一位 16 岁少年眼中的商业人工智能世界真相
2 分•作者: VanshAgenticAI•3 个月前
首先，让我介绍一下我自己。我 16 岁，是一名疯狂的科学家。12 岁开始接触编程和人工智能，那时 ChatGPT 还没出现。2022 年，在人们还不了解什么是人工智能时，我就已经构建了 AI 语音助手。12 岁时，我每天工作 10 小时，匿名地进行开发。
我有一项能力：让我构建任何东西——无论是否可能——我都会想办法实现。人工智能、数学、科学，什么都行。这就是我的全部。
15 岁时，我开始销售 AI 自动化和语音代理。我主动联系、发送邮件，连续四个月每天工作数小时。但一次销售都没有成功。没关系——销售不是我的强项。
然后我尝试构建真实的产品——我克隆了一个 OpenClaw，使其能在低端笔记本上运行，还构建了 AI 研究人员。我不是销售人员。我是那种，如果你说“从零开始构建 Claude，包括模型训练”——我就会设法完成。但如果说“构建一些很酷的东西”——我就会不知所措。
尽管如此，我还是尝试了创业的道路。以下是我发现的情况：
1) 投资人和风险投资公司：与他们听起来完全相反。他们声称要发掘人才并承担风险。但他们不这样做。A16Z 的录取率只有 1-1.5%。他们投资的对象是前谷歌员工、前 DeepMind 的首席技术官、已经创办过价值百万美元公司的创始人。那 14 岁时不知道你想要什么的孩子呢？那个在 OpenClaw 和 Claude Code 出现之前就构建 AI 代理的 25 岁年轻人呢？你们只是不乐意投资有才华的无名小卒。风险投资公司为了追求“安全”而失去了他们本来的意义。
2) 创始人工作室：一万名才华横溢的人申请。也许有 1000 人真正优秀。100 人被选中。剩下的 9900 人证明了他们有能力以任何方式构建任何东西——但你们想要的是在谷歌工作十年、拥有机器学习学位、银行里有数百万存款的人。你们宁愿要那些有头衔的蠢货，也不要那些有能力的人。
3) 公开构建——彻底死亡：在所有尝试的人中，10% 成功，而这其中 9% 是那些毫无真才实学的 AI 垃圾 YouTuber。Reddit、X、GitHub——只有那些已经拥有影响力的巨头才能被看到。一个今天开始的开发者只能获得 0-1 个观众和零个 fork。社区已经消失了。
4) 工作：小型 AI 初创公司需要多年的经验。中型公司对初级职位要求拥有十个学位。大型 AI 公司要求五年经验加机器学习学位。而要获得这些经验，你需要去小型公司——而它们也要求经验。
这个循环没有入口点。
奖学金项目说“欢迎新人，不限年龄”，然后在细则里写着：“机器学习学位、Rust/Python、18 岁以上。”他们的思维还停留在 2015 年，而他们的工具却来自 2035 年。
你知道为什么 90% 的 AI 初创公司会失败吗？它们在销售 AI，而不是使用它。在招聘应用型 AI 工程师时，它们仍然想要手动编码的程序员——而 Anthropic 花费了数十亿美元来构建在编码基准测试中名列前茅的模型。甚至那些模型公司的所有者也仍然持有 2005 年的心态：看重学位和手动编码。
如何真正招聘人才：发起一项不可能的挑战。自动化评估——答案非对即错，无需 10 天的人工审查。获胜者直接进入最终面试。就是这样。OpenAI 的 Parameter Golf 接近了这个想法。没有其他任何东西能与之匹敌。
我累了。已经很多年了。
公开构建——已死。自由职业——陷阱。投资人——过于谨慎。工作——需要经验和年龄。
没有出路了。
但如果你读到这里——谢谢你，我只能说这么多。
如果我成功了，我会再读一遍这篇文章，如果我失败了，我仍然会再读一遍。
我不知道，今天是 2026 年 7 月 10 日，我是 Vansh Sharma。
现在，我下线了。
谢谢。
我经历了很多挣扎。
我希望你们能成功。
我不知道是否有人在读。
查看原文
Let me tell you who I am first. I'm 16, a mad scientist. Started coding and AI at 12 — before ChatGPT existed. Built AI voice assistants in 2022 when nobody knew what AI was. Worked 10 hours a day at 12, building anonymously.
