# Notes on building products people love when software gets cheap
**作者**: Hiten Shah
**日期**: 2026-08-15T16:27:49.000Z
**来源**: [https://x.com/hnshah/status/2088663964116983856](https://x.com/hnshah/status/2088663964116983856)
---
I keep thinking about Meta and how ruthless they have been for years about taking mechanics that work somewhere else and making them their own.
Stories worked, so they put Stories everywhere. Short form video worked, so they went hard on Reels. Public text worked, so they built Threads. There are plenty more examples.
People usually talk about this like Meta has no shame about copying, which is fair enough, but I think there is a much more useful lesson buried in what they have been doing.
Meta is a consumer company. Their business depends on understanding what gets people to create, share, connect, consume, come back, and generally spend more time with their products.
When someone else spends years and a bunch of money discovering a new behavior that people clearly like, why wouldn’t Meta study the hell out of it?
Someone else already did a ton of the expensive discovery work for them.
In a weird way, Meta has treated the rest of the consumer internet like a giant external product lab.
I think every software company is going to need to get much better at doing some version of this.
Software is becoming way too easy to build for product development to keep working the way it has.
A lot of the scarcity used to come from implementation. You could have ten ideas that sounded good and maybe have the engineering capacity to seriously pursue two of them. The cost of building forced a bunch of prioritization on you whether you liked it or not.
That is changing fast.
Agents can already build prototypes, instrument things, test implementations, review code, fix issues, and do more and more of the production work around software. I expect that to keep moving quickly.
Eventually “can we build this?” becomes a pretty boring question for a lot of software.
Then we are left with the question that has always been harder anyway.
Should we build it?
This is where I think things get interesting.
There are thousands and thousands of software companies running product experiments for us every day.
They are spending their own time and money figuring out better ways to onboard people, create habits, collaborate, review work, build trust, share things, discover things, verify things, delegate work, build reputation, create marketplaces, personalize experiences, give feedback, make something multiplayer, and pretty much every other behavior you can imagine inside software.
We get to watch all of it.
I don’t mean make a spreadsheet of competitor features and start cloning whatever looks cool. That is probably one of the fastest ways to turn a good product into a pile of shit.
The interesting part is figuring out the mechanic underneath the feature.
Take a pull request.
On the surface it is a developer tool feature. Underneath, someone proposes a piece of work, the work becomes visible to other people, feedback gets attached directly to it, people can inspect it asynchronously, approval changes its state, and the whole history sticks around afterward.
That mechanic can travel pretty far outside software development.
You could apply versions of it to AI generated work, marketing, finance, design, legal, operations, research, and plenty of other things.
Stories are another obvious example.
The interesting part was never just that a photo or video disappeared after 24 hours. The mechanic lowered the pressure to create, which made people more willing to share, then made consuming those updates incredibly easy, showed the creator who watched, gave people an easy way to reply, and created another reason to share again later.
There is a whole behavioral system underneath the visible feature.
Once you start looking at software this way, categories get a lot less useful.
A game can teach an enterprise product something. GitHub can teach an AI company something. TikTok can teach a research tool something. Marketplaces can teach agent products about trust. Consumer social products probably have a lot to teach B2B software about getting people to contribute more without making it feel like work.
You are basically looking for machinery that has already proven it can change human behavior.
Then you figure out whether that machinery can help with a behavior that matters inside your own product.
That part is important because every company has a different thing it should actually care about.
Meta has spent a lot of time caring about attention because attention is tied pretty directly to their business.
An AI product might care about whether people trust an agent enough to delegate increasingly important work to it.
A marketplace might care about every successful transaction making the next transaction easier to trust.
An analytics product might care about turning evidence into a decision and then getting that decision turned into an improvement.
A company building software for teams might care about work leaving enough useful context behind that someone else can pick it up later without needing another meeting.
Whatever it is, figure out the behavior first. Then go hunting for mechanics that might help create more of it.
I can see companies eventually having entire libraries of these things.
This mechanic increases contribution without requiring a giant reward.
This one reduces the anxiety of sharing unfinished work.
Another one makes verification more likely.
Some mechanics build reputation as a byproduct of normal use.
Others make a workflow multiplayer.
A few only work once you have enough network density.
Some work incredibly well in games and completely fall apart when you move them into productivity software.
Over time it probably becomes more like a graph than a library, especially because AI can help us watch so much more of the software world than a product team ever could on its own.
An agent could watch product launches, changelogs, screenshots, demos, reviews, user complaints, open source projects, App Store updates, and whatever else we can throw at it. It can start finding patterns, breaking mechanics apart, connecting them to behaviors we care about, and eventually suggesting how they might translate into our own products.
Then another agent can build the prototype.
That sounds a little sci-fi when you write it all out, but we are already close enough to pieces of this that I don’t think the direction is particularly hard to see.
The funny part is that all of this makes the human side more important to me.
Judgment gets more valuable when building gets cheaper.
Taste matters more when you can add almost anything.
Knowing which behavior actually matters becomes more important when you can optimize a hundred different ones.
Understanding why something worked somewhere else matters a lot more than noticing that it worked.
