首页人物观点一些思考： 1. 在未来，一个事物完全由 AI 生成的概率，将与它的“预期生命周期”成反比。生命周期越短，越可能完全由 AI 生成；生命周期越长，越不可能。 2. 对于概念上比较简单、且注定短命的产物，人类依然会参与其中，只是参与的抽象层级发生了变化。比如我对 @Weavy_ai（Figma Weave）非常兴奋，因为它展示了这样一种可能性：把 AI 生成当作“可塑的黏土”，而不是最终成品。“工作流构建（workflow building）”正在成为一种全新的技能，值得探索和学习。 3. 如果你希望一个作品拥有很长的生命周期（比如软件、小说、电影），那么 AI 仍然可以在创作过程中辅助你。但你必须对作品投入极强的主观意图：你会反复思考不同路径，你会在意最微小的细节，你会真正沉浸在“工艺（craft）”之中。因为如果你不这样做：它不够好，它留不下来，它不会被注意，它不会被喜爱，它也不会重要。 4. 只聚焦软件来说——人们并不喜欢软件频繁改变。任何经历过一次大改版上线的人都知道这一点。所以，你当然可以在软件内部频繁生成新的内容，但你不会总是去重构软件的核心用户体验（UX）。否则，用户会非常反感。作为一个形象的比喻，可以把软件当成一栋房子。是的，你可能每周都会更换冰箱上的照片、纸条和磁铁；偶尔会重新整理、挪动家具；住久了之后，你也许会意识到空间使用上的问题，然后——带着强烈的目的性——进行一次翻修。房子里的某些部分（比如冰箱）变化很频繁，但房子的整体结构变化得要少得多。在思考“哪些东西会被 AI 生成”时，不要把整体和局部、长期和短暂混为一谈。 5. 从智识层面来看，思考“品牌是否会想为每个用户单独定制软件”是件很有意思的事。（当然，个人如果愿意，也可以为自己制作更多软件，比如 Figma Make。）但我目前非常强烈的直觉是：我们最终不会进入一个“品牌按用户逐一高度定制软件”的世界。原因在于：人们是通过其他人来学习如何使用软件的。 Snapchat 是一个极好的例子。对新用户来说Snapchat 其实相当让人困惑。你可以把这看成设计问题，也可以把它看成优势——我认为这是优势。通过坚持一套独特的交互模式，以及一个可学习但并非直觉化的界面，这个网络空间变得更加“有选择性、有意图”。如果你很年轻，你会通过看朋友怎么用 Snapchat来学会它；如果你年纪更大——那你可能本来就不是目标用户。 6. 最后总结一下：我们正处在一个软件数量呈指数级增长的世界。如果你想赢，设计才是真正的差异化武器。要投入到：设计、工艺、叙事以及鲜明而大胆的观点把 AI 当作工具来用，但不要指望它独立替你打造下一个伟大的产品。不要指望它创造出前所未见、前所未想的东西。那是你的工作。 Thoughts: 1. In the future, the probability something is generated entirely by AI will be inversely proportional to its intended lifespan. 2. For conceptually simple artifacts that are intended to have short lifespans, humans will still be involved just at a different level of abstraction. For example, I'm super excited about @Weavy_ai (Figma Weave) because it shows what's possible when you treat AI generation like clay to shape rather than the final output. Workflow building is a new skill to explore and learn. 3. If you intend for an artifact to have a long lifespan (ex: software, a novel, a movie), then AI might still aid you in your creative process. But you will bring great intention to the work. You will think through many different approaches. You will care about the smallest of details. You will lean into the craft. Because if you don't, it won't be good enough to last. It won't be noticed. It won't be loved. It won't matter. 4. Focusing just on software now... people don't like it when software changes. Everyone who has shipped a redesign knows this! So you might be generating new content within a piece of software frequently but of course you wouldn't redesign the fundamental UX of the software all the time. Users would hate it. As a grounding metaphor, consider a house. Yes, you might change the photos and papers and magnets stuck to your fridge a few times a week. Once in a while, you reorganize stuff or move furniture around. After living in the house for a while, you maybe notice issues around how you use the space and — with great intention — embark on a remodel. Some parts of the house, like the fridge, change a lot. But the overall structure of the house changes less. When asking what will be generated by AI, don't confuse the whole for the parts, the long lasting for the ephemeral. 5. It's intellectually interesting to think about whether a brand might want to adapt their software on a user by user basis. (Certainly individuals will be able to make more software for themselves if they are so inclined. For example, see Figma Make.) That said, my strong gut right now is that we will not end up in a world where brands customize software on a per user basis. People learn how to use software from other humans. Snapchat is a great example. For a new user, Snapchat is kind of confusing. You can see this as a design issue or an advantage... I argue it's an advantage. By leaning into custom patterns and a learnable (but arguably non-intuitive) interface, the resulting network is a more intentional space. If you're young, you'll learn how to use Snapchat by watching your friends use Snapchat. And if you're older, well, you might not be the intended demographic. 6. To wrap up... we are in a world where the amount of software is growing at an exponential rate. If you want to win, design is the differentiator. Invest in design, craft, storytelling and a bold point of view. Use AI as a tool, but don't expect it to build the next big thing for you on its own. Don't expect it to make something that no one has ever seen or imagined before. That's your job.

