张小珺Jùn｜商业访谈录

142. 雨森的创投观察第2集：Harness、下一个字节、2026大机会和Stanley Druckenmiller

2h 19min

In this episode, early-stage investor Dai Yusen reflects on his previous prediction of a 2026 AI return slowdown and acknowledges being wrong. He discusses the rapid evolution of AI models, the growing importance of agentic products like Harness over raw models, and identifies three major investment themes for 2026: coding agents, world models, and auto-research. He emphasizes that founders, not trends, drive his investments and that the next ByteDance will look nothing like ByteDance.

Summarized by Podsumo

🎧 Listen 🎙️ Ask about this episode

✨ Key Takeaways

1

Dai Yusen admits his 2026 AI return prediction was 'wrong' and explains why being wrong is a positive sign of learning and industry change.
2

He argues that the biggest opportunity in AI is shifting from model strength to agentic products (e.g., Harness) as the operating system layer.
3

He predicts the next ByteDance will not resemble ByteDance's short-video model but will emerge from AI-native usage patterns.
4

He highlights three 2026 mega-trends: coding agents, world models for robotics, and auto-research for AI self-improvement.
5

He warns against chasing AI hardware fads, calling many VC-backed 'wearable AI devices' solutions in search of a problem.

💬 Notable Quotes

"Being wrong repeatedly is a normal state for early-stage investors—it means the industry is changing fast and there are many opportunities."
"The real value in AI is shifting from the model to the agentic layer that orchestrates it. Harness is the new OS, not the model itself."
"If you think AI will just replace coders, you miss the point. It’s about enabling people to do things they couldn’t before, but the bottleneck is now imagination, not code."