Y Combinator CEO Garry Tan described a state of intense, AI-fueled focus during a SXSW interview, joking about 'cyber psychosis' and surviving on four hours of sleep. His excitement stems from a personal project: an open-source collection of prompts for Anthropic's Claude Code, which he calls 'gstack.'

Shared freely on GitHub, gstack structures the AI into specific roles—CEO, engineer, code reviewer—to simulate a full engineering team. The reception split the tech community. It quickly gained nearly 20,000 GitHub stars and trended on Product Hunt, with many praising its practical workflow.

The backlash was just as swift. After Tan quoted a friend calling gstack 'god mode' for finding a security flaw, critics called the claim exaggerated. Some argued the prompts were unremarkable, and the attention was due solely to Tan's position. Vlogger Mo Bitar labeled the hype a sign of 'CEO delusion.'

We asked the AI systems themselves. ChatGPT noted the value was in simulating an org structure, not magic. Gemini called it a sophisticated 'Pro' configuration for correctness. Claude approved, calling it a mature system from a heavy user.

Tan, who once used stimulants to code, now says the AI collaboration is the powerful experience keeping him up. He sees the structure, speaks, and it builds. The debate continues on whether gstack is a genuine advance or a spotlighted set of text files.

Source: TechCrunch