Garry Tan: Own Your Intelligence
Y Combinator Startup PodcastFull Title
Garry Tan: Own Your Intelligence
Summary
This talk draws parallels between the persecution of philosopher Baruch Spinoza for his radical ideas and the current approach to Artificial General Intelligence (AGI). It advocates for "personal AGI" as an owned asset that amplifies individual capabilities, rather than a rented corporate product.
The core argument is that by leveraging agents and owning one's own context and skills, individuals can achieve unprecedented levels of productivity and create new ventures, essentially "boiling the ocean" of what was previously impossible.
Key Points
- Spinoza was excommunicated and shunned for his unconventional beliefs about God and existence, demonstrating the historical cost of challenging established dogma and the temptation to silence dissenting voices with offers of comfort or threats.
- The concept of "personal AGI" is distinct from corporate AGI products, emphasizing ownership and continuous self-improvement based on personal data, contrasting with rented, externally controlled assistants that reset or are subject to company pivots.
- Individuals can achieve a 400x productivity increase by using AI agents for knowledge work, including coding, design, and product management, fundamentally changing the physics of individual output and startup creation.
- The power of AI agents lies not just in the models themselves but in the "context" and how it's provided, transforming individual capabilities from a "2x" to a "100x" level.
- Spinoza's concept of "conatus" (striving) is linked to the feeling of increased power and joy derived from using AI agents to accomplish tasks, while its absence leads to sadness and a feeling of decreased agency.
- The future of personal productivity and startups hinges on combining a rented frontier model with owned, unique personal context and a "harness" that connects them, creating an agent that acts as a hyper-efficient version of oneself.
- Building personal AI involves creating a "library" of one's own data and "skill files" which are executable instructions for agents, effectively turning individuals into managers of their own workforce.
- Markdown files are presented as the new form of "code," allowing anyone to program agents through clear English instructions, democratizing the creation of specialized tools and workflows.
- The distinction between computation happening in "latent space" (judgment, intuition) and "deterministic space" (arithmetic, queries) is crucial for building effective agent systems, with markdown files guiding latent space computations.
- Owning one's "skill files" is presented as the modern equivalent of craftsmen owning their tools, offering a path to freedom and preventing the extraction of one's accumulated judgment by employers.
- The rapid improvement of AI models makes personal context and proprietary data increasingly valuable, as these elements become the differentiators in a landscape where model capabilities are commoditized.
- The analogy of "Postgres is just B-trees" suggests that while retrieval is a primitive, the real value lies in the surrounding infrastructure of data enrichment, linking, and arbitration of conflicting information.
- Consolidating personal data into a self-controlled system is presented as a security advantage over scattered, externally controlled cloud services, emphasizing custody as the primary security model.
- The decision to open-source personal AI tools is framed as giving away the "tools of the powerful" to foster a renaissance rather than a priesthood, empowering individuals to build at an unprecedented scale.
- The core message is that the historical limitations of human cognition and resources have been overcome by personal AGI, allowing individuals to pursue ambitious projects without needing large teams or external validation.
Conclusion
The future of personal productivity and entrepreneurship is in harnessing "personal AGI" by owning your context, skills, and data, rather than relying on rented corporate solutions.
The historical barriers of needing large teams, funding, and permission to build have been largely removed by advancements in AI agents, allowing individuals to operate at vastly increased scales.
Embrace the new paradigm of building with AI agents by taking concrete steps to set up your personal library and skill files, and in doing so, secure your cognitive abilities and unlock unprecedented potential.
Discussion Topics
- How can individuals ethically and effectively build their "personal AGI" library without succumbing to data overload or privacy concerns?
- What are the potential societal impacts if the tools for massively amplified individual productivity and creation become widely accessible?
- In a future where agents handle much of the foundational work, how will the definition and value of human creativity and unique thought evolve?
Key Terms
- AGI
- Artificial General Intelligence, a hypothetical type of intelligence that could understand, learn, and apply knowledge across a wide range of tasks at a human level or beyond.
- Conatus
- In Spinoza's philosophy, the inherent striving of every being to persevere in its own being and increase its power of acting.
- Latent Space
- A compressed representation of data in a machine learning model, often used for tasks requiring intuition, judgment, or abstract understanding.
- Deterministic Space
- Computational processes that follow clear, predictable rules, such as arithmetic calculations or database queries.
- Skill File
- An executable set of instructions, typically written in markdown or a similar format, that an AI agent can follow to perform a specific task.
- Harness
- The framework or system that connects AI models with personal context and agents, enabling them to work together.
- RAG
- Retrieval-Augmented Generation, a technique that enhances AI model responses by retrieving relevant information from external data sources before generating an answer.
Timeline
Introduction of Baruch Spinoza and his excommunication for expressing forbidden thoughts, highlighting the historical cost of radical ideas.
