Skip to content
YC Root AccessYC Root Access

Building And Structuring An AI Native Company

In this presentation from Startup School Paris, Y Combinator General Partner Tom Blomfield shares findings from his research on how companies are being built with AI infrastructure placed in from the start. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs Chapters: 00:00 — Intro 01:53 — Why Roman Legions Built Your Org Chart 03:56 — Humans as the Bottleneck 05:28 — What a Real AI Loop Looks Like 07:51 — The Data Agent That Changed Everything 08:39 — The Self-Improving System 10:32 — Office Hours → Living User Manual 12:22 — The AI Employee With a VM 14:08 — What "Company Brain" Actually Means 15:52 — Humans at the Edge 17:44 — Burn Tokens, Not Headcount 18:29 — Make Everything Legible to AI 19:29 — Simulating Investor Calls

Aug 14, 202621mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Design AI-native firms with self-improving loops, not hierarchy layers

  1. Traditional org charts mirror Roman Legion hierarchies where humans route information; AI enables coordination without layered management.
  2. Most companies “bolt on” AI as Q&A or gated agents, but AI-native design treats the company as multiple closed-loop systems that sense, act, quality-check, and learn continuously.
  3. YC examples show self-improving loops in practice, like a data-query agent plus an overnight agent that fixes failures by submitting pull requests.
  4. Making work “legible to AI” (recording/transcribing meetings, reducing private channels, producing artifacts) turns organizational knowledge into a queryable, improvable system.
  5. Humans shift to the “edge”: handling high-stakes judgment, trust, culture, and real-world interactions, while the system handles routing, iteration, and optimization at scale.

IDEAS WORTH REMEMBERING

5 ideas

Stop designing companies around human information routing.

Hierarchical layers historically existed because humans were the only reliable coordination mechanism; AI can route information, run tasks, and escalate exceptions without managers acting as conduits.

AI-native means closed-loop automation, not “ChatGPT in the sidebar.”

The big shift is from one-off answers or human-gated agents to systems that ingest real-world signals, take actions via tools, check quality, and learn from outcomes continuously.

Replace human approval gates with machine quality gates where possible.

Instead of waiting for a person to unblock work at 3 a.m., use adversarial/second-model checks (e.g., prompt-injection detection, policy compliance, code review) and only escalate the truly high-stakes cases.

Build self-improving infrastructure that fixes itself overnight.

YC’s “agent-on-top-of-an-agent” pattern—reviewing daily successes/failures and submitting PRs to resolve issues—turns tooling from static productivity boosts into compounding improvement.

Turn tacit advice into a living, queryable knowledge base.

By transcribing office hours and extracting recurring guidance, YC can auto-update its internal manual and then serve consistent, multi-partner-quality advice via an agent that recalls everything.

WORDS WORTH SAVING

5 quotes

I think the main caveat I wanna start with is no one knows how to do this.

Unknown

There's an underlying assumption that organizations have to be hierarch- hierarchically organized with humans as a coordinating mechanism. And I think basically AI breaks that apart.

Unknown

If you can do this entire loop without a human, your product starts improving itself when you're sleeping.

Unknown

And then that second AI agent overnight goes and puts in pull requests to fix all of the problems from the day before. And so if you go back as a human on the second day and run the same query as yesterday, it now works.

Unknown

I'd burn tokens, not headcount.

Unknown

Roman Legions as the org-chart analogyHumans as coordination bottlenecksAI loops: telemetry → policy → tools → quality gates → learningSelf-improving agents that generate pull requestsOffice hours → living user manualAI employee with a virtual machine and persistenceCompany brain and humans-at-the-edge operating modelBurn tokens, not headcount; eliminate middle managementLegibility: record everything, fewer Slack DMs, artifactsInvestor-call analysis and simulation

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.