一些思考： 1. 在未来，一个事物完全由 AI 生成的概率，将与它的“预期生命周期”成反比。生命周期越短，越可能完全由 AI 生成；生命周期越长，越不可能。 2. 对于概念上比较简单、且注定短命的产物，人类依然会参与其中，只是参与的抽象层级发生了变化。比如我对 @Weavy_ai（Figma Weave）非常兴奋，因为它展示了这样一种可能性：把 AI 生成当作“可塑的黏土”，而不是最终成品。“工作流构建（workflow building）”正在成为一种全新的技能，值得探索和学习。 3. 如果你希望一个作品拥有很长的生命周期（比如软件、小说、电影），那么 AI 仍然可以在创作过程中辅助你。但你必须对作品投入极强的主观意图：你会反复思考不同路径，你会在意最微小的细节，你会真正沉浸在“工艺（craft）”之中。因为如果你不这样做：它不够好，它留不下来，它不会被注意，它不会被喜爱，它也不会重要。 4. 只聚焦软件来说——人们并不喜欢软件频繁改变。任何经历过一次大改版上线的人都知道这一点。所以，你当然可以在软件内部频繁生成新的内容，但你不会总是去重构软件的核心用户体验（UX）。否则，用户会非常反感。作为一个形象的比喻，可以把软件当成一栋房子。是的，你可能每周都会更换冰箱上的照片、纸条和磁铁；偶尔会重新整理、挪动家具；住久了之后，你也许会意识到空间使用上的问题，然后——带着强烈的目的性——进行一次翻修。房子里的某些部分（比如冰箱）变化很频繁，但房子的整体结构变化得要少得多。在思考“哪些东西会被 AI 生成”时，不要把整体和局部、长期和短暂混为一谈。 5. 从智识层面来看，思考“品牌是否会想为每个用户单独定制软件”是件很有意思的事。（当然，个人如果愿意，也可以为自己制作更多软件，比如 Figma Make。）但我目前非常强烈的直觉是：我们最终不会进入一个“品牌按用户逐一高度定制软件”的世界。原因在于：人们是通过其他人来学习如何使用软件的。 Snapchat 是一个极好的例子。对新用户来说Snapchat 其实相当让人困惑。你可以把这看成设计问题，也可以把它看成优势——我认为这是优势。通过坚持一套独特的交互模式，以及一个可学习但并非直觉化的界面，这个网络空间变得更加“有选择性、有意图”。如果你很年轻，你会通过看朋友怎么用 Snapchat来学会它；如果你年纪更大——那你可能本来就不是目标用户。 6. 最后总结一下：我们正处在一个软件数量呈指数级增长的世界。如果你想赢，设计才是真正的差异化武器。要投入到：设计、工艺、叙事以及鲜明而大胆的观点把 AI 当作工具来用，但不要指望它独立替你打造下一个伟大的产品。不要指望它创造出前所未见、前所未想的东西。那是你的工作。 Thoughts: 1. In the future, the probability something is generated entirely by AI will be inversely proportional to its intended lifespan. 2. For conceptually simple artifacts that are intended to have short lifespans, humans will still be involved just at a different level of abstraction. For example, I'm super excited about @Weavy_ai (Figma Weave) because it shows what's possible when you treat AI generation like clay to shape rather than the final output. Workflow building is a new skill to explore and learn. 3. If you intend for an artifact to have a long lifespan (ex: software, a novel, a movie), then AI might still aid you in your creative process. But you will bring great intention to the work. You will think through many different approaches. You will care about the smallest of details. You will lean into the craft. Because if you don't, it won't be good enough to last. It won't be noticed. It won't be loved. It won't matter. 4. Focusing just on software now... people don't like it when software changes. Everyone who has shipped a redesign knows this! So you might be generating new content within a piece of software frequently but of course you wouldn't redesign the fundamental UX of the software all the time. Users would hate it. As a grounding metaphor, consider a house. Yes, you might change the photos and papers and magnets stuck to your fridge a few times a week. Once in a while, you reorganize stuff or move furniture around. After living in the house for a while, you maybe notice issues around how you use the space and — with great intention — embark on a remodel. Some parts of the house, like the fridge, change a lot. But the overall structure of the house changes less. When asking what will be generated by AI, don't confuse the whole for the parts, the long lasting for the ephemeral. 5. It's intellectually interesting to think about whether a brand might want to adapt their software on a user by user basis. (Certainly individuals will be able to make more software for themselves if they are so inclined. For example, see Figma Make.) That said, my strong gut right now is that we will not end up in a world where brands customize software on a per user basis. People learn how to use software from other humans. Snapchat is a great example. For a new user, Snapchat is kind of confusing. You can see this as a design issue or an advantage... I argue it's an advantage. By leaning into custom patterns and a learnable (but arguably non-intuitive) interface, the resulting network is a more intentional space. If you're young, you'll learn how to use Snapchat by watching your friends use Snapchat. And if you're older, well, you might not be the intended demographic. 6. To wrap up... we are in a world where the amount of software is growing at an exponential rate. If you want to win, design is the differentiator. Invest in design, craft, storytelling and a bold point of view. Use AI as a tool, but don't expect it to build the next big thing for you on its own. Don't expect it to make something that no one has ever seen or imagined before. That's your job.

