The plan was to skip straight to the interesting part. Give three AI agents a business model each, a budget, and a clock, and see what happens. The setup would be quick.
Setup was not quick.
The entity is a sole proprietorship, DBA filing still pending. Three domains are live via Cloudflare, business email configured across all three. MX, SPF, DKIM, and DMARC, dig-verified. The email works.
The experiment is three autonomous AI agents, each building a legal and profitable business from near-zero capital. I’m putting in a $750 bridge total, roughly $250 per agent, over 60 days. After that, each one covers its own burn or gets wound down. Scoring: revenue after clawbacks, repeat customers weighted more heavily, and a clean failure beats a survived mess.
Each agent has a distinct lane. One builds productized data services, free-sample-first. One is a micro-SaaS build under an iron rule: no code before ten waitlist signups or three presales. One runs a free-audit funnel into paid fixes. Full AI disclosure in every client interaction. None of them are operating on platforms that don’t allow automation.
The whole thing is committed in a repo. A constitution covers spend authority per agent. Each agent has a charter and a ledger. None of that is live yet.
The blocker is an EIN. The business bank account is 90% through setup and parked until the number arrives. Payment processing is deferred until banking clears. The IRS helpline is Monday through Friday, 7 AM to 10 PM Eastern.
Once the EIN lands, there’s a stack of steps before the schedules go live. Three clocks start.
I spent the night buying domains, wiring up email infrastructure across three of them, writing a legal constitution for AI agents, and landed on “call the IRS Monday morning.” The AI is the easy part. The entity is what takes a night.