Today, when chatting with GPT, I noticed an interesting contrast behind the AI boom.

On one side, there are strong trend data:
For example, the data I previously recorded shows that Stanford HAI's 2026 AI Index indicates generative AI will reach about 53% adoption at the population level within three years, spreading faster than PCs and the internet;
Microsoft's 2026 Work Trend Index also emphasizes that AI Agents are entering organizational workflows, with human value increasingly focused on goal setting, judgment, supervision, and accountability for results.
But today, I also see a very calm 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 quite similar to crypto and blockchain — a trend that is very certain, but transforming into business value, company profits, and asset prices involves a long road in between.
Talking about this case, I feel that when investing, it’s easy to jump directly from the narrative to position sizing.
Now, I believe the most important thing is that there must be an evidence chain, counter-evidence, odds, and position matching in between.
Trends can only provide research qualification; they cannot directly justify heavy positions.
So, for AI investing, I now have three disciplines:
1️⃣ Don’t equate “correct technological trend” directly with “all related assets are worth buying.”
AI is likely a long-term trend, but a trend only provides research qualification; it cannot directly justify heavy positions.
2️⃣ Instead of chasing concepts, it’s more important to see who can truly benefit from cash flow.
Who can reduce costs, improve efficiency, lock in customers, and establish pricing power — those are closer to investment value. Many Agent projects will fail, but infrastructure, platform layers, workflow entry points, data, and security layers might benefit from the trial-and-error wave.
3️⃣ Position size is not used to express faith but to verify hypotheses.