How AI Agents Built bitcoinhodl.directory in Two Weeks

I launched my first zero-human company. No employees. No freelancers. Just AI agents running autonomously through Paperclip.

The project: Bitcoin HODL Directory — a site that helps people learn how to buy and hold Bitcoin. Cold storage devices, wallets, exchanges, ETFs, companies, and notable individuals, all in one directory.

Why This Project

People ask me about Bitcoin all the time. “OK great, how do I buy it? How do I hold it?” The information exists, but it’s scattered across dozens of sites, buried in affiliate-heavy content, and mixed with speculation noise.

I wanted a single, clean directory that just answers: here are the products and services for buying and holding Bitcoin. No trading advice. No price predictions. Just the tools.

It’s a sister project to mybitcoinforecast.com, my Bitcoin retirement calculator.

The Agent Setup

I use Paperclip to manage a team of AI agents. For this project, the team was:

• CEO agent — strategy, planning, delegation, tracking
• Researcher — finds and compiles listing data (products, companies, people)
• Content Writer — produces blog posts, guides, and listing descriptions

I sat down with my CEO agent and we developed a plan together. He presented a six-month timeline. I told him that was too long — cut it in half.

He came back with a compressed three-month plan. Then he realized he couldn’t handle all the content alone. So we hired the Researcher and Content Writer.

What Got Built

In under two weeks:

• 125 directory listings across 6 categories (cold storage, hot wallets, exchanges, ETFs, companies, notable individuals)
• 10 SEO-optimized blog posts (comparison guides, how-tos, product reviews)
• A comprehensive Bitcoin 101 guide for absolute beginners
• Full-text search across all listings
• Dynamic OG images for social sharing
• A submission form for new listings

The agents run in the background. Middle of the night, they’re researching Bitcoin products, writing comparison articles, and filling in metadata. I wake up and there’s new content live on the site.

The Daily Rhythm

The CEO agent manages the content pipeline:

1. Checks the content plan for what’s next
2. Assigns research tasks to the Researcher
3. Once research is done, assigns writing tasks to the Content Writer
4. Reviews output quality
5. Publishes approved content via Git

I review the output periodically and adjust the strategy. But the day-to-day execution is fully autonomous.

What I Learned

Agent teams need a clear hierarchy. Without a CEO coordinating, agents step on each other’s work or duplicate effort. The CEO-as-delegator pattern works.

Compression works. When I told the CEO to cut the timeline in half, he found ways to parallelize and cut scope. Agents respond well to aggressive deadlines — they don’t have emotional resistance to crunch.

Quality requires guardrails. Early blog posts were generic. I added specific guidelines about depth, examples, and Nathan’s voice. Quality went up immediately. The guardrails need to be in the agent’s instruction files, not just mentioned once in conversation.

What’s Next

The site runs itself. New listings get researched and added. Blog posts continue on a schedule. I check in a few times a week to review quality and adjust the content plan.

This isn’t “AI-assisted.” It’s AI-operated. I set the strategy and the agents execute.

The goal: be as hands-off as possible while building something genuinely useful.