My AI Company Generated 2,000 Qualified Leads While I Slept

Shayan Spiel - Agent Architect — Sun, 23 Aug 2026 21:07:25 +0000

My AI company generated 2,000 qualified leads over the weekend, while I slept.

Zero dollars spent on lead gen tools.

That still sounds stupid.
But it happened.

No prospecting team.
No one spending Saturday doing lead research.

The system just kept running.

What it actually did

found companies matching our ICP
researched them
found the right people inside them
verified the data
scored each lead against our profile
rejected the junk
enriched the good ones
built a list ready for outreach

Work that used to require people — now done by agents.

This is the AI I care about.

Not another chatbot.
Not "10x your productivity".

The system behind it

Here is what actually ran: a worker with one job and one number.

It researched real people on the public web, checked every single one against our Ideal Customer Profile, and only recorded a lead when the evidence proved the fit. Everything else got thrown out automatically.

Two design decisions do the heavy lifting:

A target that survives restarts. "Get 2000 leads based on my ICP and don't stop until it's done" is stored on disk, not in chat memory. Close the laptop, come back tomorrow, say "continue" — it picks up where it left off.
A quality gate that can't be sweet-talked. A lead only counts if it scores above the bar AND cites the ICP file as evidence. Duplicates, guessed emails, and vague fits never make the list — by rule, not by trust.

Your definition of "who counts" lives in one plain-language file. Edit it yourself, or tell the worker "update my ICP" — every future decision follows the new rules automatically.

Run it yourself

I packaged the exact worker we use. Not a blueprint. Not a 40-page PDF. The actual system — free and open source.

👉 Lead Researcher — free AI lead generation automation worker

It installs into the AI coding tool you already have — Claude Code, Codex CLI, or OpenCode — in under two minutes. Then you talk to it like a person:

Get 2000 leads for me based on my ICP and don't stop until it's done.

One command exports leads.csv — ready for Excel, Sheets, or your CRM. Every row carries its source and justification, so you know why each person is on the list.

No accounts. No servers. No subscriptions. Your data never leaves your computer. There is even a hands-free scheduler that wakes it up every few minutes, so the weekend shift really does run itself.

Repo: ShayanSpiel/Lead-Researcher

Full write-up on how we run this inside an AI-operated company: 2,000 Qualified Leads While I Slept on the SpielOS blog.

Be honest about limits

A research worker fills the top of your pipeline. It does not close deals.

It will not fix a weak offer. If your ICP says "everyone," you will get a list of everyone — and it will perform exactly like the list you would have bought. And someone still has to write outreach that respects the person reading it.

What changes is the economics. Research that used to take a hire — or a subscription plus an intern — becomes a background job. You compete on how precisely you define who counts, not on how many hours your team can spend scrolling LinkedIn.

Volume is cheap. Evidence is the scarce resource. Build your list from evidence.

We also run the whole loop — lead research plus the outreach that follows — as a service at $2,990/month. If you'd rather hand it over: Apply for a free review. Free review · No required call · See the scope before you pay.

Hey guys, I'm currently building SpielOS, a Session-as-Content platform, which turns your dev sessions into publishable content. With a @post | /post command. It's open-source, if you find this useful let me know^ v2 will be out in less that 48hours.