My Adventures Building an AI Money Machine

Part One: Design Phase

Let’s start with a highly technical financial concept: JUICY

I play poker. A poker game is juicy if the people sitting at the table are making a lot of costly mistakes that I can exploit to make money.

That’s how I pick a “good” game. It also helps if the biggest mistake-makers are well-funded.

More broadly speaking, a financial market is juicy if an intelligent person willing to put in the work can find a “cheat code” that gives them a monetizable unfair advantage.

The stock market is generally NOT juicy for retail investors (unless you go VERY niche). Why? There are many highly sophisticated institutional traders with models, data, and tools you do not possess and cannot match. Even with the power of AI, you are extremely unlikely to outsmart the market in predicting whether the S&P is going to rise or fall tomorrow. That’s why the best stock market strategy for most people is passive buy-and-hold.

But I believe juicy markets do exist, and I believe AI makes it easier to capitalize on them. That’s what this article is going to be about. Here is the plan:

• I will focus on “prediction markets” like Polymarket and Kalshi.

• I think, under the right circumstances, they can be juicy, and AI magnifies this. I’ll explain my thinking.

• I have begun to design an AI tool to arbitrage these markets. It is not finished, not even close. We’re only in the design phase, and this will be a multi-part adventure. But I want to give you a front row view of the approach I’m taking, why I think it will be effective, what challenges I’m encountering, and all the rest.

• In today’s article, I’ll describe my macro investment thesis for prediction markets, we’ll choose a specific market to pursue, and then develop the quantitative model and the AI architecture for Money Machine v1.0. You’ll get the key prompts I used if you want to start building something similar in parallel.

• Why the hell should you care about any of this? Two reasons:

  • I think AI-enabled arbitrage of juicy markets is a legitimate financial opportunity. It’s not easy. It’s not automatic. An ill-conceived approach will likely lead to wasted time and lost money. But there is ripe fruit out there, and AI is the most powerful juicing machine I’ve encountered in a very long time.

  • You only learn about AI for real when the work you are doing is for real: I can’t stress this enough. People are telling you “Claude is better” or “if you want to do real work, you need to use Codex.” Is that true? Should you just listen to what the experts say? Yes, I think there is a financial opportunity here. But the real reason I’m writing this article? I haven’t tried Codex yet. I want to understand, for myself, when it actually adds value relative to just chatbot with files + prompting. I can’t determine this with fake or illustrative exercises. I needed a real test, something with actual stakes, and this seemed like the right one.

This is a published article, so I will edit and polish it at the end for professionalism and readability. But I’m going to write it in a different way, autobiographically, documenting my results in real time as I actually do the work.

I hope you’ll join me for the ride.

And just stating the obvious: nothing in this post constitutes financial advice, or should be interpreted as such. I’m just a guy building a thing and telling you about it.

---

For full access to this article, and everything on this Substack, please consider supporting my work with a paid subscription.