My entire financial theory was “a billion”
Matthew Berman recently asked me how much I thought a prominent AI engineer had been hired for. I said a billion dollars. A few weeks later, he asked me what it cost to run GTC (one of NVIDIA’s tech conferences). I said a billion. He asked what SpaceX might IPO for. I said a billion again. In my head, a complex financial calculus was occurring: how much money should I have before retirement? A billion. (Queue Dr. Evil meme.)
This is meant to give you a sense of my financial newbyness, I have never been much for business or finance. I have spent most of my life in academia, then for the last five years have pivoted to tech and startups. Nobody in my family owns a business. In my family, not having a pension is basically a character flaw. I bought my first car for cash. I am an investing newb, which is the phrase I use because it sounds slightly better than “financially illiterate.”
In 2024 I started learning about money. Mostly how to save it. Then how to invest it. It was the first time I had been cash-positive in a real way, despite being in my 30s. Two graduate degrees will do that to you. A doctorate, in particular, is an expensive way to learn how to make $0 feel intellectually defensible.
Autopilot made it too easy
I started with Autopilot because it took a subject I found intimidating and made it look like picking a playlist. There were portfolios, pilots, filters, performance charts, and little buttons that implied I could participate without knowing all that much. This was exactly what I wanted, and probably exactly what I needed to be careful with.
The obvious adult path would have been to read a book, learn basic portfolio theory, and slowly build a boring index-fund habit. Instead I put a chunk of cash into Autopilot’s Nancy Pelosi autotrader and thought: sure, if it's viral, it will work.
Where I started: Autopilot made portfolios feel like something I could browse, not something I had to understand first.
The Pelosi problem
The Pelosi tracker follows publicly disclosed congressional trades, with roughly a 45-day lag. That lag should make the whole thing feel too late to matter. But it worked anyway, gloriously, at least in my small account, to give me the very unhelpful feeling that I had discovered something.
This did not make me think markets are fake, or that finance is all nonsense. It did make me understand why people get cynical so quickly. When a beginner sees a congressional-trade tracker outperform a financial advisor, the first lesson is not efficient-market theory. The first lesson is: HeeeeeeeeHawwwwww, let’s go!
Autopilot did not make me informed. It gave me a way to borrow someone else’s confidence and watch numbers move.
The Pelosi Tracker result. I redacted the personal dollar amounts, but left the return visible because that is the part that taught me the wrong lesson.
Then I made money, which did not help
The problem with early success is that it feels educational. Early failure would have been cleaner. I could have nodded, said I learned a lesson, and gone back to index funds. Instead the line went up.
I made 54.3% in the last year on the Pelosi tracker. Across the Autopilot accounts, I was up about 52%. So naturally I upgraded to other paid subscriptions with investors and companies I mostly pretended to recognize. I added Wolff’s Flagship, a GPT-5 portfolio, an AGI portfolio, rare earth metals, and even Inverse Cramer. That last one felt less like a strategy and more viral bandwagoning.
At this stage I was not really investing. I was sampling other people’s conviction. The app gave me charts, names, returns, and just enough polish to make the whole thing feel calmer than it was. I checked it too often and understood it too little. This is not an ideal combination.
been.
Then I asked ChatGPT how options work
Then I started asking ChatGPT about investments. I always did the little disclaimer dance: I know you can’t give investment advice, this is educational, I will take it with a grain of salt. Basically I wanted advice without making anyone legally responsible for the fact that I was listening to whatever came my way, with basically no baseline understanding.
ChatGPT explained my portfolio back to me. It explained options. It explained short puts. I made a few good short-put trades, then let one expire without fully appreciating what happens when you might need to buy 100 shares (1 option) and do not actually have the cash ready. Whoopsie!
This is where AI really did help. It made finance less embarrassing to learn (until sharing this now). I could ask basic questions without performing competence for another person. I could ask the same question three ways until it finally clicked. That is good. The catch is that embarrassment can be a brake. Remove the embarrassment and you may also remove the pause that keeps you from doing the thing you only half understand. So I continued blindly.
Then came Composer
Then I added Composer. Composer lets you chat with an agent or chatbot and generate trading strategies. It calls them symphonies, which made me feel very intellectual about the uninformed choices I was about to make.
I typed, more or less, “make me a symphony for AI infrastructure.” The agent came back with something that sounded reasonable: companies powering AI development, semiconductor manufacturers, cloud providers, data center operators. It had categories, logic, and a chart. It was coherent enough to make me forget that I had no idea what I was doing moving money from my bank account into a random app I found on r/singularity.
