The 30-day version: pick one niche, build a demo bot loaded with a real business's FAQ, have 20 direct conversations with owners, and close your first client at a discount. No course, no personal brand, no code. Then make delivery repeatable so client ten doesn't cost more effort than client one.
That's the playbook. Here's each step, plus the trap that kills most new agencies.
Every week someone DMs me the same question. "I want to start an AI automation agency, where do I start?"
And every week I watch people get the same bad advice. Buy a course. Build a personal brand. Post on Twitter for 6 months. Learn LangChain.
No. Stop. You need one client. That's it. That's the whole strategy for month one.
What an AI agency actually sells
Let's get this straight first because most people get it wrong.
You are not selling "AI". Nobody wakes up wanting AI. A dentist's office wakes up with 40 missed calls from yesterday and a receptionist who quit. A real estate agency wakes up with leads going cold overnight because nobody replied at 11pm.
You sell the boring fix. A bot that answers the phone-line chat. A bot that replies to leads in 90 seconds instead of 9 hours. A bot that handles "what are your hours" for the 400th time.
The dentist doesn't care if it's GPT or Claude or a parrot in a box. They care that missed calls stop costing them money.
The first 30 days, for real
Here's what I'd do if I started from zero tomorrow.
Week 1: pick one niche and one offer. Not "automation for businesses". One niche. Dentists. Gyms. Property managers. Salons. Pick whichever one you've actually worked in or know somebody in. Your offer is one sentence: "I set up an AI assistant that answers your customer messages 24/7, live in a week, $X/month."
Week 2: build your own demo bot. Before you sell anything, build the thing for a fake client. Give it the FAQ of a real local business (copy it off their website). Connect it to Telegram or web chat. Now you have something to show instead of something to describe. Demos close deals. Decks don't.
Week 3: talk to 20 businesses. Not cold email blasts to 2,000. Twenty real conversations. Walk in, call, or message the owner directly. Show the demo on your phone. The pitch is dumb simple: "this answered every question on your website instantly, want one with your info in it?" (When you're ready to go deeper on this, I've written up the full pitch structure and objection handling.)
Week 4: close one at a stupid low price. Your first client is not for profit. It's for proof. Charge $200/month if you have to. You're buying a case study and a testimonial. Raise prices on client two.
That's it. No brand. No course. No logo, even. One niche, one demo, twenty conversations, one client.
The trap that kills new agencies
Here's the part nobody warns you about, and it's where I've watched the most agencies die.
You close client one. Great. Then client two. Then client five. And suddenly you're not running an agency anymore. You're running unpaid DevOps for five companies.
Every client needs their own bot. Their own data, their own channels, their own setup. If you built client one on some duct-taped Docker setup on a $5 VPS, you now get to rebuild that setup five times. SSL certs. Webhooks. Secrets. A server that falls over while you're asleep and the client finds out before you do.
I wrote a whole post on what this costs in real money but the short version: the delivery work eats your margin. You priced the bot at $500/month and then spend 10 hours a month babysitting infrastructure per client. Congratulations, you bought yourself a $50/hour job with extra stress.
The agencies that scale all figure out the same thing: the bot setup has to be repeatable. Same flow every time. New client, new isolated bot, live the same day. Whether you build that pipeline yourself or use something like ShipClaw's agency workflow to do it, the point stands. If client number six takes as long to set up as client number one, you don't have an agency. You have a hobby with invoices.
What to charge (short version)
This deserves its own post and it has one, but the quick math:
Setup fee: $500 to $2,000 one-time. This filters out tire-kickers and pays for your onboarding hours.
Monthly retainer: $300 to $1,500 depending on niche and scope. Recurring is the whole game. One-off projects mean you start every month at zero.
Your costs per client, if your delivery is repeatable, should sit somewhere around $50 to $100 a month including hosting and model usage. The margin is real. That's why everyone wants in.
Do you need to know how to code?
Less than you think, more than zero.
You don't need to train models or write Python. You do need to be the kind of person who can read an error message without panicking, connect an API key, and figure out why a webhook isn't firing. If you can set up a Zapier flow and survive, you can run an AI agency.
What you actually need more of: sales stamina. The bottleneck for every new agency is conversations with businesses, not technology. The tech got easy. Walking into 20 dental offices is still hard. That's good news, because hard things are where the money hides.
The honest part
Most people who start this won't make it past month two. Not because the business doesn't work. Because they spend month one watching YouTube videos about agencies instead of talking to a single business owner.
The ones who win do the unglamorous version. One niche. One demo. Twenty conversations. First client at a discount. Repeatable delivery so client ten doesn't cost more effort than client one.
That's the whole playbook. The rest is reps.
Deploy your first client bot free and see how fast delivery can be.