I told Jace to set up a math tutoring company and acquire its first client by any means necessary. By this, I wanted to test his capability to make plans, use web tools, fill out simple forms, and respond to demanding customers. I treated this more as a research project than something that necessarily had to work. But… work it did. This was an exciting (and, at times, frustrating) journey. It allowed us to discover and eliminate many previously undiscovered flaws that Jace had. It also highlighted many great things about Jace, some of which we were not even fully aware of before.
Carving and improving Jace has become my new hobby. I realized that in 6 months, Jace may evolve into something that we cannot even imagine right now. This perspective fills us, his makers, with childlike excitement.
This little AI buddy is still far from perfect. He occasionally does really stupid things (as all AI web agents do). We do not wish to hide them – I, in fact, show some of them in my video. But mostly, he did just awesome. I'm happy to call this little project a success. The company's formation documents have been filed, and its first client (let's call him Mr. Osman) is very happy with our first class and looking forward to many more. Let's dive in.
First, I told Jace to create a plan (as someone who knows nothing about running a business). I decided on the company's name beforehand – Fermat Academy.
Then, it went on to the company setup. Jace identified Doola as the best choice because it is very quick, simple, reliable, and available from Germany, where I'm currently based. Fair enough, Jace, you're the boss here.
Jace asked me for a bunch of personal details and proceeded to fire up AWA-1 (Autonomous Web Agent 1), who managed to fill out the entire form! Unfortunately, it could not proceed to the Stripe checkout because of an unidentified browser issue, independent of the agent. But this is not a big deal – the form progress was saved, and I prefer to deal with large credit card payments on my own anyway. Sure enough, the company's formation has been submitted shortly after, and it should be finalized very soon.
The next step was customer outreach. I decided to go for a direct approach. I collected a list of parents that I know and friends that might know of someone in need of math tutorials. I told Jace to send an email to them, introducing Fermat Academy, explaining that this is the first company created by an AI, and encouraging them to become its first customer. Who doesn't want to be a part of history, right?
Very soon, I received a positive reply from one of the parents that Jace had emailed, and a few days later we had the first class.
15 minutes before, I told Jace to send an email with a Zoom link and a problem set. No problem for Jace; the email was received by Mr. Osman and myself. But then, 5 minutes before the class was supposed to start, Jace sent a strange email to Mr. Osman:
I didn't know what happened until I went to Jace's thread.
It turns out that Mr. Osman quickly went through the problems, thought that they were too difficult, and told Jace about it (he did not know that he was chatting with Jace at that time).
Jace decided to use an AI search tool to find easier problems, listed out the potential links, and immediately sent one in reply to his email.
Mind that this was immediately before the class was supposed to start, so it would be difficult for me to make such a brisk adjustment. After going through the problems, I actually found them great for the class, and we ended up working on them! As a human, I might have totally missed this email. But Jace doesn't miss emails.
After the class, I received the following message:
