Podcast Atlas
entity Updated 2026-07-08 Tags: Person, Investor, Ai

Dai Yusen / 戴雨森

Dai Yusen, also called Yusen in the source, is the guest in 142. 雨森的创投观察第2集：Harness、下一个字节、2026大机会和Stanley Druckenmiller. The episode presents him as an investor thinking through AI venture opportunities, public-market exposure, model-company competition, Agent Harness value, and the shape of AI-native startups after agents become more capable.

His operating style in the episode is “strong opinion, weakly held”: he says early investors must make judgments before all evidence is available, but should update quickly when new facts change the base rate. He says his earlier caution about consumer AI return was partly right, while his underestimation of agentic coding was clearly wrong once Claude Code, Codex, and long-running coding agents became more capable.

Key Points

  • Uses Stanley Druckenmiller as a secondary-market role model, especially for trading orientation and willingness to change views.
  • Treats AI return as a chain from token input to software output to business result, which connects his view to AI Investment Metrics and AI Economic Diffusion.
  • Argues that Agent Harness products can be durable because memory, context, tools, workflow habits, and user data accumulate outside the base model.
  • Frames 2026 as a period when agents may move from demos into actual task completion, making Agentic Workflow and AI Organization Design more important.
  • Warns founders not to copy ByteDance’s mobile-internet playbook when looking for the next major AI company.
  • Sees future opportunity in Agent Marketplace patterns, where agents with different context and skills can transact or collaborate.
  • Emphasizes human responsibility, trust, question-asking, and agency even as more cognitive execution becomes automatable.

Connections