Dylan FieldCo-founder & CEO, Figma

2026/01/08 22:41 UTC

观点原文

一些思考：

在未来，一个事物完全由 AI 生成的概率，将与它的“预期生命周期”成反比。生命周期越短，越可能完全由 AI 生成；生命周期越长，越不可能。

对于概念上比较简单、且注定短命的产物，人类依然会参与其中，只是参与的抽象层级发生了变化。比如我对 @Weavy_ai（Figma Weave）非常兴奋，因为它展示了这样一种可能性：把 AI 生成当作“可塑的黏土”，而不是最终成品。“工作流构建（workflow building）”正在成为一种全新的技能，值得探索和学习。

如果你希望一个作品拥有很长的生命周期（比如软件、小说、电影），那么 AI 仍然可以在创作过程中辅助你。但你必须对作品投入极强的主观意图：

你会反复思考不同路径，

你会在意最微小的细节，

你会真正沉浸在“工艺（craft）”之中。

因为如果你不这样做：

它不够好，

它留不下来，

它不会被注意，

它不会被喜爱，

它也不会重要。

只聚焦软件来说——人们并不喜欢软件频繁改变。

任何经历过一次大改版上线的人都知道这一点。所以，你当然可以在软件内部频繁生成新的内容，但你不会总是去重构软件的核心用户体验（UX）。否则，用户会非常反感。

作为一个形象的比喻，可以把软件当成一栋房子。

是的，你可能每周都会更换冰箱上的照片、纸条和磁铁；偶尔会重新整理、挪动家具；住久了之后，你也许会意识到空间使用上的问题，然后——带着强烈的目的性——进行一次翻修。

