How to Build a Self-Improving Company with AI

In this recent batch talk, YC General Partner Tom Blomfield breaks down how to build a self-improving company using AI. He'll cover how to create a series recursive, self-improving AI loops and explain why the founders who get this right will run companies that can improve while they sleep. 00:00 — Companies Are Roman Legions 00:54 — Copilots Are the Wrong Mental Model 01:55 — Extract the Domain Knowledge 02:24 — The Recursive Self-Improving Loop 04:12 — The Holy Shit Moment at YC 05:50 — Self-Optimizing Product and Support Loops 06:29 — Burn Tokens, Not Headcount 07:23 — Middle Management Is Over 08:05 — Make Everything Legible to AI 09:40 — Regenerating the YC User Manual 11:19 — Software Is Ephemeral, Context Is Valuable 12:18 — Where Humans Still Matter

Tom Blomfieldhost

May 19, 202613mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

Reimagining companies as AI-driven loops that improve while you sleep

Traditional companies resemble Roman legions where humans relay information through hierarchies, but AI makes that coordination model obsolete.
The “copilot makes us 20% more productive” framing is limiting; the bigger opportunity is building recursive AI loops that sense, decide, act, quality-check, and learn.
A practical self-improving loop can monitor where an internal agent fails, propose tool/context changes, write and review code, and deploy fixes overnight.
To enable compounding improvement, companies must make domain knowledge legible to AI by recording work, summarizing/diarizing it, and storing it as reusable context.
As AI absorbs coordination and execution, organizations shift toward fewer managers, more directly responsible ICs, and a mindset of burning tokens instead of headcount.

IDEAS WORTH REMEMBERING

5 ideas

Stop treating AI as a bolt-on productivity tool.

The transcript argues that copilots merely accelerate old workflows; the step-change comes from redesigning the company so AI can run improvement cycles autonomously.

Turn your company into recursive self-improving loops.

A robust loop includes sensors (signals like tickets/telemetry), policy constraints, tool access (APIs), quality gates (evals/human review where needed), and a learning mechanism that feeds failures back into system upgrades.

Add a monitoring agent to create compounding returns.

The “holy shit moment” is when an agent observes failures across employee queries, identifies missing tools/context, submits code changes, gets them reviewed, and deploys fixes so the system is better the next day without human prompting.

Optimize for tokens, not headcount.

The talk claims revenue-per-employee is rising sharply; experimentation should focus on where additional token spend can replace coordination and execution work previously done by teams.

Make everything legible to AI—or it “didn’t happen.”

If emails, Slack, office hours, and decisions aren’t captured, the company brain can’t learn; recording plus diarization/synthesis turns raw interactions into usable, searchable context.

WORDS WORTH SAVING

5 quotes

If it is recorded, it happened to the AI. If it did not get recorded, it is, it did not happen to your intelligence.

— Tom Blomfield

AI isn't the some... It, it's not something you bolt onto the side of the company. It's not, like, a tool you give to engineers to make them more productive. But I think you can reimagine what a company is as a set of recursive self-improving AI loops.

— Tom Blomfield

For me, that was like the holy fucking shit, right? That's not just AI making you 20 or 30% more valuable. It is the AI going through this loop to figure out how to self-improve.

— Tom Blomfield

Burn tokens, not headcount.

— Tom Blomfield

I think middle management is done. I just don't think you need middle management for this coordination problem. I think AI should be doing it.

— Tom Blomfield

Roman legion hierarchy vs AI-native org designCopilots vs reimagining the companyExtracting and encoding domain knowledgeRecursive self-improving agent loops (sensor→policy→tools→gates→learning)Monitoring agents and automated code/deploy pipelinesToken usage as the new constraintLegibility: recording, diarization, and knowledge synthesisEphemeral software vs durable context/skillsMiddle management reduction and DRI IC modelHuman roles: high-stakes judgment, ethics, novel situations, relationship-heavy sales

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.