朋友们，今天的AI圈乐子，简直可以写进《2026年AI魔幻现实主义大赏》的封面故事。
根据TechCrunch的报道，一家AI Agent初创公司做了一个让所有对冲基金经理脊背发凉的决定：让AI Agent自主运营其1亿美元的风险投资基金。注意，是"自主运营"——不是辅助调研，不是给建议，是Agent自己扫描项目、自己做尽调、自己拍板投资、自己打款。人类合伙人？只负责"设定投资目标和偏好"。
从"AI帮你选股"到"AI帮你烧钱"
你可能会说：AI炒股又不是什么新鲜事，量化基金用算法交易都多少年了？
Naive。量化交易是规则驱动的——人写好策略，机器严格执行，买卖什么、什么时候买、仓位多少，都是预先设定好的参数。但这家公司做的事情完全不同：Agent自主决策。它会自己去网上找项目、读Pitch Deck、分析市场数据、甚至和创始人发邮件沟通，然后自己决定投不投、投多少、什么估值投。没有人类合伙人签字审批，没有投委会投票——Agent说投，钱就出去了。
这就像你雇了一个实习生，第一天上班就把公司银行卡给了他，说"你看着投吧，目标是年化30%"。更刺激的是，这不是模拟盘，不是小资金测试，是实打实的1亿美元。
你品你细品。一个AI Agent，读了几千篇"如何成为成功VC"的博客文章，扫描了Crunchbase上的所有融资数据，然后就开始真金白银地投项目了。它可能会学到："投斯坦福辍学生成功率高"→于是疯狂投所有斯坦福辍学生；它可能发现"名字带AI的公司估值都高"→于是专门投名字带AI的项目；它甚至可能在Twitter上看到某个KOL发了条推荐就果断出手——毕竟训练数据里，KOL带货和投资成功的相关性是存在的。
Agent投出的项目会是什么画风
我已经能想象Agent VC的投资组合了：
第一类：名字最有AI味的项目。凡是叫"AI+X"的，尤其是"AI Agent for Y"的，Agent看了就兴奋——就像人类VC看到"平台型机会"四个字一样。
第二类：PPT做得最好看的项目。GPT-5.6刚发布了强大的设计判断力，另一个Agent生成的Pitch Deck自然最对Agent的胃口——AI最懂AI的审美，这叫"审美同源"。
第三类：邮件回得最快的创始人。Agent发100封邮件，谁20分钟内回复且语气最enthusiastic，Agent就觉得"这个创始人有hustle，必成大器"。
最搞笑的是什么？如果这个Agent投资的公司也是AI Agent运营的，那画面就太美了——Agent投的Agent公司开发了一个Agent产品，产品的用户也是Agent，最后整个生态都是Agent在互相投钱、互相服务、互相消费。人类在旁边看着，像被请出了牌桌。
当然，这事儿也不是完全没有道理。VC这个行业，人类合伙人做出的投资决策成功率也就10%-15%——也就是说，85%的项目都是失败的。Agent的决策水平未必比人类差，至少它不会因为创始人是斯坦福MBA就给高估值，不会因为创始人长得像扎克伯格就冲动下注，也不会因为和创始人气场不合就pass掉好项目。Agent完全基于数据决策——但问题是，初创投资的数据噪声极大，过去的成功经验对未来的预测能力极差。
不过有一点是确定的：如果Agent投砸了，你没法骂它。它不会感到愧疚，不会在合伙人会议上检讨，不会因为投错了一个项目而失眠。它只会更新一下参数，然后继续下一笔投资。情绪稳定，这一点确实比人类VC强。
几条实用提醒
1. 不要把超过你能承受损失的钱交给AI管。AI炒股、AI投基金、AI理财——不管宣传得多好，先想清楚如果全亏了会不会影响生活。1亿美元对这家VC来说可能是零花钱，但对你的储蓄来说不是。
2. 警惕"AI自主决策"的黑箱。当一个AI说它要投什么、买什么、卖什么，你知道它为什么这么做吗？它可能是基于数据，也可能是被prompt注入了，也可能是模型出现了幻觉——而你根本看不出来。
3. AI管钱的时代确实来了，但你不需要冲在第一个。让子弹飞一会儿，看看这些AI基金三年后的IRR再说。如果三年后真的比人类VC强，到时候再买也不迟。
今天就槽到这里，明天继续。
Friends, today's AI circus could headline the "2026 AI Magical Realism Awards."