房子里的某些部分（比如冰箱）变化很频繁，但房子的整体结构变化得要少得多。

在思考“哪些东西会被 AI 生成”时，不要把整体和局部、长期和短暂混为一谈。

5. 从智识层面来看，思考“品牌是否会想为每个用户单独定制软件”是件很有意思的事。（当然，个人如果愿意，也可以为自己制作更多软件，比如 Figma Make。）

但我目前非常强烈的直觉是：我们最终不会进入一个“品牌按用户逐一高度定制软件”的世界。

原因在于：人们是通过其他人来学习如何使用软件的。

Snapchat 是一个极好的例子。对新用户来说Snapchat 其实相当让人困惑。你可以把这看成设计问题，也可以把它看成优势——我认为这是优势。

通过坚持一套独特的交互模式，以及一个可学习但并非直觉化的界面，这个网络空间变得更加“有选择性、有意图”。如果你很年轻，你会通过看朋友怎么用 Snapchat来学会它；如果你年纪更大——那你可能本来就不是目标用户。

最后总结一下：我们正处在一个软件数量呈指数级增长的世界。如果你想赢，设计才是真正的差异化武器。

要投入到：设计、工艺、叙事以及鲜明而大胆的观点

把 AI 当作工具来用，但不要指望它独立替你打造下一个伟大的产品。不要指望它创造出前所未见、前所未想的东西。

那是你的工作。

Thoughts:

In the future, the probability something is generated entirely by AI will be inversely proportional to its intended lifespan.

For conceptually simple artifacts that are intended to have short lifespans, humans will still be involved just at a different level of abstraction. For example, I'm super excited about

@Weavy_ai

(Figma Weave) because it shows what's possible when you treat AI generation like clay to shape rather than the final output. Workflow building is a new skill to explore and learn.

If you intend for an artifact to have a long lifespan (ex: software, a novel, a movie), then AI might still aid you in your creative process. But you will bring great intention to the work. You will think through many different approaches. You will care about the smallest of details. You will lean into the craft. Because if you don't, it won't be good enough to last. It won't be noticed. It won't be loved. It won't matter.

Focusing just on software now... people don't like it when software changes. Everyone who has shipped a redesign knows this! So you might be generating new content within a piece of software frequently but of course you wouldn't redesign the fundamental UX of the software all the time. Users would hate it.

As a grounding metaphor, consider a house. Yes, you might change the photos and papers and magnets stuck to your fridge a few times a week. Once in a while, you reorganize stuff or move furniture around. After living in the house for a while, you maybe notice issues around how you use the space and — with great intention — embark on a remodel.

Some parts of the house, like the fridge, change a lot. But the overall structure of the house changes less. When asking what will be generated by AI, don't confuse the whole for the parts, the long lasting for the ephemeral.

It's intellectually interesting to think about whether a brand might want to adapt their software on a user by user basis. (Certainly individuals will be able to make more software for themselves if they are so inclined. For example, see Figma Make.)

That said, my strong gut right now is that we will not end up in a world where brands customize software on a per user basis.

People learn how to use software from other humans. Snapchat is a great example. For a new user, Snapchat is kind of confusing. You can see this as a design issue or an advantage... I argue it's an advantage.

By leaning into custom patterns and a learnable (but arguably non-intuitive) interface, the resulting network is a more intentional space. If you're young, you'll learn how to use Snapchat by watching your friends use Snapchat. And if you're older, well, you might not be the intended demographic.

To wrap up... we are in a world where the amount of software is growing at an exponential rate. If you want to win, design is the differentiator. Invest in design, craft, storytelling and a bold point of view.

