YC Partner Diana Hu lays out the thesis that AI should not be a tool your company uses but the operating system your company runs on. Every workflow, decision, and process should flow through an intelligent layer that learns and improves. The talk is a 10-minute Startup School episode aimed at early-stage founders.
Key learnings
-
Closed loops over open loops. Old companies ran open-loop: make a decision, execute, rarely measure or adjust systematically. AI-native companies run closed-loop: every process captures information, feeds it back into an intelligent system, and self-improves. Borrowed from control systems theory.
-
Make your company queryable. Every important action should produce an artifact the intelligence layer can learn from. Record meetings, minimize DMs, embed agents in all channels, build dashboards across revenue, sales, engineering, hiring. If the AI can't see it, it can't improve it.
-
Software factories replace TDD. Humans write specs and tests that define success. Agents generate code and iterate until tests pass. Some YC companies already have repos with zero handwritten code, just specs and test harnesses. Strong DM's AI team is a cited example.
-
The 1,000x engineer is real. One person surrounded by a system of agents can build what used to require an entire team. The shift is from maximizing headcount to maximizing token usage. Run an uncomfortably high API bill because it replaces far more expensive human overhead.
-
Middle management disappears. If your company is queryable and artifact-rich, you don't need human middleware to route information. Jack Dorsey at Block is restructuring around this. Three archetypes remain: IC (builder-operator), DRI (strategy and outcomes), and AI founder (builds, coaches, leads by example).
-
Startups have a massive edge. No legacy systems, no thousands of people to retrain, no SOPs to unwind. You can design systems and culture around AI from day one. Large companies have to maintain live products while transforming, which is much harder.
Notable quotes
"AI should not be a tool your company just uses. It should be the operating system your company runs on." - Diana Hu
"Maximizing token usage, not headcount, will be the critical shift. The best companies will be the ones that are token maxing." - Diana Hu
Why it matters for me
This maps directly to how I think about building at Kaltura. The closed-loop concept applies to our sales process, engineering sprints, and customer feedback loops. The "queryable company" idea is basically what I've been doing with agents across Slack, CRM, and call recordings. Also validates the bet on running high API costs versus headcount.