Do You Know a Guy

Author: Yoni Rechtman URL: https://99d.substack.com/p/paying-the-agency-tax One-line: AI agents will eliminate the consumer surpluses hidden in financial products by making everyone an optimizer, but the “knowing a guy” layer of human relationships will become more valuable, not less, as a result.

Key claims

• Every consumer financial product currently contains hidden subsidies funded by inattentive users. Credit card rewards are paid by merchant interchange fees; mortgage servicers profit from sticky borrowers who don’t refi; banks earn spread on deposit inertia. These subsidies flow from the many (inattentive) to the few (attentive/savvy) — and AI agents will abolish this cross-subsidy by making everyone attentive.
• Morgan Stanley estimates AI agents could double mortgage refinancing frequency. The mechanism: friction removal. Refi today requires intentional effort across multiple steps. An agent can pre-stage everything and execute on instruction — so the optimal refi moment becomes the actual refi moment.
• Apollo has speculated that AI agents could trigger bank runs by enabling simultaneous optimization of deposit placement. If millions of consumers’ agents simultaneously identify that higher yields are available elsewhere and move funds, the coordination effect could amplify into a bank run scenario.
• The “agency tax” is the cost of being represented by someone else. Real estate agents, mortgage brokers, insurance agents, financial advisors — all extract a fee for coordination and information advantages. AI agents compress this cost, but they also create their own agency layer: an AI agent that routes you somewhere is doing so for reasons you may not fully understand.
• “Knowing a guy” becomes more valuable in a world of AI agent optimization, not less. When everyone has access to the same optimization tools, the remaining edge is relationship-based: the deal that isn’t publicly listed, the variance in outcome between a mediocre and an excellent execution, the human judgment call in an ambiguous situation. “Showing up in a room” retains its value precisely because agents can’t replicate it.
• The consumer surplus from agent optimization is real but it comes from somewhere. The interchange income that funds credit card rewards, the deposit spread that funds banking infrastructure, the advisory fees that fund wealth management — all compress. The question is whether the efficiency gain to consumers exceeds the revenue loss to the intermediary.

Notable quotes

“Every consumer financial product is, at some level, a subsidy from inattentive users to attentive ones. AI agents will make everyone attentive. The subsidies disappear.”

“Morgan Stanley thinks agent-driven refi optimization could double the frequency at which Americans refinance their mortgages. Not because rates will be better, but because the friction will be gone.”

“Apollo raised the possibility of AI-agent-triggered bank runs — not because people panic, but because they optimize simultaneously.”

“Knowing a guy still matters. The agent can find the best public offer. It cannot get you the deal that isn’t public yet, or perform better in the adversarial situation where the other side has also automated.”

“You are paying an agency tax on almost every financial transaction you make. AI agents will compress that tax — but they will charge you a different one, in attention and trust, in return.”

“When the optimization layer commoditizes, the differentiation moves to relationship and judgment. The era of AI agents may paradoxically be the era when ‘who you know’ matters more than ever.”

How it connects

• AI Agents — the technology that drives the attention-optimization thesis
• Credit Card Rewards — one of the cross-subsidies that agent optimization will erode
• Morgan Stanley — source of the 2× refi frequency estimate
• Apollo — source of the bank-run speculation from agent-driven optimization
• Matt Levine — wrote the complementary “AI Agents Will Pay Attention” piece for Bloomberg Money Stuff from the industry side
• Financial Services — the industry most affected by attention optimization
• Social Capital — the “knowing a guy” thesis about what human relationship networks are worth in an agent-optimized world