Gary Tan shares his experience returning to software development after years as an investor, highlighting how advanced AI tools and frameworks like GStack enable rapid, efficient, and sophisticated software creation by augmenting human capabilities through concepts like “token maxing.” He envisions a future where AI democratizes software building, empowering individuals to innovate and automate extensively, transforming productivity and creativity in the tech industry.
In this special episode of The Lyone, Gary Tan shares his remarkable journey of returning to software building after a 13-year hiatus spent as an investor. Despite running Y Combinator full-time, Gary managed to ship hundreds of thousands of lines of code and create popular open-source projects like Gary’s List, which combines blogging with high-quality investigative journalism powered by AI. His motivation stemmed from personal concerns, such as improving education access in California, and he leveraged advanced AI tools like Claude Code and OpenClaw to build sophisticated platforms quickly and efficiently, demonstrating a new paradigm in software development.
Gary emphasizes the transformative power of “token maxing,” a concept where developers leverage large-scale AI capabilities by burning more tokens to achieve deeper, more comprehensive results. This approach allows software to perform complex tasks like extensive research, cross-referencing multiple sources, and generating detailed, well-sourced reports far beyond what a human could do in a reasonable time. He highlights how AI can augment human agency rather than replace it, enabling builders to focus on high-level goals while machines handle repetitive or exhaustive tasks, thus accelerating productivity and innovation.
The discussion also delves into Gary’s creation of GStack, a skill-based AI development framework that automates many aspects of coding, testing, and quality assurance. GStack integrates various AI models, including Claude and Codex, to simulate roles like CEO, designer, and developer experience, streamlining the software creation process. Gary shares insights into balancing deterministic code with AI-driven latent space operations, stressing the importance of thorough testing (aiming for 80-90% coverage) to maintain reliability despite the inherent brittleness of current AI tools. He likens using OpenClaw to driving a Ferrari—powerful and exhilarating but requiring a skilled mechanic to maintain.
A significant theme is the democratization of AI-powered software development, where individuals can build and control their own AI tools rather than relying on opaque corporate platforms. Gary envisions a future akin to the personal computer revolution, where personal AI agents become ubiquitous, customizable, and integral to knowledge work. He acknowledges current challenges, such as the cost and complexity of using cutting-edge models, but encourages embracing these hurdles as investments in productivity and innovation, much like paying premium rent for proximity to valuable resources and networks.
Finally, Gary reflects on how his limited time as YC’s CEO forced him to innovate and automate extensively, turning scarcity into a catalyst for efficiency. He encourages others to adopt a similar mindset, leveraging AI to “borrow time” from machines and exponentially expand their capacity to build and create. This philosophy underscores a broader shift in technology and work, where human creativity and judgment combine with AI’s computational power to unlock unprecedented possibilities, marking an exciting era for builders and engineers worldwide.