Not subscribed? Sign up to get it in your inbox every week.

Rameel and I started a business last Saturday.

Rameel had pitched me the idea a few times, and I kept putting it off. Then I had a few free hours on a Saturday and finally said yes. The idea is simple: pick a real business, and see how much of it AI agents can run end to end, from finding customers to getting the work done.

The business is seasonal. We generate leads with Facebook ads and subcontract the actual work, which makes it a good test case because most of the job is marketing, quoting, and coordination. Those are all things an agent can plausibly do.

Here's where things stand 72 hours in.

We built a simple website with a lead form using ChatGPT Sites, bought a domain on Namecheap, and connected Zapier so that every new lead lands in our inboxes. The whole setup was quick and easy, and none of it needed a developer.

The ads were the interesting part. We used Fal.ai and OpenAI's image model to build an automated pipeline that generates the whole creative package: the personas, the storyboards, and the finished ads. A 15-second video ad currently costs us around $6 to $7 to produce. We made four video ads and two static ads, and they were running on Facebook by the end of the day. (If you're interested in how this works, send me an email!)

A few years ago, that last step alone (finding a creator, writing a brief, waiting on revisions) would have taken a couple of weeks and a few thousand dollars.

We set a budget of $50 a day and let it run.

The first lead came in within 12 hours. Two more arrived over the next two days, and as of this writing we've sent quotes to two of them. All in, we went from "okay, let's do this" to three leads in under 72 hours, for less than $300 (everything including the domain, ads, ad spend, etc).

Three leads won't make anyone rich, and the real test is whether those quotes turn into paid jobs. But the speed caught me off guard. The gap between having an idea and having a customer used to be weeks of setup, and here it was an afternoon and a day of waiting.

Once leads started coming in, I started thinking about how to automate what happens next.

I ended up with Grok Bot as my assistant, connected to an inbox from AgentMail (an email provider built so AI agents can send and receive mail easily). I'm still in the loop for now. When a lead comes in, I do a bit of pre-work, then pass it to Grok Bot to handle the rest: requesting quotes from subcontractors, following up with the prospect, and keeping the thread moving.

Here's how the work splits today:

The obvious next question is whether I can take myself out of that table entirely. The biggest limitation right now is that Grok Bot doesn't have a phone number, and a lot of customers and subcontractors in a business like this still want to text or talk on the phone. I'm planning to try Inkbox, which gives an agent its own email, phone number, and iMessage under one identity, to see whether it can take the calls too.

The thing I keep thinking about is how little stood between us and a working business.

We didn't need a team, a developer, an agency, or much of a budget. The tools are good enough now that the real constraint is whether you're willing to try. I think we're entering the age of agency, where getting something started mostly takes the effort to start, and most people who try will get a lot further than they expect.

The biggest blocker I see, in myself included, is the fear of failure. Rameel pitched this to me more than once before I bit the bullet, and the only thing that changed was that I had a free Saturday. The downside turned out to be a few hours and a couple hundred dollars.

So here's the Monday test for this week. Pick the idea you've been sitting on, whether it's a side business, an internal tool, or a process you've wanted to automate. Give it one afternoon and a few hundred dollars, and see how far you get by Wednesday.

I'll keep you posted on ours.

Till next time,
Chris