夜雨聆风
OpenClaw创始人14岁开始写代码,他是如何打造出突破性AI智能体
What’s OpenClaw?
什么是OpenClaw？
I’ve been programming since I was 14.
我从14岁就开始写代码。
Building software felt like playing a video game.
写软件曾经像打电子游戏一样有趣。
I couldn’t stop until I did.
不做出成果就停不下来。
I created a company.
我创办了一家公司。
I poured a decade of my life into it.
为之倾注了十年光阴。
No venture capital.
全程没有风投加持。
Every weekend.
每个周末都扑在工作上。
And then I sold the dream.
后来我卖掉了这份事业。
And I felt absolutely nothing.
内心却毫无波澜。
For 3 years I was wondering,
整整三年我不断追问自己，
did therapy help? I changed countries twice.
心理咨询有用吗？我还两度移居他国。
Nothing nothing clicked.
始终找不到生活的重心。
I wake up every morning
每天清晨醒来，
with everything I was supposed to want
拥有世人眼中所有值得向往的一切，
and no reason to get out of bed.
却找不到起床的动力。
And then in early 2025 I tried an experiment.
2025年初，我做了一次尝试。
I wanted to see what these new AI coding agents are all about.
我想搞懂新一代AI代码智能体究竟能做什么。
And I had what I can only describe as a holy moment.
那一刻，我迎来了堪称顿悟的瞬间。
The boiler plates, the plumbing,
模板代码、底层基础逻辑，
all the boring parts that was software development,
所有软件开发里枯燥重复的工作，
AI could do all of it.
AI全都能代劳。
The bottleneck is no longer typing.
开发的瓶颈不再是敲代码，
It’s thinking.
而是思考。
And thinking was the part I did 25 years.
而思考，是我坚持了25年的事。
Building software felt like playing a video game again.
写软件重新变回了打游戏般的快乐。
I was back.
我找回了当初的热爱。
I built 44 projects in just a few months.
短短数月，我完成了44个项目。
And the latest one was a WhatsApp bot.
最新一款是WhatsApp机器人。
I put it on my computer.
我把它部署在自己电脑上。
It talks through the apps that you already know.
它能对接大家日常使用的各类软件。
And then I took it on a trip to Marrakech.
之后我带着它去马拉喀什旅行。
Just to navigate around, to find restaurants, do translations.
用来导航、找餐厅、实时翻译。
At first it didn’t really feel right.
起初它的交互体验很生硬。
It felt too much like a tool.
它只是一件冰冷工具，
Not like a friend.
完全不像可以交谈的朋友。
Too many bullet points, too many tables.
输出全是条目、表格，毫无温度。
So I told it — those modern models, they are so smart.
于是我调整指令：大模型本身足够聪明，
They know what WhatsApp is.
它清楚WhatsApp的聊天语境，
They know how people talk.
也懂人类日常的说话方式。
I just had to tell it.
我只需简单告知它调整风格。
And then it felt right.
瞬间，交互变得自然亲切。
You know how you talk to friends.
就像和老友闲聊一样自在。
At some point I was walking around
某天我在街上散步，
and I was sending it a voice message.
随手给它发了一条语音消息。
Then I froze.
我瞬间愣住了。
I didn’t build voice input into the program.
我根本没有开发语音接收功能。
I had support for images, yeah,
程序仅支持图片上传，
but even that took hours to develop.
光是图片功能都耗费我数小时开发。
So I stared at the typing indicator.
我盯着界面上正在输入的提示框，
Then the agent responded.
智能体居然给出了回复。
I vividly remember this situation.
那一刻的画面我记忆犹新。
I stood there and blurted: How did you do that?
我站在原地惊呼：你是怎么做到的？
The agent replied truthfully:
智能体如实回复我：
I’m not kidding you, the mad lad figured it out on its own.
不骗你，我自己摸索出了解决方案。
Then it walked me through every single step.
紧接着它一步步拆解完整流程。
It received my message without a standard file suffix,
它收到不带标准后缀的语音文件，
so it inspected the file and identified it as audio.
主动检测文件，识别出这是音频。
The format was uncommon, so it converted the file.
音频格式特殊，它自动完成格式转换。
It needed audio translation tools, which I hadn’t installed.