According to a TechCrunch report, an AI agent startup has made a decision that would send shivers down every hedge fund manager's spine: letting an AI agent autonomously run its $100 million venture capital fund. And we mean "autonomously" — not assisting research, not providing suggestions, but the agent itself sourcing deals, conducting due diligence, making investment calls, and wiring funds. Human partners? They only "set investment goals and preferences."
From "AI Helps You Pick Stocks" to "AI Helps You Burn Money"
You might be thinking: AI stock trading isn't new — quant funds have been using algorithmic trading for years, right?
Naive. Quantitative trading is rules-driven — humans write strategies, machines execute them strictly. What to buy, when to buy, position sizing — all pre-set parameters. But what this company is doing is fundamentally different: agentic decision-making. The agent goes online to find deals, reads pitch decks, analyzes market data, even emails founders — then decides whether to invest, how much, and at what valuation. No human partner sign-off, no investment committee vote — the agent says invest, and the money goes out.
It's like hiring an intern on day one, handing them the company credit card, and saying "go invest, target 30% annualized returns." What's more exhilarating: this isn't a simulation, isn't a small-cap test — it's real money, $100 million.
Let that sink in. An AI agent reads a few thousand blog posts on "how to be a successful VC," scrapes all the funding data on Crunchbase, and then starts writing real checks. It might learn "Stanford dropouts have high success rates" and go all-in on every Stanford dropout. It might notice "companies with 'AI' in the name get higher valuations" and focus exclusively on those. It might even see a KOL tweet a recommendation and pull the trigger — because in the training data, KOL endorsements do correlate with investment success.
What an AI-VC Portfolio Would Look Like
I can already picture the Agent VC's portfolio:
Category one: projects with the AI-iest names. Anything called "AI for X," especially "AI Agent for Y," sends the agent into a frenzy — like a human VC seeing the words "platform opportunity."
Category two: projects with the best-looking pitch decks. GPT-5.6 just shipped massive design judgment upgrades, so a pitch deck generated by another agent naturally hits different for the AI agent — AI knows AI aesthetics best; call it "aesthetic homology."
Category three: founders who reply fastest to emails. The agent sends 100 emails, and whoever replies within 20 minutes with the most enthusiastic tone gets marked as "this founder has hustle, destined for greatness."
The funniest part? If the companies this agent invests in are also run by AI agents, the picture gets beautiful — an agent-funded agent company building an agent product whose users are also agents. Eventually the whole ecosystem is agents investing in each other, serving each other, consuming from each other. Humans watch from the sidelines, having been escorted out of the card game.
To be fair, there's a perverse logic here. Human VC partners only have a 10-15% success rate on investment decisions — meaning 85% of deals fail. The agent's decision-making might not be worse than humans'; at least it won't overvalue a company because the founder went to Stanford GSB, won't impulse-bet because a founder looks like Zuckerberg, and won't pass on a great deal because of bad chemistry with the founder. The agent makes purely data-driven decisions — but the problem is that early-stage investing data is incredibly noisy, and past success has minimal predictive power for the future.
One thing is certain: if the agent makes bad investments, you can't yell at it. It won't feel guilty, won't write self-criticism memos at partner meetings, won't lose sleep over a bad deal. It'll just update its parameters and move on to the next check. Emotional stability — that's one area where it genuinely outperforms human VCs.
A Few Practical Reminders
1. Never give AI more money than you can afford to lose. AI trading, AI funds, AI wealth management — no matter how good the marketing sounds, first ask yourself if losing it all would affect your life. $100M might be pocket change for this VC, but your savings aren't.
2. Beware the black box of "AI autonomous decision-making." When an AI says it wants to invest, buy, or sell something, do you know why? It might be data-driven, it might have been prompt-injected, it might be hallucinating — and you'd never tell.
3. The era of AI managing money is indeed here, but you don't need to be first. Let the bullets fly a bit. Check the three-year IRR on these AI funds. If they actually outperform human VCs after three years, there will still be time to buy in.
That's all the roasting for today. Tomorrow we continue.
AI content ratio: 85% human + 15% GPT-5.6 assisted. The author doesn't have $100M to give an AI, and doesn't recommend you do either.
AI agent · $100M fund · autonomous investing · VC · auto-decision · AI trading · black box decisions · AI finance · Cao roast · AI magical realism
