AI native startups: how to build a 20x company
Y Combinator's playbook for AI-native companies that outcompete teams 100x their size, and how 1ar labs runs on the same principles.
Y Combinator published a video that perfectly captures what I’ve been building toward with 1ar labs: The New Way To Build A Startup.
The thesis is simple. In the AI era, startups aren’t winning by hiring faster — they’re winning by automating as many internal functions as possible. Code, support, marketing, sales, hiring, QA — all of it. YC calls them 20x companies.
All of us devs manage anywhere between three and eight Claude instances implementing features, fixing bugs, or researching potential solutions. — Boris Cherny
Some numbers from real YC-backed companies:
- “When we got DoorDash as a customer, we were approximately four to five engineers going against players who had 100x engineers.”
- “We’ve grown 4x in the past year, but we haven’t hired a single net new person.”
- “Right now we have only a single human FTE within the company.”
- “Phase Shift right now is a 12-person team and we’re going up against companies that have been around since 2006.”
YC breaks the approach into three strategies. I run all three at 1ar labs.
1. Build an AI teammate (internal agent)
A powerful internal AI agent that functions like a full-time employee.
Giga ML built an agent called “Atlas” — it can access browsers, edit policies, and write code. It handles boilerplate tasks, doubling or tripling each engineer’s scope. It also manages customer service for massive accounts (like DoorDash) autonomously, letting a single human employee run 10+ Fortune 500 pilots.
How I do it: Claude and OpenClaw are my employees. I manage multiple Claude Code instances in parallel — each one implementing features, fixing bugs, or researching solutions. My custom statusline setup helps me keep track of all of them. The agents handle the work; I handle the direction.
2. Build an AI-integrated source of truth
A unified internal interface that instantly pulls all relevant context across systems.
Legion Health built a custom interface for their care operations team — patient history, scheduling, insurance codes, communication logs, all in one view. Result: 4x revenue growth without hiring a single new person for operations.
How I do it: Craft is my source of truth. All notes, project specs, client briefs, research — everything lives there. And it’s not just for me to read. I connected Craft via OAuth-protected MCP so my AI agents can access scoped, structured context directly. Graft — a knowledge graph tool I built for Craft documents as part of the global Craft hackathon (I won the grand prix) — lets me visualize how everything connects. The agents don’t guess; they pull real context from a real knowledge base.
3. Build custom agents for each workflow
Tailoring automation to the specific manual tasks of individual roles.
Phaze Shift (automating accounts receivable) asks employees to document their daily manual tasks, then builds custom AI agents for each one. They avoided hiring a designer entirely by using AI tools for front-end design.
How I do it: I build custom tools for specific workflows. Autaxy is one — a time-zone-aware scheduling calculator that I use daily and ship to clients who need it. export4ai is another — it packages folder contents into AI-ready payloads for faster iteration. Sumr started as a personal tool to summarize articles in Safari and turned into a product. When I hit a recurring friction point, I build a tool. When the tool is good enough, I ship it.
Why this matters
This is not a future trend. This is happening right now. A single person or a tiny team, with the right AI infrastructure, can compete with companies that have been staffed for decades.
1ar labs is one person. I build and ship products (cin, Sumr, Graft), take on client work (agentic AI systems, bespoke platforms), and maintain everything — because the bottleneck is no longer headcount. It’s how well you set up your agents and how good your source of truth is.
If you’re building something and want this kind of infrastructure — or want to see how AI-native operations work in practice — get in touch.