The "Detroit-to-Human" Moment in Private Equity: When AI Takes Over Alpha, What's Left for Human Fund Managers?

Author: Yuanchuan Investment Review

A recent unemployment report from Anthropic sent chills down the spines of finance professionals.

The report shows that the job replacement rate in finance is as high as 94%, ranking second among all professions, but the current actual replacement rate is only 28%, indicating huge potential for future growth. Fortunately, 30% of professions are almost unaffected, and finance professionals can still consider re-employment opportunities such as dishwashers or plumbers.

Staying in the industry for a long time can lead to anxiety – financial professionals live in a world of comparison, with sales targets and performance rankings weighing heavily on them every day, and a sense of unease arises if they don't keep learning.

For example, after the Spring Festival holiday, when the Financial Securities Depository and Clearing Corporation (FSDC) returned to its workstation, it was still chatting with a chatbot, while its colleague at the next table, Xiao Deng, had already raised eight lobsters and was passionately arguing about the rise and fall of crude oil prices.

The financial industry has never rejected efficiency, from manual order placement to algorithmic trading, from offline bank sales to online distribution. But this time, AI isn't replacing inefficient financial tools, but rather the inefficient people behind those tools. After all, the highest cost in the financial industry is people; the profits of asset management companies are based on how to manage more money with fewer people.

As a result, private equity firms began embracing advanced production methods: Diewei Asset Management offered online courses teaching people how to train "digital researchers" who work 24/7; Mingxi Capital used Manus to automatically generate promotional leaflets for its dividend index, with layouts that rivaled the sophistication of the magazine era. Even clients became more cautious; after a financial advisor had just recommended a popular private equity firm, they would immediately turn around and ask Doubao whether they should buy it.

The private equity industry is gradually entering a Detroit-like era, with replacements already occurring at every link in this mature chain: investment research, operations, and sales.

Salary vs. Token Costs

In a competitive environment where operating costs are high and alpha is becoming increasingly difficult to obtain, the per capita efficiency ratio is a metric that private equity managers rack their brains to optimize every night before going to bed.

In the private equity industry chain, researchers' salaries are generally quite high. According to data from Mulifang, the annual salary of quantitative stock researchers is usually between 800,000 and 1.5 million yuan. The salary of subjective researchers is slightly lower, but there are also occasional shocking incentives. For example, at the beginning of the year, a subjective researcher with tens of billions of yuan in assets received a year-end bonus of more than 20 million yuan because he recommended Nvidia.

If private equity firms can successfully leverage AI-powered investment research, they can save tens of millions of dollars in costs. If it can work 24/7, it can reduce hourly wages while generating greater output. The AI ​​won't take a penny of the money that would otherwise be deducted from the boss's earnings, such as travel expenses, overtime, transportation costs, and meal allowances.

In the asset management field, all technological advancements boil down to two words: improving efficiency and reducing costs. Private equity managers don't care whether AI can truly think like a human; they only care whether the work can be completed.

In response, Howard Marks did the math: if the analysis could produce the equivalent of a $200,000 annual salary for a research assistant, then for the person paying the salary, it doesn't matter whether it's genuine thought or just pattern matching; the key is whether the work is reliable enough to have practical value.

After returning from the Spring Festival, the financial research teams of eight securities firms collectively released a "crayfish farming" tutorial, personally accelerating the process of human researchers being replaced. They tested OpenClaw and found that it could proactively produce research results like humans.

On the Jinmen APP, a roadshow by OpenClaw Financial Engineering, titled "OpenClaw: From Beginner to Expert," was played 4,839 times; Xu Jianhua from Northeast China recommended 20 skills that can boost investment research efficiency by 10 times; Cao Chunxiao from Founder Securities used lobsters to reproduce the PB-ROE strategy, cup and handle pattern stock selection strategy, and fully automated factor mining and backtesting.

This is terrifying to think about; it's like simultaneously acquiring the skill sets of Buffett, O'Neil, and Simmons.

A trader who loves learning

The sellers worked hard to popularize science, and the buyers also learned very actively. A private equity firm in Beijing was afraid that its main machine would be contaminated, so it gave each of its investment research staff a new computer and a subsidy of 50,000 yuan in tokens, which were used specifically to raise lobsters[1].

Yang Xinbin of Xueqiu Asset Management has trained two lobster researchers. He said that he talks to AI much more every day than he talks to people. The AI ​​Agent he trained can do more work in two days than a mature quantitative researcher can do in six months, and may even have greater potential.

Paul Wu of Qinyuan Capital is gradually integrating AI into various departments. He feels that AI can complete closed loops in some job roles and operate independently and iterate. He foresees that in the near future, the company's expenses will become purchasing and maintaining an Apple analyst AI agent, and perhaps later a portfolio advisor named Paul.

In the past, many private equity firms have experienced wear and tear on their investment research and development—researchers think fund managers are incompetent, and fund managers think researchers are useless. The emergence of OpenClaw has shown private equity owners a completely new possibility for the first time—they no longer have to endure the internal friction of repeatedly working with mediocre researchers, nor do they have to worry about their core researchers being poached by competitors with high salaries.

In terms of characteristics, Lobster fulfills all the ideal expectations fund managers have for researchers: working around the clock, without holidays or slacking off; having a long-term memory and being able to recite key data fluently; being absolutely loyal and obedient, and not setting up their own independent firm with core strategies; and continuously iterating themselves, unlike the old researcher Deng who became obsessed with his own path dependence and was then eliminated by the times.

