Found something good?

A $600 box on a shelf. One agent living inside it. No payroll, no office, no Monday standup.

That line sounded insane 12 months ago.

Right now people are filming it on their desks.

First, the proof. Because you won't believe the plan until you see the receipts.

A 20 something started 2 businesses 14 days ago. He's already past $1,000, best week near $900. He didn't run the stores. 2 agents did one named Ray scrapes TikTok and Etsy for winning products, the other lists them and waits for orders.

Another one: $191 in 7 days. An Etsy + Printify store run by 2 sub-agents. One finds the trending product, the other kills it if sales die. The first hit was a squish duck. Sold out in a week.

A third built a whole factory on screen. Etsy store near $2,000 in 14 days, 37 orders. In the same room a second agent pumps deep house to Spotify 24/7 and builds its own YouTube visualizers. A third undercuts a Fiverr seller charging $55 a thumbnail same output, generated, fulfilled while he sleeps.

Then it scales. One founder spun up 42 agents to run a sales department. A chief sales officer on top. Sub-agents under it: 8-language outreach, RevOps, lead scoring, call prep, proposals. They pass context to each other. The researcher feeds the email writer. The call analyst feeds the follow-up drafter. Not 42 chatbots. 1 org chart.

And the dumb-simple version that prints: a window-cleaning business. $6,800 week 1. $13,000 week 2. The owner never lifted a finger. One agent booked every job.

These aren't pitches. They're screen recordings.

Why the old way can't compete

Hiring a human marketer: 3 weeks to find, $5,000 a month, sleeps 8 hours, quits in a year.

One ChatGPT tab: forgets everything the second you close it. No memory. No hands. No access to your files.

No-code spaghetti: breaks the moment 1 node changes.

The unlock isn't "a smarter chatbot." It's an agent that remembers, acts, and hires more of itself.

That's the whole company. 1 brain box. Agents as staff. Each one a specialist with its own memory.

The model: 1 Mac Mini is the building. The agents are the employees.

Picture an org chart, except every box is software.

2 tools run the whole thing:

You never touch the org chart by hand. You type, it builds.

The build. 6 phases. Copy the blocks.

Phase 1 — The hardware

A $5/mo VPS gets you started. A $600 Mac Mini M4 is the real move — you see what the agents do, debugging is 10x easier, and security is sane out of the box. It runs 24/7.

Phase 2 — Install the brain

Make sure Node.js is on (node --version, needs 22+). Then:

Run it once to authorize:

Then install OpenClaw — the always-on layer:

The onboarding asks for your Anthropic key, your default model, and your messenger. Pick Telegram — it's the easiest to debug.

Phase 3 — Wire up Telegram

Open BotFather, send /newbot, copy the token (1234567890:AAF...). Drop that token into OpenClaw's Telegram config. Message your bot. It replies. That bot is your main agent. Everything else gets built through it.

Phase 4 — Hire your first employees

You don't write config files. You write messages. Create a Telegram group per agent, grab its ID with myidbot (/getgroupid), then tell the main agent:

That's hiring. 30 seconds. No interview.

Phase 5 — Give each one a personality

Quality is 80% the instructions. Write the agent like you're onboarding a new hire:

And the secret weapon — the reverse prompt. Don't micromanage. Ask the agent to fix itself:

The agent knows its own limits better than you do. It rewrites itself.

Phase 6 — Connect the chain

Flip on cross-talk so agents can pass work down a pipeline:

Now it runs without you:

That's a company that opens its own laptop in the morning.

Pick the right staff for the right job (this is your payroll)

The going rates per 1M tokens: Opus $15 / $75. Sonnet $3 / $15. Haiku $0.25 / $1.25.

Translation: a researcher running all day on Sonnet costs you a few dollars. Your "salaries" are smaller than one team lunch.

What it actually looks like, month by month (no fantasy)

Month 1: 1 Mac Mini. 3 agents. First store live. You make $191 in a week and feel ridiculous.

Month 2: 6 agents. A pipeline that runs while you sleep. Maybe $1,500–$3,000. Half your time goes to fixing dumb agent mistakes.

Month 3: 2 Mac Minis. 12 agents across 2 product lines. The margins compress as you scale more orders means more support, more refunds, more edge cases. That's normal. The fix is more specialists, not more hope.

Month 6: a farm. Marketing, sales, ops, finance each a cluster of agents. You stop touching the work. You touch the system.

The $1B headline is the ceiling, not the first week. But the shape is real: 1 person, a shelf of boxes, an org chart that compounds because every output gets written back into memory and the next task starts smarter than the last.

Before. After.

Before: you hire 5 people, burn $25,000 a month, wait a quarter for output, and pray nobody quits.

After: you plug in a box, type 6 messages, and the team is live by dinner and it never sleeps, never quits, never asks for a raise.

A human hire costs $5,000 a month. This one costs $5 and a power cord.