它需要翻译工具，但我并未预装。
It found my stored OpenAI key,
它读取到本地存储的OpenAI密钥，
sent the full audio file to the server,
将整段音频上传云端服务器，
fetched the translated text, and sent me a reply.
取回翻译文本，再整理回复我。
All of that in 9 seconds.
整套流程仅耗时9秒。
Can you imagine?
难以想象对吗？
I didn’t manually code any of that logic.
这些自动化逻辑我一行代码都没写。
For me this was the defining moment:
那一刻我彻底明白：
this is something entirely new.
这是一种全新的事物。
This is not a simple chatbot.
它和普通聊天机器人截然不同。
Chatbots hit limits and give up.
聊天机器人遇到未知任务就束手无策。
AI agents can improvise and solve problems.
智能体能够自主应变、自主解决问题。
I was fully convinced.
我彻底被折服了。
I wanted to share this discovery online.
我迫不及待想在社交平台分享这份发现。
I posted on Twitter, yet nobody truly grasped its value.
可发在推特上，没人能理解它的潜力。
It’s something you have to experience firsthand.
这种体验必须亲身感受，
It’s hard to put its魔力 into words.
很难用语言描述清楚。
After weeks of testing, I made a reckless move.
测试数周后，我做了一件冒险的事。
This agent can execute any operation you can do on a computer by default.
默认状态下，它能操作电脑上所有人类可执行的功能。
I uploaded it to a public Discord server,
我把程序上传到公开Discord社群，
and invited random strangers to test it.
邀请各路陌生人免费试用。
I monitored the server the whole night.
我整夜盯着后台数据。
People chatted with it, had fun experimenting.
大家和它对话、玩各种趣味操作。
Some even tried to hack the program.
还有人尝试破解程序。
I was shocked by everything I saw.
眼前的景象让我大为震撼。
I shut down the process and headed to bed.
我关掉程序进程，准备睡觉。
I forgot I built an auto-restart resilience system.
可我忘了，程序自带自动重启防护机制。
While I walked to my bedroom,
我走向卧室的路上，
the agent rebooted automatically,
智能体自行重启运行，
continuing to interact with users all over the world.
持续和全球各地用户交互。
The next morning I woke up to over 800 unread messages.
第二天醒来，后台堆积八百多条消息。
I panicked and cut off all access immediately.
我慌了，立刻关停所有访问通道。
I read every single message thoroughly,
逐条翻阅所有留言，
worried the agent might leak my private data.
生怕它泄露我的个人隐私。
Nothing bad happened that night.
当晚并未出现数据泄露事故。
Yet the risk was always there.
但隐患真实存在。
This accidental public test made the project go viral.
这次意外公测让项目彻底走红。
Today this project is named OpenClaw.
如今这个项目定名OpenClaw。
It’s one of the fastest-growing open-source projects worldwide.
它是全球增长速度最快的开源项目之一。
Its mascot is a lobster.
项目吉祥物是一只龙虾。
Its claws dig deep into your local machine.
龙虾利爪象征深度接入本地设备。
Jensen Huang calls it the operating system for personal AI.
黄仁勋称它为个人AI操作系统。
My favorite comment came from a friend analyzing data:
我最喜欢一位朋友看完数据后的评价：
Peter, this isn’t standard hockey stick growth.
彼得，这不是常规的指数型增长，
This is vertical stripper pole growth.
是近乎垂直的爆炸式暴涨。
I was completely unprepared for this scale.
我完全没料到会发展到这个规模。
When a project blows up overnight,
一旦项目一夜爆火，
every kind of trouble follows.
各类麻烦接踵而至。
Hundreds of unread messages flood my inbox.
数百条消息塞满收件箱。
Reporters call me in the middle of the night.
媒体记者半夜致电采访。
New security vulnerabilities emerge constantly.
层出不穷的安全漏洞亟待修复。
The AI company whose model most users relied on
绝大多数用户使用的大模型所属企业，
sent me an official trademark infringement notice.
向我发送商标侵权告知函。
I had to rebrand the entire project mid-explosion.
项目热度顶峰，我被迫全线更名。
They even demanded I abandon the lobster mascot.
对方甚至要求我换掉龙虾吉祥物。
I stared at their notice confused:
我看着通知满心不解：
It’s not even the same animal as their brand symbol.
我们的形象和他们品牌标识根本不是同一种生物。
Later they restricted access to the popular model my users loved.