I HAVE 1 THING: ASK ME TO BUILD ANYTHING — IMPOSSIBLE OR POSSIBLE — I WILL FIGURE IT OUT. AI, MATHS, SCIENCE, WHATEVER. THAT'S IT.
At 15, I started selling AI automation and voice agents. Cold called, cold emailed, worked hundreds of hours for 4 straight months. Never got a single sale. Fine — selling wasn't my strength.
Then I tried building real products — made an OpenClaw clone that ran on low-end laptops, built AI researchers. I'm not a salesman. I'm the guy who if you say "build Claude from scratch including model training" — I'll somehow do it. But "build something cool" — I freeze.
Still, I tried the startup routes. Here's what I found:
1) INVESTORS & VCs: Completely opposite of what they sound. They claim to find talent and take risks. They don't. A16Z has 1-1.5% selection. They invest in ex-Google employees, ex-DeepMind CTOs, founders who already built million-dollar companies. What about the 14-year-old who didn't know what you wanted? What about that 25-year-old building AI agents before OpenClaw and Claude Code existed — you just weren't comfortable investing in talented unknowns. VCs lost their entire meaning trying to be "safe."
2) FOUNDER STUDIOS: 10,000 ultra-talented people apply. Maybe 1,000 are genuinely strong. 100 get selected. The other 9,900 showed they can build anything, anyhow — but you wanted people with 10 years at Google, ML degrees, millions in the bank. You wanted morons with credentials over builders with capability.
3) BUILD IN PUBLIC — COMPLETELY DEAD: Among 100% who try, 10% succeed, and 9% of those are AI slop YouTubers with zero real skill. Reddit, X, GitHub — only giants with existing traction get seen. A developer starting today gets 0-1 viewers and zero forks. Community is gone.
4) JOBS: Small AI startups want years of experience. Medium companies want 10 degrees for junior roles. Giant AI companies want 5 years experience + ML degree. And to GET that experience you go to small companies — who also want experience.
THE LOOP HAS NO ENTRY POINT.
Fellowships say "freshers welcome, any age" then in fine print: "ML degree, Rust/Python, 18+." Their brains are stuck in 2015 while their tools are from 2035.
You know why 90% of AI startups fail? They're selling AI, not using it. They still want manual coders in interviews for Applied AI Engineer roles — while Anthropic spent billions building models that score top on coding benchmarks. And even the owners of those model companies still have 2005 mentality: degrees and manual coding.
HOW TO ACTUALLY HIRE TALENT: Launch an impossible challenge. Automated evaluation — answer is right or wrong, no 10-day human review. Winners go straight to final interview. That's it. OpenAI's Parameter Golf was close. Nothing else comes close.<p>I'm tired. It's been years.
Build in public — closed. Freelancing — trap. Investors — too cautious. Jobs — need experience and age.
NO WAY LEFT.<p>But if somehow you read this far — THANK YOU, THAT'S ALL I CAN SAY
IF I BECOME SUCCESSFUL, I WILL READ THIS POST AGAIN, IF I DIDN'T , STILL I WILL READ THIS POST AGAIN
I DONT KNOW MAN, ITS TODAY IS 10 JULY 2026, I AM VANSH SHARMA
NOW , I AM SIGNING OFF
THANK YOU
I HAVE STRGGLED A LOT
I HOPE YOU GUYS SUCCEED
I DONT KNOW IF ANYONE IS READING OR NOT