And knowing when to leave the damn product alone might become one of the most valuable product skills of all.
I also think this means ruthless prioritization gets more important, which sounds backwards at first.
For a long time engineering scarcity saved us from ourselves. A bunch of mediocre ideas never got built because there simply wasn’t time.
We are losing that protection.
We are about to become extremely good at building bad ideas quickly.
That is probably going to create a whole new class of bloated products made by teams that confuse the ability to build something with a reason for it to exist.
The companies I would bet on are the ones that get really good at the learning loop.
See something interesting somewhere in the world. Understand what is actually happening underneath it. Form a hypothesis about why it might matter for your users. Build the smallest thing that lets you test it. Put it in front of people. Measure what changed. Keep the learning whether the idea worked or not.
Then do it again.
Meta figured out a version of this because their business forced them to. Their old Little Red Book is full of ruthless prioritization, keep shipping, the quick inherit the earth, code wins arguments, everything is up for debate, and a bunch of other ideas that make a lot more sense to me when I look at them through this lens.
They built an organization that could watch human behavior change, move quickly when they saw something important, and keep learning.
AI is going to make some version of that useful for pretty much every software company.
The entire software industry is becoming an external R&D lab that you get to learn from.
Everyone else can help discover what works.
The code to try it yourself is getting cheaper every day.
Which means knowing what deserves to exist is going to matter a hell of a lot more.
## 相关链接
- [swyx reposted](https://x.com/swyx)
- [Hiten Shah](https://x.com/hnshah)
- [@hnshah](https://x.com/hnshah)
- [299K](https://x.com/hnshah/status/2088663964116983856/analytics)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [12:27 AM · Aug 16, 2026](https://x.com/hnshah/status/2088663964116983856)
- [299.8K Views](https://x.com/hnshah/status/2088663964116983856/analytics)
- [View quotes](https://x.com/hnshah/status/2088663964116983856/quotes)
---
*导出时间: 2026/8/16 12:31:12*
---
## 中文翻译
# 关于软件变得廉价时如何打造人们喜爱的产品的思考
**作者**: Hiten Shah
**日期**: 2026-08-15T16:27:49.000Z
**来源**: [https://x.com/hnshah/status/2088663964116983856](https://x.com/hnshah/status/2088663964116983856)
---
我一直在思考Meta，以及他们多年来是如何不遗余力地将其他地方行之有效的机制拿来为己所用。
Stories模式奏效了，于是他们把Stories功能铺得到处都是。短视频模式奏效了，于是他们全力投入Reels。公开文字模式奏效了，于是他们打造了Threads。这样的例子还有很多。
人们通常谈论这件事时，会说Meta在抄袭上毫无羞耻心，这么说也没错，但我认为他们所做的这些事情背后，隐藏着一个更有价值的教训。
Meta是一家面向消费者的公司。他们的业务依赖于理解是什么促使人们去创造、分享、连接、消费、回访，以及总体上花更多时间在他们的产品上。
当别人花费数年时间和大量金钱，发现了一种人们明显喜欢的新行为时，Meta为什么不好好研究个透呢？
别人已经替他们完成了大量昂贵的探索工作。
从某种奇怪的角度看，Meta把整个消费互联网当成了一个巨大的外部产品实验室。
我认为，每家软件公司都需要变得更擅长做类似的事情。