Connection drawn between Spinoza's story and startups, emphasizing the importance of building and creating without permission.
Introduction of Spinoza's concept of "conatus" as the driving force of all living things.
Spinoza's heresy about God being inherent in nature is paralleled with the current misconception of AGI as a singular event.
The assertion that AGI is already present, not as a grand event, but as infrastructure and agents.
Definition of "personal AGI" as intelligence for an individual, distinct from marketable chatbots or assistants.
The concept of personal AGI as an asset that grows with use, unlike rented corporate AGI.
Explanation of why personal AGI is emerging now, tied to the capabilities of agents, particularly in encoding.
Personal example of a 400x productivity increase using AI agents for coding.
The productivity multiplier of AI agents applies to all knowledge work, not just coding.
Observation of a YC batch where 25% of companies had AI-generated codebases, leading to faster growth.
Spinoza's definition of joy as an increase in the power of acting, linked to the feeling of using AI agents.
The formula for the next decade: a rented model + owned context + a harness = a personalized agent.
Reinforcement of YC's core advice: make something people want and do things that don't scale, with agents amplifying the latter.
Comparison of human working memory (7 items) to AI agent memory (1,000 pages/tokens) and its implications.
The crucial question for an agent's intelligence is who or what decides which context is used.
Introduction of "Gbrain" as a system for managing personal knowledge and context.
Example of an agent pulling relevant past conversations and portfolio data to assist with a founder's crisis.
Description of an agent processing an inbox overnight, providing a briefing rather than just sorted emails.
Agent preparation for meetings, including context from previous interactions and relevant research.
Mention of G-Stack, an agentic coding framework, as a top open-source project.
The core components of the architecture: skill files, a controllable browser, and markdown.
Detailed breakdown of a "skill file" for processing meeting transcripts and extracting information.
The concept that clear English instructions are now a form of programming via language models.
The shift from being a programmer to a manager of agents.
The two types of computation: latent space (model-driven) and deterministic space (code/database).
Example of using latent space for seating arrangements and deterministic space for managing schedules for a large conference.
The creation of the talk's opening using an agent to synthesize biographies of Spinoza.
The "compendium skill" as a personal tool for deep research and knowledge synthesis.
The genesis of the talk's idea from a folder of markdown files and personal history.
Managing a workforce of markdown files, where skill files are employees and resolvers are organizational charts.
The ability to run an organization of one with agents, breaking traditional business math.
The new approach: scratch your own itch because building tools is nearly free, and these tools can become entire companies.
A caveat: a brain without curation is a garbage dump; memory needs hygiene and provenance.
Practical steps to start: pick a harness, run an agent, start a library, and write skill files.
The library grows with time and agent assistance, with the first click of using personal context being a turning point.
The process of writing a skill file by explaining a task in plain English to an agent.
Setting up skill files as recurring jobs, changing the unit of work from a day to what can be imagined.
The discipline of skillifying everything to avoid repeating tasks and build cumulative knowledge.
The trajectory of adopting personal AGI: week one (toy), week four (flywheel catches), week twelve (advanced library and skills).
The political aspect: skill files are externalized cognition, and their ownership determines the future.
Maya's story: owning her 40 skill files as her judgment versus them belonging to the company.
The doctrine: Own your skills, because if you don't, your job becomes a skill file.
Spinoza's refusal of a professorship with conditions as a parallel to declining offers that limit intellectual freedom.
The political question of who commands one's power of acting, personal AGI as staying under one's own power.
Spinoza's possessions as a contrast to his intellectual ownership and control over his ideas.
Objection 1: Models are improving too fast; the answer is that context becomes more important as models improve.
Objection 2: Is this just RAG? RAG is a primitive, not the product; the value is in what is retrievable from.
Objection 3: What about leaks? This is precisely why personal control and custody are essential.
The decision to open-source tools is to give away the tools of the powerful and prevent a "priesthood."
The creed: Say the things other people won't, fund the people other people won't, build the buildings other people won't, write and give away code other people won't, leave behind institutions other people won't.
Spinoza's final chapter: Leibniz visiting him and later writing extensively about him while downplaying the visit.
The "dunks" (criticisms) are part of the adoption curve; learn the pattern of building in public.
Example of a father building a personal AGI for his son with a rare form of epilepsy.
Nobody will build your personal AGI for you; the historical barriers have fallen away.
The mantra: "It's all made up, but you get to make it up."
The difference for founders today is the ability to build with a laptop and personal history, bypassing the need for many believers initially.
The machinery allows striving to go straight to work, one person, no intermediaries, no permission.
Spinoza's closing words: "All things excellent are as difficult as they are rare." The difficulty has collapsed, rarity is up to you.
Episode Details
- Podcast
- Y Combinator Startup Podcast
- Episode
- Garry Tan: Own Your Intelligence
- Official Link
- https://www.ycombinator.com/
- Published
- August 6, 2026