[Week 39 of 2025] Access and Agents

Welcome back to Price and Prejudice with a few musings from Week 39 of 2025.

Private Access, Public Problems

The SEC’s Investor Advisory Committee (IAC) just floated recommendations on letting retail investors into private markets. Their big idea is: don’t throw Grandma into a venture fund; instead, let her buy a registered fund that holds a slice of private assets, with all the usual protections—audited financials, diversification, liquidity rules, etc. The report is careful about investor protections but surprisingly brief on retirement accounts, which has been the subject of much political discussion as well.

Another wrinkle is the role of financial advisors and sales incentives. The IAC nods toward FINRA and state regulators, urging them to police whether brokers and advisers actually recommend these products in a client’s best interest. But history suggests that if you dangle higher fees in front of intermediaries, they will happily explain to Grandma why an illiquid interval fund is perfect for her rainy-day savings. Fees here do double duty: they create conflicts of interest on the distribution side and quietly chip away at household returns once the product is in the portfolio. (A paper of mine highlights precisely this issue). The point isn’t just disclosure—it’s that the distribution machine has its own equilibrium, and it tends to end with retail investors holding the most lucrative products for everyone but themselves.

Then there’s the general equilibrium angle. One: returns in private markets are endogenous—high when they’re scarce, compressed once everyone piles in. The Yale endowment can sell its stake into your interval fund precisely because too much retail demand props up valuations. Two: if access rules change the calculus for companies, why go public at all? Liquidity via semi-public, retail-facing private funds could make the “private” markets the new de facto public markets. In which case, we’re not "democratizing" private markets but just shifting the IPO window sideways.

Agents in the Wild

There is no shortage of speculation about which jobs AI will replace. Analysts predict research associates, compliance officers, and even junior bankers will see their workflows automated away. But predictions don’t get you very far; what matters is what actually happens when the tools are dropped into a live organization. That’s why Citi’s pilot is interesting: it’s a controlled field experiment, not another white paper. Over the next month or so, 5,000 employees will test how “agentic” AI performs in practice, rather than in theory.

The basic pitch is straightforward. Instead of prompting a model step by step—first gather the data, then summarize it, then translate it—an AI agent chains those tasks together. Citi’s internal platform stitches in external models like Gemini and Claude, then lets employees delegate multi-part assignments in one go. Executed properly, the scale would be quite impressive, with thousands of small tasks compressed into a single instruction, repeated across the bank.

The deeper economic question is what happens when agents move from demos to daily life. Citi’s trial isn’t so different from Waymo testing robotaxis in Manhattan: both are dropped into messy systems to prove they can survive. With cars, the pitch is fewer accidents; in finance, it’s fewer bad incentives. Maybe machines don’t cut corners on compliance or oversell a pitch deck. But accidents still happen, and moral hazard has a way of reinventing itself. It would be interesting to see whether we’re getting safer “drivers” of financial workflows, or just new mistakes—at scale.