Practicing Wisdom — Issue #21

A distillation of the most interesting things I explored, learned, and thought about.

1. What I Learned This Time

A customer who never checks the price is a wonderful customer to have.

They leave cash earning very little, auto-renew insurance policies and keep paying for a subscription they meant to cancel. None of this requires stupidity or a lack of care, it revolves around distraction. It requires a life: children, work, dinner, a hundred decisions more urgent than finding out whether a previously good choice has slipped into a bad one.

A surprising amount of profitability sits inside that perfectly reasonable allocation of attention.

Marc Rubinstein’s Agents Revolt made me wonder what happens when we leverage AI to never get bored or stop paying attention. He describes using an AI agent to shop for car insurance, gathering alternatives while he slept. The interesting change is that the customer no longer has to devote an evening to a task whose payoff is uncertain. The work of comparison starts to become cheap enough to repeat. That creates a useful question for almost any business: Would our best customers still be our best customers if they understood every alternative?

Some businesses should welcome that world. Their economics depend on serving customers efficiently and sharing enough of the benefit to keep them coming back. Others have quietly counted on the customer failing to do the math. Both can report excellent retention, but that retention means something very different.

We explored part of this in Issue #18, when I wrote about tedium protecting software moats. This week’s readings push the question further. Once an agent makes comparison easy, who decides what gets compared? Who defines the objective, and who pays the agent?

Ben Thompson supplies the complication in Frontier Overhangs. As models become capable enough for many everyday tasks, convenience, integration, and the relationship with the user can matter more than having the most impressive model. A personal agent that knows your preferences, holds your context, and participates in your daily routines may become difficult to replace. Put those two arguments together and another possibility emerges: the agent that helps me escape my bank’s inertia could acquire an inertia advantage of its own. Could AI become the stickiest and most attention grabbing middleware we have ever created?

That does not make the service bad. Context has real value. So do reliability and convenience. It does however change where I should direct my skepticism. If my assistant recommends an insurance policy, I want to know whether it searched the whole market, whether the coverage is actually comparable, and whether someone paid to appear in the answer. A lower premium is an incomplete objective if the deductible doubles. It also continues to highlight how much of our lives could be directed by an extremely small set of companies. So much of the economy now rides on the market cap of a few isolated giants - how much more concentrated will that become if those companies start making all of our choices for us?

Rubinstein’s other piece, The Art of Doing Financial Engineering, brings that same question to the people funding the AI buildout. His examples show how leases, guarantees, and separate financing vehicles can connect enormous projects with investors willing to fund them.

There is real economic value here. A lender may be better equipped to hold a predictable stream of contracted payments than a technology company is to finance an entire project from cash. Good structures can distribute risks to people who understand them and can bear them,  yet the labels on the structure do not tell us everything we need to know.

In the Meta financing Rubinstein describes, the company’s exposure includes a commitment to cover a specified shortfall if it terminates a lease and the affected assets are sold for too little. In his IREN example, a large customer contract helps finance the equipment, while shareholders remain exposed to what the infrastructure is worth after the contract ends. Surprise, surprise - these exposures show up in disclosures and not on the balance sheets themselves. Why would that be?

Where does the cash ultimately come from? What happens if the customer leaves? Who owns the equipment when the contract expires? Who is still obligated to pay? You can spend a great deal of time studying a financing diagram without answering those questions. Several legally distinct claims may still depend on the same customer, the same technology, and the same optimistic assumption about future demand.

This is where attention and incentives come apart. A consumer may accept a poor deal because investigating it takes too much time. A sophisticated investor may understand a structure quite well and still accept it because the yield, mandate, or competitive pressure makes participation attractive. Better analysis addresses the first problem more directly than the second.

Howard Marks approaches a related issue in Shall We Repeal the Laws of Economics - Part III. His criticism of efforts to suppress long-term yields rests on the distinction between changing a market price and addressing the fiscal conditions behind it. Buying bonds can influence their price but the spending commitments and financing needs remain.

