally transforming software development. On the other, skeptics argue that the hype — especially from influential tech figures — often outpaces reality. Whether gstack is revolutionary or just well-packaged prompts, one thing is certain: the debate over how humans and AI should collaborate on code is only getting louder.
Tan did not respond to multiple requests for comment.
