Y CombinatorHow to Build Superintelligence Inside Your Company
At a glance
WHAT IT’S REALLY ABOUT
YC’s blueprint for organizational superintelligence: tools, context, transparency, trust
- YC’s internal AI stack began with a finance workflow pain point and evolved into a general agent harness with a shared tool registry that now contains 350+ organization-specific tools.
- A major unlock was giving agents read-only SQL access to a single Postgres database containing core institutional context, enabling non-technical staff to ask complex questions and triggering “Jevons Paradox” (more and bigger questions get asked).
- They argue legacy companies can move faster by consolidating context (or “denormalizing” data for agents) and standardizing how agents discover capabilities via resolvers/skill registries that stay DRY and MECE.
- YC is pushing beyond the “single-player era” of agents toward a “multiplayer” organizational brain by making agent conversations broadly visible internally, using transparency plus a high-trust culture as a governance mechanism.
- They describe a self-improving loop (“dream cycle”) where nightly agents review prior conversations and artifacts (e.g., transcripts) to refine prompts/skills—illustrated by a two-sentence pitch skill that improved via meeting transcripts and meta-prompting.
IDEAS WORTH REMEMBERING
5 ideasTreat AI as infrastructure, not a feature.
YC’s approach isn’t “add an AI button” to existing workflows; it’s building an agent harness, tool registry, and context layer that becomes the substrate for many workflows across teams.
Unified, queryable context is a force multiplier.
Agents became dramatically more useful once they could query a single Postgres database containing companies, founders, notes, and transactions; the agent can answer arbitrary business questions when the schema and models are accessible.
Lowering query cost increases organizational curiosity (Jevons Paradox).
When asking complex questions no longer requires hours of SQL or waiting on a data team, people ask far more questions—and attempt more complex analyses—raising decision quality and speed.
Build a shared tool/skill registry so agents can actually do work.
A tool registry turns generic LLMs into workplace-capable agents; as YC teams kept adding tools, they reached 350+ capabilities, reusable both in internal agents and local harnesses like Claude Code.
Use resolvers and keep skills DRY + MECE to prevent sprawl.
They highlight “Skillify” (turn repeated workflows into callable skills) and a “Check Resolvable” discipline to avoid many overlapping tools—prefer one well-parameterized tool over ten redundant ones.
WORDS WORTH SAVING
5 quotesPart of the key thing is not to just use AI as a copilot. This is the, the thing where you use it as the building layer for everything, and you need to start recording all the artifacts.
— Diana Hu
What if we just gave the thing, like, access, complete access to the production database where it could just, like, trample on anything? And I sort of, like, surreptitiously pushed it out maybe late at night.
— Jared Friedman
We have this general agent that every night will go and read through all of the agent conversations that employees have had and look for, uh, things it could have done better- and pieces of context that if it had up front, it would have done more efficiently.
— Pete Koomen
How do you build super intelligence inside a company? You do that on everything you do, and it's not more complicated than that. Like, you literally just compose everything that you do, and any given thing that any given person can do, you combine that in aggregate and in this particular process, and, like, you have a super organization.
— Garry Tan
If you want to create this type of organization, you have to be relatively egalitarian, and you also have to be trust by default.
— Garry Tan
High quality AI-generated summary created from speaker-labeled transcript.