If the cost of silicon-based tokens is far lower than that of carbon-based salaries in the future, how can private equity tycoons refuse an obedient, easy-to-use, and trainable AI researcher?

The replacement is not just because of lobster

While private equity firms are still weighing whether token costs are worthwhile, large quantitative trading firms have already compressed token costs to extremely low levels thanks to their self-built computing infrastructure. However, they remain unusually calm in the face of this surge in popularity.

"For the quantitative trading community, OpenClaw is nothing more than a half-finished toy," a leading quantitative trader in Shanghai told me. Its significance lies in lowering the technical threshold for subjective institutions and retail investors, and providing a clear cost recovery path for large model companies' huge upfront infrastructure investments, but it has little meaning for the serious production environment of quantitative investment.

Another leading quantitative analyst put it more bluntly: OpenClaw is operating like a pyramid scheme in the financial world. Its randomness, lack of systematicity, and low security bring enormous uncertainty to the entire quantitative trading system.

OpenClaw is not considered cutting-edge in the quantitative trading community, and Cui Yuchun of XunTu Technology believes there's no need to be anxious about it.

Lobster's capabilities in agent optimization and tool usage (including research browsers, writing, and data analysis tools) are significantly weaker than those of agents like Manus and Kimi. For a researcher without a programming background, it takes 5-10 hours to deploy and get started, and most tasks fail to achieve a score above 60.

When retail investors use China Stock Analysis Skills to select stocks, it's as if a new world has opened up for them. Quantitative trading has built a multi-agent platform, leveraging a richer arsenal of agents to crush retail investors. However, the operation of this powerful system may not necessarily require more humans.

Traditional quantitative investment research systems typically employ a pipeline architecture: data cleaning → factor calculation → model prediction → portfolio optimization. With the advent of the AI ​​era, some institutions, like top overseas quantitative investment groups, have simplified this to a system of role division → tool usage → workflow design. Standardized, repetitive tasks are gradually being replaced by AI agents, eliminating the need for researchers to be distorted in factor sweatshops.

For example, Apollo AI Multi-Agent System, invested by Xiyue, embeds AI Agents into various aspects of investment research, data, trading, and operations. Founder Zhou Xin described it as being like having seven or eight hundred more AI employees.

With quantitative "unmanned factory"-like sci-fi crushing power ahead and retail investors using OpenClaw to reduce information gaps behind, subjective fund managers, who are in the middle of the efficiency gap, are in a rather awkward position. They are looking at the information that researchers have worked hard to produce, but they are being attacked by quantitative quantification from above and are being pressed by retail investors from below. They inevitably fall into the anxiety of AI FOMO.

During the Spring Festival, I reviewed the annual report of a leading Shenzhen-based fund manager, who lamented that fund managers have excessively high expectations of research analysts.

Fund managers expect researchers to be market-sensitive, promptly highlighting opportunities and providing research and judgments that are ahead of their peers; they even need to be constantly on the "inner circle." If researchers can perform this level of work, why would they need fund managers? Why would someone need to work for a fund manager when they can make a fortune trading stocks on their own?

Therefore, he lowered his expectations—the researchers are only responsible for researching specific targets and issues, and they do not need to discover opportunities or give investment advice; these are all part of his job as a fund manager.

Conversely, if subjective fund managers only need someone who doesn't penetrate the core circles of the industry and relies solely on desk analysis to track targets, then wouldn't such researchers be replaced by AI agents in the next step?

end

Being in the A-share market, the past two years have felt like they've been put on an accelerator.

The first half of the year was particularly hectic. Last year, Deepseek launched during the Spring Festival, Trump imposed harsh taxes during the Qingming Festival, and this year, everyone was raising shrimp during the Spring Festival. Before the Lunar New Year was even over, war broke out in the Middle East. The brains of finance professionals have been constantly overloaded; I can't even remember the last holiday when I didn't have to study. At least for me, as an editor, my mental processing power is insufficient.

In my memory, when I talked to fund managers about writing articles two years ago, I would often hear them happily describe their work with an awkward sentence: "I go to work every day by tap dancing." But in the past two years, when we talked, they would talk about the "iteration" of team organization, the "iteration" of investment philosophy, and the "iteration" of industry knowledge without a smile.

With AI developing so rapidly and competitors progressing so quickly, it seems that only through iteration can one avoid being eliminated.

The industry is still too anxious.

AI doesn't understand human nature; it can't predict whether the trading in the A-share market, where retail investors are concentrated, is based on the third or fifth derivative at this very moment. AI lacks empathy; it can't understand why some people have been trapped by the two oil giants for so many years, yet still hold on, just waiting for the day they can break even. AI cannot take responsibility; it won't be blocked at its door by investors for losing 30%, nor does it need to write an apology letter to reflect on its soul or examine itself.

If AI replaces all fund managers and researchers in the future, then the efficient market hypothesis will hold true, there will be no more Alpha, and there will almost certainly be no more Warren Buffett.

So the real question is, in the future asset management industry, when AI takes over data mining, model running, and report writing, what will be left for humans? What will remain is precisely the passion for investing, the intuition for uncertainty, and the reason for choosing to stay even when criticized for not being as good as AI in research.

We cannot change the trend of AI's increasing proportion, but we can change the mindset of being busy responding and struggling to catch up.

Just like in the game "Detroit: Become Human", the final choice that players have to make is not to destroy the AI, nor to submit to it, but to decide what roles humans and AI should play.