Composer turning a loose prompt into a trading strategy. The embarrassing part is not that it worked; it is how quickly I accepted that it might.
It’s crazy how the distance between curiosity and action is getting so short. I do not think that is automatically bad, but it does change what a beginner can get themselves into.
Small account, large-looking chart. Emotionally, this was all the evidence I needed. Intellectually, it should not have
My friends performed a wellness check
This is around when I told my friends what I had been doing. Their reaction was not subtle. “Are you willing to lose it all?” one of them asked. The answer was an addict’s ‘no’, obviously.
They were not wrong to be concerned. In a pretty short window I had gone from not investing to copying congressional trades, subscribing to automated portfolios, asking ChatGPT about options, and prompting an app to build day-trading strategies. I could explain each step individually. The sequence, taken together, made me sound like someone who was about to buy an NFT apartment on the upper-east side of the Metaverse version of Manhattan.
Still, I do not think the whole thing is just stupidity. Finance has been locked behind jargon, confidence, and class signals for a long time. AI tools can lower that barrier. They can explain risk in normal language and make people less afraid to start learning. The problem is that they also make it easier to act before the learning has settled, and trust me, I was still settling.
The dashboard got personal
Then I connected to ChatGPT Finances. This felt different from asking a model about a trade. Now it could see spending, categories, risk exposure, and some actual life context. It was not just looking at the portfolio. It was looking at the rent money I was spending testing AI.
ChatGPT Finances made the advice feel more grounded. It also made the whole thing feel much more personal.
That context is useful. A model that does not know your cash flow can only give generic warnings. My real risk is not just whether a trade can go down. It is whether it goes down in the same month something expensive happens in real life. Once the system can see more of the picture, it can give better answers. So in other words: ChatGPT, my partner’s savior.
The agentic trading cliff
Now I’m also messing around with the idea of building my own AI trader, using random finance and trading repos and a connected agent to Robinhood. Robinhood now lets you connect your brokerage account to your personal AI agent, which feels like one of those features that is either the future of finance or the setup to a congressional hearing. Possibly both, especially if it leads to even more market volatility than day trading.
This is where the experiment gets less cute. There is a big difference between asking an AI, “What does this trade mean?” and letting the AI actually place the trade. One is a conversation. The other is a transaction. If AI gives me a bad explanation, I can ignore it, misunderstand it, or complain about it later. If an agent buys something, the market does not care that I was “just exploring.” Robinhood isn’t giving me a refund because I called them to complain about my OpenClaw or Hermes version being out of date.
So if I do this at all, it has to be boring first. Tiny limits. No margin. No options. Human approval on every order. A full log of why it wanted to do what it did. Basically, every safety feature that makes the project less fun is probably the exact thing that makes it sane. But please, check in with me in a week to see what guardrails I took down in favor of being lazy.
You can now add an agentic model to Robinhood.
What I actually learned
I do think AI taught me to be a better investor. Although maybe a bit overconfident given my success with tools that require little understanding. It taught me enough vocabulary to become a more confident beginner. That is better than being an intimidated beginner, but it is not the same as wisdom.
The bigger story is that AI makes financial experimentation feel legible, and potentially lucrative – I have made a pretty penny. It turns the market into something you can chat with, browse, prompt, and configure. That is powerful. More people can learn. More people can ask basic questions. More people can see behind language that used to make finance feel intentionally unfriendly.
The risk is not that beginners are dumb. The risk is that beginners can now move very quickly with partial understanding. You can learn one concept, get a plausible explanation, see a green chart, and move money before humility has time to show up.
There is also something uncomfortable about what I wanted from these tools. I did not only want returns. I wanted to feel like the kind of person who understood returns. AI is extremely good at giving you that feeling. It explains things patiently. It uses your context. It makes the next step feel available.
That might be the real product: confidence. Not in a scammy way necessarily. Confidence can help people start. But confidence can also make a half-formed idea feel complete. The agent keeps whispering: you are not guessing, you are configuring.
So I am not anti-AI trading tools, definitely not – I will keep playing until I am the majority shareholder of SpaceXAi. I am also not ready to pretend they are harmless. They make finance easier to enter, which is good. They also make it easier to do dumb things with professional-looking dashboards, which is less good. Both are true.
This is not financial advice. It is more like therapy notes from someone who finally started learning about investing and AI-augmented trading and immediately let the apps get a little interesting. I am still an investing newb. Just a better-instrumented one. And if Matt asks me what valuation SpaceXAi will IPO for, I still know my answer: probably a billion.
