Key Moments
Garry Tan: "Personal AGI Is How You Stay Under Your Own Power"
Y CombinatorKey Moments
Personal AGI, not corporate chatbots, will redefine startups by giving individuals unprecedented leverage, allowing solo founders to achieve what once required teams and funding. Own your intelligence, don't rent it, or risk becoming an extractable skill file.
Key Insights
Garry Tan claims he is now 400x more productive in coding than in 2013, attributing this to AI agents, even after applying significant penalties for verbosity and scaffolding.
In YC's Winter 2025 batch, a quarter of companies had codebases that were 95% AI-generated, and this batch is on track to be one of YC's fastest-growing and most profitable.
Personal AGI compounds daily as the agent learns more about the user's life, unlike corporate AGI which only improves when the company releases updates.
A skill file, defined as a page of English instructions, allows individuals to manage AI agents like a workforce; if a smart intern can follow it, an agent can run it.
Companies built with personal AGI are achieving unprecedented revenue per person, with one company hitting nine figures in revenue with only 15 people.
Own your skill files, as they represent your externalized cognition. If you don't control them, your job becomes a skill file that can be extracted without you.
Spinoza's heresy and the dawn of personal AGI
The talk begins by drawing a parallel between the 17th-century philosopher Baruch Spinoza, excommunicated for his radical ideas about God being inherent in the universe, and the current misconception of Artificial General Intelligence (AGI). Spinoza, ostracized and nearly killed for his 'evil opinions,' responded by grinding lenses to see further and writing a dangerous book, symbolizing a drive to build and express truth despite immense pressure. Garry Tan posits that AGI is not a singular, future event like a 'god in a data center,' but rather it's already here, diffused and accessible as personal infrastructure. He calls this 'personal AGI'—intelligence for one person, not for everyone at once. This is distinct from rented AI services like chatbots, which are corporate assets that can be reset or altered by their providers. Personal AGI, conversely, runs on your infrastructure, uses your owned memory, and compounds knowledge over time, making it a personal asset rather than a consumed product.
The 400x productivity multiplier
Tan illustrates the power of personal AGI by recounting his own productivity increase. He estimates he is now 400 times more productive in coding than he was in 2013, even after accounting for potential overestimation and verbosity. This multiplier effect isn't limited to coding; it applies across all knowledge work, including design, product management, and growth. He notes that at Y Combinator, a quarter of companies in a recent batch had codebases that were 95% AI-generated and are now among YC's fastest-growing. The key differentiator for these high-leverage founders is treating AI not as autocomplete, but as a workforce, emphasizing that the true leverage comes from the context provided to the AI, not just the model itself.
Compounding joy and the power of acting
Drawing from Spinoza's definition of 'conatus'—the striving in every living thing to keep going and increase its power to act—Tan connects this to the feeling of 'joy' when AI agents significantly amplify one's capabilities. Experiencing an agent complete a week's worth of work in an afternoon is described not just as a convenience, but as a direct increase in one's power of acting, a larger conatus. Conversely, he links the feeling of 'sadness' to a decrease in this power, akin to quiet quitting. This personal AGI framework is built upon three components: a rented frontier model (which is becoming a commodity), your unique and owned context (your personal data and history), and a harness that connects them. The crucial insight is that while model quality is rented, your personal context and the intelligence it generates are owned.
The library and the librarian: managing your working memory
Human working memory is limited to about seven items, a constraint that has historically led to the creation of prosthetics like checklists and org charts. In contrast, an AI agent can hold a million tokens—roughly a thousand pages—simultaneously. Tan emphasizes that the true power lies not just in the agent's capacity, but in who or what decides which 'three books' are open on its desk. This is where personal AGI, or 'GBrain' as he calls it, comes in. GBrain acts as both a vast library of your personal data (emails, notes, conversations) and a librarian that curates and synthesizes this information. This system allows an agent to act with the full context of your life's knowledge, transforming it from a simple assistant into a colleague.
Skill files as executable cognition
Tan introduces 'skill files' as the core mechanism for building a personal AGI workforce. These are essentially pages of plain English instructions that define a specific task or capability for an agent. If a smart intern could follow the instructions, an agent can execute them. This democratizes programming, enabling individuals outside of traditional tech roles, like media or finance professionals, to become 'managers of agents.' A skill file transforms a repeatable task into an executable employee. Tan argues that 'markdown is code,' with the language model acting as the compiler. This system allows for complex operations, like custom scheduling for thousands, by combining latent space computation (judgment, interpretation) with deterministic computation (arithmetic, database queries). The system's effectiveness relies on the careful orchestration of these two types of computation.
