@garrytan: How does one engineer become a 1000x founder? @sdianahu and I give you the real goods here Thanks @AnjneyMidha for havi…

How does one engineer become a 1000x founder? @sdianahu and I give you the real goods here Thanks @AnjneyMidha for having us! https://youtube.com/watch?v=Lri2LNYtERM&list=PLoROMvodv4rN447WKQ5oz_YdYbS74M5IA…

TL;DR

Garry Tan and Diana Hu shared on Stanford CS 153 how YC standardized seed round financing with the SAFE agreement, and how in the AI era a small team can achieve exponential growth—from 10 people generating $20M in revenue over two years, to 6 people reaching $10M in just five days. The core is the “software factory” concept and tools like Gstack.

Opening: A Special Lecture

This session is part of Stanford CS 153 (Secure Large-Scale Systems), co-taught by Mike and the host starting four years ago. It draws inspiration from several landmark courses: Peter Thiel’s Zero to One (CS 183), YC’s startup class, and Terry Winograd’s Computers and Open Society. Garry Tan and Diana Hu return to the classroom as YC representatives—Garry himself is a Stanford ’03 grad who used to attend and even sleep in this very room. The host emphasizes that system design applies beyond engineering: capital and standards are also systems, and YC’s SAFE agreement is a prime example of such systems thinking.

The SAFE Agreement: Redefining Capital Allocation Standards

In 2011, the venture capital world was in a “pre-standard” state: chaotic, inefficient, and tough for founders to meet VCs and secure good terms. Paul Graham and Jessica Livingston introduced the SAFE (Simple Agreement for Future Equity)—a two-page legal document, offered free and publicly, defining how YC funds startups. Many dismissed it as just a piece of paper, but looking back, it became a pivotal turning point in Silicon Valley history.

The essence of SAFE is standardization, much like how AC/DC electrical standards and the power grid shaped the Industrial Revolution. It removed the capital bottleneck, allowing the abundance from the cloud and SaaS era (AWS, GCP lowering compute costs) to truly flow to innovators. The host remarked, “Without that document, Silicon Valley’s trajectory would be vastly different.”

The AI Era: From 10-Person Teams to 6-Person Teams, 1000x Efficiency

Garry shared his startup journey: joined YC in 2008, raised $4M, hired 10 people, spent two years building Posterous (a simple blogging platform), and eventually sold to Twitter for $20M. Today, the same software costs him $200/month for a Claude Code Max subscription, and anyone can replicate it in five days.

Diana added: many YC portfolio companies now go from zero to tens of millions in revenue within a year—traction that previously took four to five years to reach Series B level. It’s a completely different world.

Garry quoted Steve Yegge: engineers using AI coding agents are 10 to 100 times more productive; Anthropic’s internal engineers are 1,000 times more efficient than a 2005 Googler.

Gstack: From Code Generation to Software Factory

Garry started experimenting with Claude Code late last year, writing roughly a million lines of code (which some misinterpreted as “AI junk,” but he stresses it isn’t simple generation). He built Gstack, an open-source tool whose core is extracting “specific roles from latent space” to guide AI agents in generating high-quality code.

Key points:

• Test coverage must reach 80-90%: This is the core of the “plan-eng-review” skill. Garry uses this skill about 20 times a day, ensuring what’s delivered isn’t toy code but genuinely usable production code.
• Lines of code are a bad metric: The real measure is whether you and your customers are actually paying to use it. Gstack’s code emphasizes compactness and conciseness, avoiding deliberate verbosity.
• Gstack’s “Office Hour” skill: Distilled from transcripts of thousands of YC partner conversations over three to four months, then compressed by 90%, and finally open-sourced. It simulates YC’s office hour flow: What’s the problem? Who are the customers? How do you know? What are you building?

Garry’s personal stats: Gstack has 87,000 GitHub stars, Gbrain has 13,000, about 15,000 daily users, and hundreds of thousands of skill invocations.

Core Skill: Testing, Planning, and Review

Garry emphasizes that this is no longer the “copilot” era—it’s the “software factory” era. To combat AI hallucinations and bloat, you must:

• Default to preventing garbage output (LLMs tend to be verbose; control is needed).
• Achieve high test coverage to avoid shipping unusable things.
• Continuously iterate using the plan-eng-review process to hit 80-90% test coverage.

Conclusion: The Future of Founders and New Standards

Garry and Diana believe that while SAFE was a legal tool, the tools of today are code. The previous generation built the internet, mobile, and social networks; today’s students will build the cognitive layer for society. YC hopes students will put the “software factory” concept into practice and reshape society. They believe the people in this room will be the ones to set new standards.

Source: YouTube video (https://www.youtube.com/watch?v=Lri2LNYtERM&list=PLoROMvodv4rN447WKQ5oz_YdYbS74M5IA)