Use AI as a tool, but don't expect it to build the next big thing for you on its own. Don't expect it to make something that no one has ever seen or imagined before. That's your job.

ABAB AI 解读

这组“Thoughts”不是在讨论“AI 能做什么”，而是在给出一套判断“什么值得由人类主导、什么可以交给 AI”的长期方法论。它的核心不是技术，而是时间、意图与审美。

我从时间维度 → 人类角色迁移 → 产品与品牌 → 设计作为护城河四个层级解读，并结合具体案例。

一、第一性判断：生命周期决定创作主权

完全由 AI 生成的概率 ∝ 1 / 预期生命周期

这是一个极其有力的判据。

短命产物（临时海报、一次性原型、活动页、A/B 测试素材）

→ AI 全权生成，速度与覆盖最重要。

长命产物（软件、小说、电影、品牌系统）

→ AI 只能辅助，人类必须主导。

📌 时间是过滤器：越想“留下来”，越需要人类的强意图与审美取舍。

二、人类角色的迁移：从“出图的人”到“塑形的人”

第 2 点的关键不在“生成”，而在把生成当作“黏土”。

以 Figma Weave / Make 为例：价值不在“AI 给你一个最终稿”，而在你能否构建一个可反复试错、可迭代的工作流。

这意味着新技能栈出现了：

Workflow Building（工作流构建）
约束、反馈、选择与组合
把“生成”变成“可控材料”

📌 高手不再比“谁出图快”，而比“谁更会塑形”。

三、为什么“长寿作品”必须回到工艺（craft）？

第 3 点说的是一个残酷现实：如果你不在意细节，作品不会被注意、不会被爱、也不会重要。

AI 能给你：80 分的“合理答案”，但永远给不了“值得被记住的偏执”

历史反复证明：

经典软件 → 对细节的长期克制与一致性
经典小说/电影 → 明确的作者意图与审美选择

📌 “意图密度”是长寿作品的核心变量。

四、软件的反直觉：生成≠频繁重构 UX

第 4 点用“房子”的隐喻非常准确。

冰箱上的照片、磁贴 → 可频繁变（内容层）
房子的结构、动线 → 极少变（体验层）

软件亦然：

内容可以天天变（文案、推荐、素材）
核心 UX 不能频繁变，否则用户会厌恶

📌 把“整体”和“局部”、“长期”和“短期”混为一谈，是很多 AI 产品失败的原因。

五、为什么“按用户定制软件”大概率走不通？

第 5 点给了一个少见但重要的判断。

直觉上，AI 似乎应该“千人千面”。但现实的学习路径是：人们通过“看别人怎么用”来学软件。

以 Snapchat 为例：

新用户觉得“反直觉”
但这反而形成了社会学习与代际筛选
使用门槛塑造了更有意图的网络

📌 过度个性化会破坏“共同语言”，从而削弱网络效应。

六、终局判断：软件指数增长，设计成为唯一差异化

第 6 点给出结论：当软件数量指数级增长时，设计、叙事与观点，决定谁能被记住。

AI 会：降低生产门槛、放大“平均水平”

而胜负手在于：设计的一致性、叙事的清晰度、观点的大胆性

📌 AI 是放大器，不是原创者。

七、可执行的判断清单（把方法论落地）

在任何一个项目上，问自己四个问题：

它打算活多久？

短 → AI 主导
长 → 人类主导

AI 是终稿，还是材料？

终稿 → 风险高
材料 → 可控

哪些部分可以常变，哪些必须稳定？

内容变、结构稳，是否需要“共同学习”？需要 → 避免过度个性化

一句最高认知总结

AI 会极大地加速“生成”，但真正决定价值的，仍然是：

你选择留下什么、舍弃什么、坚持什么。

不要指望 AI 帮你做“前所未见”的事。那仍然是人的责任——也是人类在 AI 时代不可被替代的核心工作。