Building your organization of one
The framework described—skill files, resolvers (org charts for agents), and a growing library—allows individuals to build an 'organization of one' even before incorporating a company. This significantly breaks the old math of startups, enabling much higher revenue per person. Companies built on this model, like Emergent and Retail, have achieved nine-figure and $60 million annualized revenues respectively with remarkably small teams. The core principle is that software development becomes nearly free, allowing founders to 'scratch their own itch' with personalized tools. Tools built for oneself might become products that others demand, forming the basis of a new company. This shift means founders can now execute unscalable tasks at scale, fundamentally changing the physics of startups.
The political dimension: ownership of skills
Tan highlights the critical political implication: who owns the generated skills? If skill files reside in a company's repository, an employee's expertise can be extracted and retained by the company when they leave, leaving the individual with no lasting asset. Conversely, if skill files are owned by the individual, they take their compounded judgment and experience to new roles or ventures. This distinction is framed as the difference between having a career and undergoing extraction. The doctrine is to 'own your skills,' as the alternative is to have your job become a skill file controlled by others. This echoes historical parallels, where craftsmen owned their tools, and the factory system broke that ownership.
Taking custody and building in the open
Tan addresses objections, emphasizing that as models improve, the differentiator shifts to context (your personal library). He clarifies that personal AGI is more than just retrieval-augmented generation (RAG); it's about the entire ecosystem of data curation and management. Regarding privacy, he argues that consolidating personal data into a controlled, personal repo under one's own keys is more secure than scattering it across multiple cloud providers with misaligned incentives. He open-sourced his personal OS because powerful tools should be given away to foster a renaissance, not a priesthood. The final message is that the historical barriers to building—team, funding, permission—have collapsed. With a laptop and personal history, individuals can now build directly, unmediated, realizing their 'conatus' and creating organizations of one with unprecedented leverage, transforming personal striving into tangible creation.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●People Referenced
Building Your Personal AGI: A How-To Guide
Practical takeaways from this episode
Do This
Avoid This
Common Questions
Personal AGI refers to a general intelligence that runs on your own infrastructure, using your owned memory and executing procedures you define. It's distinct from corporate AGI, which is typically a rented service, lacks true ownership, and resets when you close the tab.
Topics
Mentioned in this video
A 17th-century philosopher whose excommunication and life choices serve as an analogy for modern tech principles, emphasizing self-reliance and intellectual integrity.
Referenced for his belief in Spinoza's God, highlighting Spinoza's philosophical influence.
Co-founder of Y Combinator, whose advice to 'make something people want' and 'do things that don't scale' is revisited in the context of personal AGI.
Canadian philosopher known for his theories on media and technology, quoted for his idea that 'technology is an extension of man'.
Co-founder of Apple, described as calling a computer a 'bicycle for the mind', relating to the idea of technology augmenting human capabilities.
The speaker, CEO of Y Combinator, discusses his vision for personal AGI and how individuals can leverage AI agents for personal and professional growth.
A harness or framework used to wire together AI models and context, presented as a key component for personal AGI.
A platform for software development where the GStack coding framework achieved significant popularity.
The startup accelerator where Garry Tan is CEO, serving as a backdrop for observing AI adoption and its impact on startups.
An AI agent framework mentioned as part of the 'harness' for personal AGI.
A large language model that can be used as part of the personal AGI harness, with its effectiveness depending on context and usage.
An AI system by OpenAI, mentioned as a potential component of the personal AGI harness.
A personal knowledge management system and AI agent framework developed by the speaker, designed to store and leverage personal data.
A relational database system mentioned as an example of a deterministic computing tool, contrasting with latent space computation in AI.
A version or feature of Claude focused on code generation, mentioned as an option for the AI harness.
A platform or tool from which meeting recordings are processed by AI agents for transcription and summary.
More from Y Combinator
View all 619 summaries
37 minBuilding the First Data Centers in Space
50 minWaymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Work
31 minPatrick Collison: "What If You Succeed?"
58 minJeff Dean: The 1% Rule for Building in AI
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free