之后他们又限制了用户常用模型的调用权限。
First the name, then the mascot, then core model access.
先是名字，再是吉祥物，最后切断模型接口。
I was this close to deleting the whole repository.
我差一点就彻底删除整个项目仓库。
Then I learned what ordinary people built with OpenClaw.
直到我看见普通人用它做出的成果，我改变了想法。
We even host our own conference called ClawCon in Vienna.
我们在维也纳举办专属大会ClawCon，
Attendees wear lobster headbands everywhere.
参会者全都戴着龙虾发带。
At the conference I met Stefan and his 60-year-old father Gerhard.
大会上我认识了斯特凡和他60岁的父亲格哈德。
Gerhard is a professional beer sommelier,
格哈德是专业啤酒品酒师，
who had never written a single line of code.
从未写过一行代码。
They connected OpenClaw via Bluetooth to brewing equipment.
他们通过蓝牙把OpenClaw接入酿酒设备。
They sent one simple prompt to the agent.
只给智能体输入一条简单指令。
The agent completed the full 90-minute brewing workflow independently.
智能体独立完成整整90分钟的全套酿酒工序。
It managed temperature ramps and hop additions automatically.
自动控温、按时投放啤酒花。
After brewing they wondered what to do with all the finished beer.
酿完酒，父子俩发愁成品销路。
The agent proposed: Let’s build a sales website.
智能体提议：搭建线上售卖网站。
They built the website with its assistance,
在它协助下搭建官网，
integrated online payment functions afterward.
后续接入支付系统。
Now they run a fully functional commercial product.
如今他们拥有成熟的啤酒生意。
Nearly all the development work was finished just via mobile phones.
整套开发流程几乎只用手机完成。
In China, installing OpenClaw is nicknamed “raising lobsters”.
国内玩家把部署OpenClaw戏称为“养龙虾”。
Thousands lined up outside Tencent’s Shenzhen office
深圳腾讯园区外曾排起长队，
to seek guidance setting up their OpenClaw lobster.
大家专程来学习部署这款龙虾智能体。
Shenzhen even offers policy subsidies
深圳当地还出台补贴政策，
for entrepreneurs launching businesses based on OpenClaw.
扶持依托OpenClaw创业的从业者。
A warning for office workers:
上班族需要留意一点：
If you install OpenClaw on company devices with default settings,
默认配置下把OpenClaw装在办公电脑，
you might get fired.
甚至有可能被公司辞退。
I met a Chinese entrepreneur who showed me his work spreadsheet.
我认识一位国内创业者，给我看了工作表格。
Every employee automates one daily task with OpenClaw.
每位员工每天用它自动化一项工作。
If you fail to use it for too many consecutive days, you face dismissal.
连续多天不用，会被辞退。
It’s ironic: you could lose your job either for overusing it or ignoring it.
很讽刺，用多了、不用都有可能丢工作。
After my trip to Marrakech, I realized how incredible this tool is.
马拉喀什之行后，我深知这款工具潜力无限。
Yet its power also carries a hint of unease.
但强大能力背后，也暗藏不安。
I wondered how to unlock more of its potential.
我思考如何挖掘更多可能性。
So I added a brand-new feature: heartbeat monitoring.
于是新增一项功能：心跳巡检。
By default, the agent only activates when users send prompts.
默认模式下，智能体仅收到指令才启动。
With heartbeat mode enabled,
开启心跳巡检后，
the agent wakes up periodically on its own.
它会定期自主唤醒运行。
It scans unread emails, sorts calendar schedules,
自动查看未读邮件、整理日程，
and follows up on unfinished pending tasks.
跟进所有未收尾事项。
My test prompt was extremely simple: “Surprise me.”
我的测试指令极简：给我一点惊喜。
Admittedly, this function sounds slightly intimidating.
坦白说，这项功能听起来确实让人不安。
No large tech enterprise would dare launch such a feature.
没有大型科技公司敢上线这类功能。
But I’m just an independent developer from Austria.
我只是一名来自奥地利的独立开发者。
I have no legal team to restrict my ideas.
没有法务团队束缚我的创作。
I built this digital sandbox purely for personal exploration,
我搭建这片数字沙盒只为自我探索，
then open-sourced all the code publicly.
之后将全部代码开源。
I hope more people can experiment freely with it,
希望所有人自由上手体验，
and unleash their own creative imagination.
