Wall Street veterans Alex Popa and Charles Elkan are raising money for an equity hedge fund run primarily by artificial intelligence agents and models. Poesis, to which the two belong, is trying to move AI from a research assistant to the core of portfolio management. (Previous summary: He once said that AI is garbage! Citadel boss Ken Griffin was shocked by the facts that happened "inside the company") (Background supplement: Citadel founder Ken Griffin shouted that "generative AI is useless": Alpha cannot be discovered, and Wall Street still has to beat the market by hand) Wall Street's AI applications are moving from helping analysts read data to directly participating in fund allocation. A summary of the Bloomberg report on November 10 stated that Alex Popa and Charles Elkan are raising funds for a new stock hedge fund, which will be mainly operated by artificial intelligence (AI) agents and models. Poesis, the investment company to which the two belong, said on its official website that its AI strategies have been put into real markets, with agents generating investment signals, testing strategies, allocating capital and managing risks. This design is not just about organizing summaries for fund managers, but putting research, portfolio and risk management into the same system. Capital Group’s investment experience combined with Goldman Sachs’ machine learning background Poesis founder Alex Popa has worked at Capital Group for 16 years as a partner and managed its Growth Fund of America. Another core member, Charles Elkan, is the chief scientist of Poesis. He was a former managing director and global head of machine learning at Goldman Sachs, and also led machine learning research and development at Amazon. The combination of the two involves active stock selection and asset management on one end, and model research and system construction on the other. The goal of Poesis is to allow AI to continuously learn from market feedback; the actual investment of funds becomes the field for testing model judgments. First use real money to test whether the agent can complete the entire set of tasks. Poesis has previously partnered with ASUS and Huida to conduct experiments. Asustek’s public case introduction shows that two groups of AI investment teams conducted real money transactions for a week, adopting long-only and market-neutral strategies respectively. The experiment split the work among different agents: some look for signals, some manage portfolios, some execute trades, and others have independent checks and risk gates. The entire process operates on a local workstation, and the focus is on testing whether the agent can work together within a given investment framework, rather than just showing a stock-picking answer. In addition to research efficiency, fund management also has to withstand market fluctuations. This type of architecture advances the task of AI from generating recommendations to influencing position and risk allocation. For investment companies, the potential value lies not only in reading the information faster, but also in whether the research results can be connected to transactions and whether risks can be continuously adjusted after changes in positions. However, short-term experiments can test whether the process works, but they cannot cover the complete market cycle. From real capital testing to fund raising, the test questions Poesis faced also changed: not just whether the agent can pick stocks, but whether the entire investment system can continue to operate under market reversals, transaction costs and risk restrictions. Related reports Interactive Brokers has gathered three major AIs: Claude, ChatGPT, and Grok can now be used to help you select stocks, trade options, and risk allocation for free. OpenAlice’s open source AI trading engine has become popular, and “Wall Street for One” has gained 3,100 stars on GitHub. The actual measurement details are not in the program. Wall Street’s first non-human employee! Goldman Sachs hires AI engineer Devin to "work more than four times more efficiently" and can collaborate with 12,000 developers at the same time. "These two Wall Street veterans are raising funds to build a hedge fund, and they want AI to operate it directly." This article was first published in BlockTempo "Dongqu Trend - the most influential blockchain news media".