Podcast Atlas

130. 张月光创业两年首次访谈：妙鸭不是AI Native产品、流程到上下文设计、One Way Door和乙女游戏

source Updated 2026-07-08 Tags: Podcast, Ai, Product, Agents, Games

Summary

This 张小珺Jùn｜商业访谈录 episode interviews 张月光 about leaving Alibaba, the explosion of 妙鸭, and the two years of exploration that led toward AI Friend Products, AI Otome Games, and Docky. The central product argument is that 妙鸭 was a strong internet product powered by AI, but not fully AI Native Product Design: AI-native products shift from deterministic flow design toward Context Engineering, open input/output, model boundaries, and mixed product-design-engineering exploration. The episode also adds One Way Door Product as Zhang’s standard for products that make users unwilling to return to the old workflow.

Key Claims

  • 张月光 frames his career path through Alipay, ByteDance, Alibaba, a first startup, and 妙鸭 as a search for a long-term product direction rather than a sequence of disconnected launches.
  • He says his first startup tried short-video commerce too early and from the wrong angle, because selling through short video was more naturally an extension of short-video platforms than a separate recreated platform.
  • His earlier Live2D product 原因 convinced him that virtual companions could create real user emotion, payment, and ritual behavior, later feeding his interest in AI Otome Games.
  • The Kyoto decision in October 2023 is presented as a personal operating reset: from ages 35 to 45 he wants to work with the same group on the same thing, which pushed him out of large-company assignment logic.
  • He distinguishes large-company promotion through external battles from promotion through internal competition, and says AI reopened the kind of external game he prefers.
  • 妙鸭 chose AI portraits because the technology was mature enough, the category was narrow, and consumer portrait photography had direct pricing logic compared with ecommerce or model images.
  • 妙鸭’s quality target was “true, alike, beautiful”: the image should not look AI-generated, should resemble the user, and should make the user look better than ordinary photos.
  • Zhang now argues that 妙鸭 was not fully AI-native: requiring AI is necessary, but not sufficient, because the underlying product still resembled a bounded internet flow with a known output.
  • His AI-native product-design distinction is that users can enter open-ended language or intent, outputs are less predetermined, and the product must first ask what context the model needs to produce a good result.
  • He argues AI-native teams cannot rely only on a linear “product writes requirements, design makes mockups, engineering implements” pipeline; product, design, and engineering have to jointly define taste, context, model effect, and boundary conditions.
  • He interprets 妙鸭 as a portrait business and possible new consumer brand, with potential for higher pricing, human-in-the-loop service, and offline pop-up pressure on traditional studios.
  • His second startup began with the idea of “AI population”: AI should not only be a tool or service, but can become an individual-like product object.
  • He argues AI companionship has real commercial demand but faces high user-entry cost, weak monetization, and stagnant character growth; game structure can lower entry, improve monetization, and provide character progression.
  • In AI Otome Games, he treats IP as a bundle of values, visual packaging, authored expression, and user interaction; AI is more suited to daily reachable interactive IP than to low-frequency, star-like premium IP alone.
  • Docky starts with PPT because existing AI PPT products have not become One Way Door Product experiences: templates constrain content, code-generated slides are hard to edit, and image-generated slides look polished but are not editable.
  • For Docky, Zhang wants to polish the smallest generated unit and the needed context before reverse-engineering the user flow.
  • He frames agents less as long offline task runners and more as short, high-frequency, low-latency loops that help users cross ability boundaries.
  • The company is now all in on Docky, except for the longer-line game direction, because Zhang sees “an AI friend that helps me break through ability boundaries” as the startup meaning he is willing to commit to.

Key Quotes

“真、像、美” — 张月光 on 妙鸭’s portrait-quality standard.

“没有这一代 AI 技术就做不了” — described as necessary but not sufficient for AI Native Product Design.

“One Way Door” — Zhang’s product standard for a new solution users do not want to reverse.

“能力突破” — the agent value Zhang wants Docky to pursue.

Connections

Contradictions

  • No direct contradiction with prior wiki content. The source reinforces existing Context Engineering, AI Interactive Entertainment, and Agentic Workflow themes while adding a sharper product-design distinction: AI capability alone does not make a product AI-native.
  • The source qualifies broad “AI game” optimism by arguing that current AI reliably helps only narrow play spaces, with female-oriented dialogue companionship as the most credible near-term game landing point.