Y Combinator CEO Garry Tan released his personal AI coding setup — a collection of Claude Code prompt configurations he calls “gstack” — on GitHub under an MIT license on March 12. Within days, the repository accumulated nearly 20,000 stars and over 2,200 forks. It trended on Product Hunt. And it started an argument that hasn’t stopped.
gstack consists of 13 specialized “skills” that configure Claude Code to operate like a simulated engineering team. One skill evaluates startup ideas from a CEO’s perspective. Another writes code as an engineer would. A third reviews that code for bugs and security vulnerabilities. Additional skills handle design mockups and technical documentation. Tan has been actively expanding the collection since launch.
The setup debuted alongside candid remarks at SXSW, where Tan told venture capitalist Bill Gurley in an onstage interview that AI-driven coding has changed his work habits to the point of obsession. “I sleep like four hours a night right now,” Tan said, per TechCrunch’s coverage. “I have cyber psychosis, but I think like a third of the CEOs that I know have it as well.” He compared his AI-augmented productivity to past startup-building experiences: “It’s like I was able to recreate my startup that took $10 million in VC capital and 10 people, and I worked on that for two years.”
The Praise and the Backlash
Initial response was enthusiastic. The GitHub metrics speak for themselves. But the tone shifted when Tan posted on X a testimonial from a CTO friend who reportedly called gstack “god mode” after it discovered a cross-site scripting vulnerability, predicting that “over 90% of new repos from today forward will use gstack.”
That claim drew immediate fire. Sherveen Mashayekhi, founder of startup talent agency Free Agency, posted on X that “Garry should be embarrassed for tweeting this” and suggested the CTO in question “should be fired immediately.” YouTuber Mo Bitar released an analysis titled “AI is making CEOs delusional,” arguing gstack is “a bunch of prompts” in a text file. Mashayekhi also wrote on Product Hunt: “Garry, let’s be clear and honest: if you weren’t the CEO of YC, this wouldn’t be on PH.”
The criticism has a point. gstack is, technically, a curated set of system prompts. When TechCrunch asked AI models to evaluate it, ChatGPT described the skills as “reasonably sophisticated prompt workflows” but cautioned they’re not “magical,” noting the real insight was that “AI coding works best when you simulate an engineering org structure.” Claude itself — the model gstack runs on — called it “one of the better examples of Claude Code skill design out there.”
What the Debate Actually Reveals
The gstack controversy is a proxy for a much larger question: what does it mean to be a software developer when AI writes most of the code?
Tan represents one side of the answer. In a subsequent X post, he wrote: “I took modafinil just to stay awake longer to be able to turn the momentary crystalline structures I had in my brain into lines of code. I love coding but I love coding with AI even more.” This is the founder-as-composer view — the human provides direction, taste, and architectural judgment; the AI handles execution.
The opposing camp sees this as skill atrophy dressed up as productivity. If a CTO calls a prompt configuration “god mode,” what does that say about the CTO’s own ability to find an XSS vulnerability? If 90% of new repositories use someone else’s prompt templates, are developers building software or assembling it from pre-chewed instructions?
The tension is real: companies want AI-augmented productivity and confidence that their engineers can function without it. gstack forces the question — can you have both?
The YC Signal
Tan isn’t just any developer sharing dotfiles. As YC’s CEO, his public endorsement of a specific AI coding workflow carries weight across the startup ecosystem. The 2,200 forks suggest thousands of early-stage founders are now adopting his configuration — or adapting it — as a starting point.
YC’s latest batch, covered by Forbes this week, tilts heavily toward AI agent infrastructure: observability, reliability, and governance tooling. Respan, a YC-backed startup, raised $5 million today for proactive AI agent observability — monitoring agent behavior in production so teams can catch hallucinations and failures before users do. The batch signal and gstack signal point in the same direction: YC sees the future of software as AI agents supervised by humans, not humans writing code supervised by linters.
Whether that future produces better software or just faster software remains the open question. gstack, for all the noise, is a 13-file GitHub repo. The debate it triggered is about something much bigger than prompt engineering. It’s about whether the craft of programming survives the tools that automate it — or whether “craft” was always just a word for “the slow way.”