The YC president open-sourced the stack he builds with. What it says about taste

Last week I was going through a repo that collects skills for coding, several of them. Most share one theme: helping AI write code in a systematic way, and faster.

But one made me stop longer than the rest, called gstack, for two reasons. One: its owner, Garry Tan, president and CEO of Y Combinator, took the stack he actually builds with every day and opened it for free. Two: it does not sell "code faster," it sells "review before you ship."

Once I actually opened it, it was not just a toolbox but one of the clearest examples of an idea I have been interested in for a while. On the day AI can write code very fast, the bottleneck of the work is no longer speed.

I will tell it in three parts, starting with what it is, then what gstack believes, and closing with lessons for people who build products, not just people who write code.

Part 1What gstack is

Garry Tan describes gstack in the README plainly, as the way he works.

"It turns Claude Code into a virtual engineering team: a CEO who rethinks the product, an eng manager who locks the architecture, a designer who catches AI slop, a reviewer who finds production-grade bugs, a QA lead who opens a real browser, a security officer who runs OWASP and STRIDE, and a release engineer who ships the PR. Twenty-three specialists and eight power tools, all slash commands, all Markdown, free, MIT-licensed."

Read it and you see it is not a single "coding assistant" but a team with clearly divided roles. What matters more than the headcount is that it forces the work through a multi-lens review before shipping. The CEO lens asks whether this is really a 10-star product or a 3-star one dressed up. The engineer lens asks whether the architecture holds up to the edge cases. The design lens asks whether you even know what "good" looks like. The devex lens asks how easily someone else can pick it up and build on it. Then it closes with steps to test, ship the PR, and watch after deploy.

Part 2What gstack believes

gstack does not just tidy up the tools. It has a stance, and that stance is the heart of what makes it interesting. Garry writes in the ETHOS file:

"The engineering wall has fallen. What remains is taste, judgment, and the willingness to do the complete thing."

He claims that this year he ships code hundreds of times faster than he did 13 years ago (his own figure, self-measured, not a central benchmark). But his point is not about speed. It is that once writing costs almost nothing, deciding what to build, and refusing to ship slop, becomes the whole job.

You can see it in the principle he calls Boil the Ocean. The old advice was "don't do the complete thing," because engineer time was expensive. But once that cost is gone, cutting corners and fixing later turns into debt. If the complete version takes only a few more minutes, do the complete thing.

The principle I like most, though, is that a person has to be the one who decides. gstack puts it this way:

"AI models recommend. Users decide." and "Two AI models agreeing on a change is a strong signal. It is not a mandate."

This is exactly why its multi-lens review pipeline matters. Each lens exists to challenge the work, not to rubber-stamp it, and the person is the one who chooses which challenge lands. Put another way, gstack is taste and the priority of decisions written into software. Most teams skip these lenses almost entirely; solo builders skip all of them. He brought them back as required gates.

Part 3Lessons for product builders

Even if you do not write code for a living, three things here are worth stealing.

• Give yourself review gates, even working alone. A team argues by nature; a solo builder has no one to argue back. Set up two or three review lenses and have the AI play each role to question your own work: the business angle, the user angle, the person who has to maintain it later. It really does catch what you overlook.
• Do the complete thing, now that completeness is cheap. What you used to cut short for lack of time now costs far less to finish. Don't ship a rough version when the complete one takes only a little longer.
• Keep the final call for yourself, especially when the AI looks confident or two models agree. That is the moment that tempts you to let go, which is exactly the moment to be most careful.

Where to start

You do not have to install all of gstack. Try borrowing just the ideas first.

1. On the next thing you are about to ship, have the AI play a single role, for example "review this as a CEO," and see what it asks that you had not thought of.
2. When you hit a spot you want to cut short, ask yourself how much longer the complete version really takes. If it is only a few minutes, do the complete thing.
3. Next time the AI is firmly certain, practice pausing to ask yourself first: do I agree because it is right, or because it is confident?

That Garry open-sourced his own stack is not a small thing. It is a signal from someone who has watched thousands of startups, that the bottleneck in building has moved, from "can you write the code?" to "can you decide what to build, and keep the slop out?" This is one in a series where I have been going through repos that pack expertise into skills. More are coming, and a pillar piece that draws the lines between all of them will close it out.

Read on in the series

• GTM Strategist skills by Maja Voje the whole go-to-market cascade broken into steps you can run.
• product-discovery-skills in the Teresa Torres vein continuous discovery turned into a repeatable pipeline.
• pm-skills by Pawel Huryn around 68 PM skills you can call with a single command.
• The AI product management skills guide the pillar that draws the lines across the whole series.