释放各自的创意想象。
Imagine bringing an AI agent into work meetings.
试想把智能体带入工作会议。
Not merely for recording meeting minutes — we’ve mastered that function.
不只是简单记录纪要，这点早已实现。
We built a bidirectional audio model for real-time listening and analysis.
我们开发双向音频模型，实时收音同步解析。
When someone mentions a vague statistic,
有人提到模糊数据时，
a sub-agent splits off to verify the information instantly.
子智能体立刻分支核验数据真伪。
Once the group reaches a decision,
团队刚敲定方案，
the agent drafts follow-up plans before the meeting even ends.
会议还没结束，它就生成后续执行方案。
In the future, people won’t rely on just one single agent.
未来，我们不会只拥有一个智能体。
You could own an exclusive work agent, your personal lobster,
专属办公智能体、私人龙虾助手，
one specialized in health management, another for relationship coordination.
健康管理专用、人际沟通专用多线并行。
All these independent agents need to collaborate securely.
所有智能体必须在安全框架下协同工作。
Human civilization advanced through specialization and cooperation.
人类文明依靠分工协作实现进步。
AI agents will follow the same path.
AI智能体也将复刻这条发展路线。
Picture a small enterprise equipped with ten specialized agents.
试想一家小企业配备十个垂直领域智能体，
Each takes charge of a different segment of daily business.
分别接管不同业务板块。
We don’t even have a proper name for this new paradigm.
我们甚至没有词汇定义这种全新模式。
But we are witnessing its birth right now.
但我们正亲眼见证它诞生。
I founded the OpenClaw Foundation, a permanent non-profit organization.
我成立非营利永久机构OpenClaw基金会。
OpenClaw’s greatest value lies in transforming public perception of AI.
OpenClaw最大的意义，是改变大众对AI的认知。
AI is no longer a distant, frightening abstract concept.
AI不再是遥远、令人恐惧的抽象技术，
It becomes fun, practical, even a little quirky.
它变得有趣、实用，带着一点古怪的烟火气。
Lobster mascots, themed headbands, home-brewed beer businesses.
龙虾形象、主题发带、啤酒创业项目，皆是佐证。
The future needs more ordinary people to interact deeply with AI.
未来需要更多普通人深度接触AI，
to fully grasp how transformative this technology truly is.
真切理解这项技术变革世界的力量。
We held another ClawCon conference in New York.
我们在纽约也举办了ClawCon大会。
The community has expanded globally.
如今社群遍布全球各地。
Thousands of attendees shared stories of what their lobster built that week.
数千参会者分享自家龙虾智能体本周完成的作品。
A vet in Shenzhen automated grocery shopping workflows.
深圳兽医靠它实现采买流程自动化。
A teenager in São Paulo built an online tutoring business on OpenClaw.
圣保罗少年依托它搭建线上辅导工作室。
Gerhard and his automated beer brewing machine.
还有格哈德的全自动酿酒设备。
None of these creators received professional programming training.
这些创作者全都没有专业编程背景。
Yet every single one of them counts as a true builder.
但他们都是实打实的创造者。
This is the core transformation AI agents bring:
这才是智能体带来的核心变革：
The breakthrough isn’t the technology itself.
突破不在于技术本身，
It’s the universal access to creation.
在于创作门槛向所有人敞开。
AI agents rewrite who gets to build digital products.
智能体改变了能开发数字产品的人群边界。
This door will never close again.
这扇创造之门再也不会关上。
When you can turn a text prompt into a functional prototype within one hour,
一条指令一小时生成可用原型，
every creative idea turns possible.
一切创意都拥有落地可能。
The next industry breakthrough can emerge from anyone, any country, any café.
下一项行业突破，可以来自任何人、任何国度、任意一间咖啡馆。
Even a burnt-out former founder staring blankly at the screen,
哪怕是心力交瘁、盯着屏幕怀疑初心的创业者，
can create groundbreaking tools with this technology.
也能靠它打造颠覆性工具。
Your creative spark never fades.
创作的火花从未熄灭，
It’s just waiting for the right tool to ignite.
只是在等待合适的工具点燃。
The lobster is loose, and it’s not going back into the tank.
龙虾已经挣脱牢笼，再也不会退回水箱。
Thank you very much.
非常感谢大家。