0DTE options on SPY whipsaw back and forth on every little price swing. They are torqued up and ready to run, cutting either for or against your P&L with every tick. I was certainly terrified by them, but I also understood their potential to amass (or destroy) capital, so I decided to try my hand at trading them manually. After some mixed results — okay, it was one positive month followed by a down month where I lost all of my first month's profits — I knew that it wasn't for me. The stress of trying to time SPY moves and make trades was too much. But I knew that the potential was still there.
Of course I'd been inundated with stories of the magic of vibe coding and how it allowed anyone to slop a project together, but I was skeptical. I had used a few AI tools in the past with varying degrees of success — there was no way that I could actually deploy anything vibe-coded to trade my own money… right?
In late February 2026, after swearing off trading 0DTEs on my own, I decided to enlist Claude to see if it could help me. It's simple, right? SPY goes up → buy calls, SPY goes down → buy puts. It started with a simple prompt: "how can I identify momentum as it starts?" I didn't know about any of the existing infrastructure that could be leveraged to take that simple concept and turn it into a functioning tool.
Claude (with some help from Gemini) helped me to learn a lot about Charles Schwab's REST API and WebSocket framework, the backend tools to get the plumbing all connected, and also about Flask and web tools to be able to visualize everything as it happens. After realizing that I could get second-by-second SPY and SPY options ticks, and that I could actually place orders nearly instantaneously, I knew that it was an avenue worth pursuing.
Over the next couple of weeks, I used Claude to help me research, plan out, and build a custom platform to continuously stream market data, analyze the quotes it retrieves, and place orders, all autonomously. It began as a crude tool: a Python script that used a basic strategy and wrote everything to the terminal. That made my eyes bleed, so I prompted my way into creating a dashboard to house all of the live market data I was logging along with trackers for all of the different strategies I wanted to test.
As I logged these early paper trades, I continued to dig into trading strategies and implemented refinements as I went. At one point, I was running 26 different test variants concurrently during live market hours! I quickly learned that being able to backtest different configurations against the SPY tick data that I had accumulated was a must, so I had Claude help me build that out, which led to discovering that the variant using a Kalman filter was outperforming the rest at detecting SPY momentum early.
Diving into multiprocessing allowed me to build out a robust simulation tool to sweep over dozens of config tweaks for weeks of SPY tick data at a time, which I then used to fine-tune the parameters the signal algorithm uses. But after deploying, it was apparent that a single parameter set was overfitting to the sample data that I had collected, so I studied up on tuning strategies. Claude helped me cook up a Bayesian lookup table that dynamically adjusts signal parameters in response to varying market conditions. As I accumulate more trade data daily, it feeds back into the table to refine its recommendations for future sessions.
So where are we now? I'm currently running this tradebot daily with real (albeit small) money trades, and it's showing a lot of promise. The slippage on real fills for my orders is eating into the edge that the tradebot had when it was only creating paper trades, but it is consistently bringing in capital. I'm excited to continue to optimize the bot's performance in hopes that it can continue to scale up.
This was my first big vibe coding project, and while it isn't yet the infinite money glitch that I had imagined, I'm really proud of what I've accomplished with it. Along the way, I have learned so much about system architecture, testing and debugging, data analysis, and quantitative concepts. But perhaps most importantly, I have gained a deeper appreciation for the power of the tools at my disposal. I now feel much more confident in my ability to leverage AI to learn and to create new things, and I know that's going to be priceless in our ever-accelerating world.