Building an AI-native organisation: YC's playbook for creating a shared organisational brain

Summary

Y Combinator's general partner Pete Koomen describes the internal AI infrastructure YC built over the past 18 months to transform a 50-year-old firm into an AI-native organisation. Starting from a concrete finance team problem—replacing hand-coded deterministic software with agentic tools that let staff build their own workflows—YC moved to a shared architecture: one centralised database, a growing registry of shared tools (now over 350), and self-improving skill loops that capture and refine the firm's operational knowledge. The playbook emphasises that superintelligence compounds through recording all organisational artifacts (meetings, decisions, emails), making that context accessible to agents, and operating in a high-trust, egalitarian mode where security is managed through transparency rather than lockdown. Koomen frames this as the "personal computer moment for AI": organisations that build open, user-facing agentic systems will outrun those that treat AI as a feature bolted onto legacy software. Read the full conversation at the YC Lightcone episode.

Key Claims

• YC started building internal agent infrastructure when its software engineers realised agents could let staff encode their own workflows in natural language instead of waiting for engineers to code purpose-built deterministic tools.
• Centralising all critical organisational context into one database—every company funded, founder, financial transaction, internal notes—allowed agents to answer arbitrary business questions, and removing the back-and-forth cost prompted staff to ask far more complex questions (Jevons paradox).
• YC now maintains over 350 shared tools in a registry; when tools live in one place, they become available both to internal agents and to Claude Code on individual machines, turning agent superpowers from solo into organisational scale.
• Self-improving skill loops run autonomously each night, reading past agent conversations to identify missed opportunities and context, then refining skills—e.g., a skill for writing two-sentence pitches has become better than Koomen at the task after learning from partner feedback.
• Most organisations resist by locking context down for "safety"; the path forward is recording all artifacts (meetings, emails, planning) and framing this as collective capability, not surveillance, which requires egalitarianism and trust-by-default as core traits.
• A new employee can reach productivity far faster because agents automatically show how staff actually do things; new hires effectively apprentice with the organisation's best people through AI.
• Organisations can choose a decentralised future where individuals control and program their own AI as an extension of themselves, or a centralised one where a few large firms gate advanced compute and AI; the coming wave favours the decentralised path.

Quotes

• "When you remove the back-and-forth between teams to get a thing done, you ask more questions."
• "It is denormalisation; you are taking data and putting it into a format optimised for the agent to ask questions. When you give the agents a soul and the data, and it knows you and what you care about, these things suddenly have insane wings."
• "Building superintelligence inside a company isn't about using AI as a co-pilot. That is the thing where you use it as the building layer for everything. You have to start recording all the artifacts."
• "The real idea was that the potential for AI is to shift control from the developer to the user. As we get better at building AI-native software, it will look like the agent wrapping deterministic tools, rather than deterministic software wrapping an AI."
• "You have to be willing to spend $10,000 to $100,000 a year on tokens. But if you do it and you invest in skills and operate in an open way, you can live in 2028. You can leapfrog every incumbent by doing this."
• "It is a shared organisational brain. It is the closest thing to us being able to connect our brains."