There are legitimate reasons to intervene in markets, as Marks acknowledges. Now is probably not that time, and seems more of a play on ‘extend and pretend’ rather than dealing with uncomfortable underlying truths.

I find that idea useful well beyond the government. A company can improve the presentation of its capital intensity while accepting future obligations. A platform can make choosing between alternatives effortless while concealing how choices are ranked. An organization can produce a beautiful dashboard that nobody wants to question. The information may be available. The arrangement may be entirely disclosed. Yet understanding still requires someone to follow the consequences through. I don’t get the sense that any of the technology we are racing at building is going to make humans better at this subtlety over time.

Benn Stancil’s How to Have Ideas approaches the human side of this. He uses Codenames to explore how unusual connections emerge: a sparse collection of words forces you to find relationships you would never search for directly. Carry a half-formed question into a walk, a conversation, or an unrelated book, and something unexpected may help it take shape.

I read this as a theory of creative work, rather than a settled explanation of brains or language models. What resonated was the role of participation: an idea becomes useful through the connections we make with it. Receiving the finished answer can spare us effort while also skipping the experience that would have made it ours. 

As an aside, that is basically the point of this distillation endeavour - finding the hidden connections between seemingly disparate topics to create unique viewpoints and insights. Yes, all of this content I cover exists independently, but has anyone ever combined it before? 

There is plenty of tedious work worth delegating and I am happy to delegate it to agents. The harder responsibility is deciding what deserves scrutiny once the busyword is abstracted away. I want to understand whether a business benefits when its customer becomes better informed, whether a financing still works under disappointing assumptions, and whether an assistant serves the objectives I actually care about.

The why will become more available as the what and how are automated away.

Sources Referenced

Frontier Overhangs - Stratechery (link)

How to Have Ideas - Benn Stancil (link)

The Agents Revolt - Net Interest (link)

The Art of Doing Financial Engineering - Net Interest (link)

Shall We Repeal the Laws of Economics - Part III - Howard Marks Memo (link)

2. Key Distillations

• Loyalty becomes more convincing when leaving becomes easy.

• A guarantee deserves the same curiosity as a loan.

• Cheap comparison makes the role of the comparator critical.

• Disclosure is an invitation to understand, not evidence that anyone has.

• An answer can be delivered. Understanding takes participation.

3. One Contrarian Viewpoint

AI could make the intermediary more powerful even as it makes switching providers easier.

The appealing story is that agents empower consumers by searching, negotiating, and switching on their behalf. I think much of that can happen, but the service performing those actions also gains influence over which providers customers encounter.

My hypothesis is that we may see intense competition among banks, insurers, and other suppliers beneath a relatively small number of trusted assistants. A provider could win a customer through better pricing while losing ownership of the relationship to the agent that recommended it.

The crucial questions would then concern the assistant’s incentives: how it earns revenue, which alternatives it can access, how it ranks them, and how easily a customer can take their history elsewhere. Notice that none of this is controlled by the user. A capable agent with conflicted incentives can make a poor recommendation remarkably convenient.

4. One Investable Idea

Build a household financial assistant whose economics improve when the customer saves money and stays well served.

The opening I see is a service that monitors recurring financial decisions: insurance renewals, idle cash, subscription changes, and opportunities worth investigating. Start with a narrow, measurable task, such as reviewing an insurance renewal while preserving the customer’s required coverage.

The product should show the existing arrangement, the alternatives considered, the material differences, and the expected benefit after fees. It should ask for approval before consequential changes and make its recommendation understandable.

I would evaluate the idea through realized savings after fees, errors and reversals, repeat usage 

The opportunity is to make careful financial housekeeping economical for households that cannot justify hiring someone to do it. The durable advantage would have to come from trustworthy execution and accumulated context; a recommendation screen would be easy to copy.

5. From the Archives: A Recall Highlight

“Markets can be irrational without their participants being irrational. The incentives can be crazy even when the people aren't.”

— Practicing Wisdom, Issue #20

This applies just as well to the agent choosing the product and the banker arranging the financing. More intelligence leaves the question of incentives very much alive.