Today, while chatting with GPT, I came across an interesting contrast behind the AI boom. On one hand, there’s some strong trend data: For example, based on the data I previously recorded, Stanford HAI’s 2026 AI Index shows that generative AI is expected to reach about 53% population-level adoption within three years, spreading faster than PCs and the internet. Microsoft’s 2026 Work Trend Index also emphasizes that AI Agents are being integrated into organizational workflows, with human value increasingly focused on goal-setting, judgment, oversight, and accountability for outcomes. But today, I also noticed a much cooler reality on the other side: Gartner predicted in June that over 40% of Agentic AI projects will be canceled by the end of 2027 due to rising costs, unclear business value, or insufficient risk control. This is actually quite similar to Crypto and blockchain—while the trend itself is undeniable, there’s a long road between that and its transformation into business value, company profits, or asset prices. I’m sharing this example because, when it comes to investing, it’s easy for us to jump straight from the narrative to taking positions. Right now, I feel the most important thing is to go through the chain of evidence, counterarguments, odds, and position sizing in between. Trends can only grant you the qualification to research, not the qualification to go all-in. So when it comes to AI investments, I now follow three principles: 1️⃣ Don’t equate “the tech trend is right” with “all related assets are worth buying.” The AI direction is very likely a long-term trend, but trends only grant you the qualification to research, not the qualification to go all-in. 2️⃣ Instead of chasing concepts, focus on who can actually capture cash flow—it’s more important. Whoever can lower costs, improve efficiency, lock in customers, and establish pricing power is closer to investment value. Many Agent projects will fail, but infrastructure, platform layers, workflow entry points, data, and security layers might actually benefit from the wave of trial and error. 3️⃣ Positions aren’t for expressing beliefs—they’re for testing hypotheses.