It's a bit embarrassing to say that I started using AI coding agents in depth, which was actually the result of being stimulated by our PM six months ago. One day, our PM suddenly knocked on me and wanted me to see his arbitrage program written in Claude Code. I thought it was just a little fuss, but when I clicked on it, I was scared out of my wits. The completion rate of the project he wrote is extremely high, mainly divided into two core areas: Find arbitrage targets Compare the funding rates of all perpetual contracts on Binance and Bybit, automatically select the trading pairs with the largest differences and long-term consistent directions for backtesting, and find the trading pair with the best ROI and MDD. Cross site brick moving monitoring Detect the margin status of both exchanges every 5 seconds. Once the water level is imbalanced, the system will automatically allocate funds from the profit-making party to make up for it, ensuring the safety of the position. This is already the level of commercial products. Many paid arbitrage tools on the market have the same underlying logic, the difference is that others use them to sell money, while they use them for their own purposes. He lost $50000 and went on to make real money almost every day. When the market was good, the annualized turnover exceeded 100%, and when it was bad, it was also 30%. He himself was also puzzled: 'Did I miss any risk? How could making money be so easy?' I gave him some practical advice, including the liquidity limit for small coins, the risk of the exchange closing deposits and withdrawals, and risk control details such as earning too much and being automatically reduced by the system's ADL (automatic position reduction). After listening to it, he turned back and optimized the program, and in the end, this system withstood the extreme market of the large-scale liquidation of 1011. What shocked me the most was that our PM graduated from the Department of Industrial Design and had no programming background at all. He even learned Git while doing it, and with Claude Code, he broke into the door of quantitative trading because he continued to make money. Recently, some investors came to him willing to give him money to operate. After that day, I put aside the arrogance of a professional and diligently studied Claude Code and Codex. A week later, I released the CoinKarma backtesting system using Vibe Coding. The development speed and execution efficiency are completely different from the experience of repeatedly copying and pasting code using ChatGPT windows in the past. This is the most important lesson I learned from our PM at home: If you truly have talent, please be grateful for living in this era. This may be the era with the lowest probability of 'talent not being met' in human history. In the past, an astonishing business intuition often died from not understanding code, not being able to afford engineers, and dying from long development cycles; But in the face of AI, the technological monopoly has completely collapsed, and the development threshold has been flattened. This means that as the cost of implementation approaches zero, the value of ideas tends towards infinity. Previously, technology was a wall against mediocrity; Nowadays, technology is the sail that carries genius. In today's world where knowledge and tools have been democratized, the only thing that limits your achievements is no longer your degree or background, but your insight into the world and the courage to invest real money and silver to verify it after logical clarity. The top-notch business intuition has finally achieved the highest redemption rate in history. This is a golden age where no one can bury you as long as you 'understand the industry'.