软件正变得太容易构建了，以至于产品开发不能再像以前那样运作。
过去，稀缺性很大程度上来自于实现层面。你可能会有十个听起来不错的主意，但只有能力认真推进其中两个。构建的成本迫使你进行优先级排序，不管你愿不愿意。
这种情况正在迅速改变。
智能体已经能够构建原型、进行数据埋点、测试实现、审查代码、修复问题，并承担越来越多围绕软件的生产工作。我预计这一趋势会持续快速发展。
最终，“我们能构建这个吗？”对于很多软件来说会变成一个相当无趣的问题。
到那时，我们只剩下那个一直更难回答的问题：
我们应该构建它吗？
我认为有趣的地方就在这里。
每天都有成千上万的软件公司在为我们进行产品实验。
他们花费自己的时间和金钱，去探索更好的方式来引导用户入门、培养习惯、协作、审查工作、建立信任、分享事物、发现事物、验证事物、委派工作、建立声誉、创建市场、个性化体验、提供反馈、让产品具备多人协作功能，以及你能想到的软件中几乎所有其他行为。
我们都能看到这一切。
我不是说要做一个竞争对手功能清单，然后开始克隆任何看起来酷的东西。那可能是把一个好产品变成一堆垃圾最快的方式之一。
有趣的部分在于弄清楚功能底下的机制。
以拉取请求（Pull Request）为例。
表面上，它是一个开发者工具功能。在底层，有人提出一项工作，这项工作对其他人变得可见，反馈直接附加在上面，人们可以异步审查，批准会改变其状态，之后整个历史记录都会被保留下来。
这种机制可以远远超出软件开发的范围。
你可以将它的各种变体应用于AI生成的工作、市场营销、财务、设计、法律、运营、研究以及许多其他领域。
Stories是另一个明显的例子。
有趣的部分从来不只是照片或视频在24小时后消失。这种机制降低了创作的压力，让人们更愿意分享，然后让消费这些更新变得极其容易，向创作者展示谁看过，给人们一种简单的回复方式，并创造了以后再次分享的另一个理由。
在可见的功能之下，有一个完整的行为系统。
一旦你开始用这种方式看待软件，类别的界限就变得不那么有用了。
一个游戏可以教给企业产品一些东西。GitHub可以教给AI公司一些东西。TikTok可以教给研究工具一些东西。市场平台可以教给智能体产品关于信任的知识。面向消费者的社交产品可能有很多东西可以教给B2B软件，关于如何让人们更多地贡献而不觉得像是在工作。
你基本上是在寻找那些已经被证明能够改变人类行为的机器。
然后你弄清楚那台机器是否能帮助实现你自己产品中重要的行为。
这部分很重要，因为每家公司都有自己真正应该关心的不同事物。
Meta花了很多时间关心注意力，因为注意力与他们的业务直接相关。
一个AI产品可能关心的是，人们是否足够信任智能体，从而将越来越重要的任务委派给它。
一个市场平台可能关心的是，每一笔成功的交易都能让下一笔交易更容易被信任。
一个分析产品可能关心的是，将证据转化为决策，然后让该决策转化为改进。
一家为团队构建软件的公司可能关心的是，工作是否留下了足够有用的上下文，让其他人之后可以接手，而无需再开一次会。
不管是什么，先弄清楚行为。然后去寻找可能有助于创造更多这种行为的机制。
我可以看到公司最终会拥有这些机制的完整库。
这个机制可以在不需要巨大奖励的情况下增加贡献。
这个机制可以减少分享未完成工作的焦虑。
另一个机制让验证更有可能发生。
有些机制将声誉建设作为正常使用的副产品。
还有一些机制让工作流程变得多人协作化。
有些机制只有在网络密度足够大时才能发挥作用。
有些机制在游戏中效果极佳，但当你把它们移到生产力软件中时，它们就完全失效了。
随着时间的推移，这可能会更像一个图谱而不是一个库，特别是因为AI可以帮助我们观察到比产品团队自己所能观察到的多得多的软件世界。
一个智能体可以观察产品发布、更新日志、截图、演示、评论、用户投诉、开源项目、App Store更新，以及我们能提供给它的任何其他内容。它可以开始发现模式，拆解机制，将它们与我们关心的行为联系起来，并最终建议它们如何转化到我们自己的产品中。
然后另一个智能体可以构建原型。
当你把这些都写出来时，听起来有点科幻，但我们已经足够接近这些碎片了，我认为这个方向并不难看清。
有趣的是，所有这些都让我觉得人的一面更加重要。
当构建变得更便宜时，判断力变得更有价值。
当你可以添加几乎任何东西时，品味变得更加重要。
当你能够优化一百种不同的行为时，知道哪种行为真正重要变得更加关键。
理解某件事为什么在别处奏效，比注意到它奏效了要重要得多。
而知道什么时候该让产品保持原样，可能会成为最有价值的产品技能之一。
我还认为这意味着无情的优先级排序变得更加重要，这乍听起来有点反直觉。
很长一段时间以来，工程资源的稀缺性让我们免于自食其果。一堆平庸的想法从未被构建，因为根本没有时间。
我们正在失去这种保护。
我们即将变得极其擅长快速构建糟糕的想法。
这可能会催生出一整类臃肿的产品，由那些把构建能力误认为存在理由的团队制造。
我会押注的公司，是那些真正擅长学习循环的公司。
在世界的某个角落看到有趣的东西。理解它底下实际发生了什么。形成一个关于它为什么可能对你的用户重要的假设。构建最小的东西来测试它。把它放到人们面前。衡量发生了什么变化。无论想法是否奏效，都保留这个学习成果。
然后再次重复。
Meta想通了这一点的一个版本，因为他们的业务迫使他们这样做。他们那本旧小红书里充满了无情的优先级排序、持续交付、敏捷者继承世界、代码赢得争论、一切都可以讨论，以及许多其他想法，当我透过这个视角来看时，这些想法对我来说更有意义了。
他们建立了一个组织，能够观察人类行为的变化，在发现重要事物时迅速行动，并不断学习。
AI将使某种形式的这种能力对几乎每家软件公司都变得有用。
整个软件行业正在成为一个你可以从中学习的外部研发实验室。
其他所有人都可以帮助发现什么有效。
自己尝试的代码成本每天都在降低。
这意味着，知道什么值得存在将变得无比重要。
## 相关链接
- [swyx reposted](https://x.com/swyx)
- [Hiten Shah](https://x.com/hnshah)
- [@hnshah](https://x.com/hnshah)
- [299K](https://x.com/hnshah/status/2088663964116983856/analytics)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [12:27 AM · Aug 16, 2026](https://x.com/hnshah/status/2088663964116983856)
- [299.8K Views](https://x.com/hnshah/status/2088663964116983856/analytics)
- [View quotes](https://x.com/hnshah/status/2088663964116983856/quotes)
---
*导出时间: 2026/8/16 12:31:12* 
