Practicing Wisdom — Issue #22

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

1. What I Learned This Time

Who Keeps the Gains?

There is an assumption buried inside much of the conversation about AI: if the technology becomes more useful, the people building it, buying it, and investing in it should become more prosperous.

How confident should we be that this is true?

• A company can become more productive and lose pricing power. 

• An economy can grow while needing less of an asset that once made its growth possible. 

• A business can deliver on its promise while its shareholders lose money.

We tend to ask whether a technology will work but it is at least as interesting to ask what happens to everyone’s bargaining position when it does.

Jerry Neumann’s AI Will Not Make You Rich starts with the distinction between creating value and capturing it. His historical comparison is container shipping: standardized containers made it cheaper and more predictable to move goods around the world, but competition and heavy capital requirements made those improvements difficult for many shipping investors to monetize. Much of the opportunity appeared elsewhere, in businesses able to reorganize around cheap transportation.

The more provocative part of his argument concerns the timing. Early personal computing gave small companies room to experiment before established businesses understood what was happening. AI arrived with the incumbents already watching, spending, and defending their positions. A technology can be revolutionary without giving its inventors a long period of protection from competition.

I’m less certain than Neumann that this settles AI’s eventual industry structure. One of the most frustrating parts of this AI moment is that everyone seems to be able to force it into a different type of box to make a point. But it does force a useful question: how much of the apparent opportunity depends on competitors failing to notice it.

Imagine a service business whose cost of completing a task falls by half. Initially, that looks like a wonderful margin story. Then a competitor cuts its price. Another offers more service for the same fee. Customers begin to expect what recently seemed extraordinary. The company may end up working faster, serving more people, and earning roughly the same return on each dollar of revenue. That would still represent progress, but the benefits would show up in the customer’s life rather than entirely in the producer’s income statement.

Ben Thompson’s Apps, Agents, and Aggregation adds a complication. Even as capabilities become more widely available, control over the customer relationship may concentrate. His argument begins with a familiar irritation: people accumulate applications because they want things done, and then have to learn each application’s way of doing them. An agent could handle those interfaces on the user’s behalf, or generate a small custom tool when necessary. In that arrangement, the underlying service might remain useful while becoming much less visible. The service still performs the work. Someone (or something) else increasingly decides when it gets the work.

Thompson sees an opportunity for the agent to become both the user’s interface and the intermediary directing demand. Access to personal context, files, and accounts could make that relationship sticky even if the underlying models are replaceable. He extends the argument further: when execution becomes abundant, helping people decide what they want becomes a source of power.

That is where his optimism about the agent business sits in conflict with Neumann’s skepticism about AI investment. Cheap intelligence could create fierce competition among suppliers while strengthening the company that chooses between them. Consumers might receive better service, and a gatekeeper might still collect a substantial share of the gains. The distribution depends on how readily customers can switch agents and how much suppliers need the access those agents provide. A supplier can therefore gain efficiency and lose independence at the same time.

Dror Poleg’s What’s Eating the Office? brings the same problem into the physical world. For a long time, a growing business was a reasonably dependable source of demand for more employees and more desks. Poleg argues that remote work, distributed teams, greater output per employee, and uncertainty about future staffing have weakened that relationship.

The distinction between remote and distributed work matters. A company can bring people into offices every day while spreading a team across several cities. Returning to an office does not necessarily restore demand for the particular office, neighborhood, or headquarters that previously housed the work.

Nor does rising revenue guarantee proportional hiring. Poleg is careful that high revenue per employee does not, by itself, establish how much productivity came from AI. Landlords cannot assume that business growth will translate into space requirements in the old proportions.

There is also a mismatch in commitments: a company whose teams and priorities change frequently may value the ability to add or subtract space, while a landlord financing a building values dependable, lengthy commitments. Someone has to absorb the uncertainty between those two preferences.

The office may remain valuable as a place to collaborate, think, and build relationships. But those purposes do not automatically support the amount, location, and lease structure of the existing stock. A market can even show improving occupancy because obsolete supply has been removed. Recovery in the statistics need not mean restoration of the old business model.

Then Permanent Equity’s Guys, It’s a Grocery Store adds the investor’s final problem: even correctly identifying a successful business is insufficient.

The piece recalls the enthusiasm around Whole Foods as something larger than a grocer - a lifestyle brand with a visionary leader and an expansive future. Yet its underlying operations still involved the constraints of selling groceries. Revenue growth did not spare investors who bought at expectations the business could not satisfy.

A persuasive description of the future can quietly substitute for an explanation of the returns.

Put these pieces together and there are several distinct questions to answer. 

• Will the technology create value? 

• Which businesses will retain some of that value after customers and competitors respond? 

• What has an investor already paid for those gains?

Confidence in the first answer can make us careless about the other two.

There are human reasons for this. Learning new tools feels like preparation and owning something associated with an important trend feels like participation. This offers relief from the discomfort of watching the world change but usefulness, participation, and economic advantage are different things.

After the world has changed, why does this particular business still get paid?

For an application, the answer may depend on whether customers still choose it directly. For a landlord, it may depend on whether the building serves the way tenants now organize work. For a service company, it may depend on whether lower costs allow it to reach customers who could not previously afford its help. And for an investor, the answer must survive the purchase price.

There is plenty of room to be excited about what becomes possible but it seems we need to be more precise about who has a durable claim on the result.

Sources Referenced

• Jerry Neumann, AI Will Not Make You Rich, Colossus

• Ben Thompson, Apps, Agents, and Aggregation, Stratechery

• Dror Poleg, What’s Eating the Office? 

• Permanent Equity, Guys, It’s a Grocery Store

2. Key Distillations

• Progress can lower costs faster than it raises profits.

• A customer you can serve is not necessarily a customer you control.

• Growth can bypass the assets that once made it possible.

• Flexibility is uncertainty someone else agrees to carry.

3. One Contrarian Viewpoint

Widespread AI adoption may tell us very little about the returns to adopting it.

We often treat adoption as evidence of an attractive opportunity, but a business may have to adopt a technology simply to remain competitive. If every rival gets the same cost reduction, customers may eventually capture much of the benefit through lower prices or better service.

AI could become indispensable to running a business without making the average business more profitable.

Perhaps a company reaches an underserved market, develops a learning advantage from its own operations, or earns enough trust to retain customers when alternatives multiply. “We use AI” tells us very little about any of those things. Adoption demonstrates that the tool matters. Retained earnings will tell us more about for whom it matters.

4. One Investable Idea

The low-cost producer in a world of tireless shoppers.

If agents begin making more purchasing decisions, I would look for businesses whose advantage becomes more valuable when customers compare alternatives carefully. A supplier with structurally lower costs and dependable delivery could benefit as buyers find it more easily. A supplier earning a premium because nobody bothers to shop around could face the opposite.

Look for the business that benefits when its customers become harder to fool. Then apply the grocery-store test: pay for the operating economics that exist, with any benefit from agent adoption as upside.

5. From the Archives: A Recall Highlight

“AI may not eliminate software demand. It may eliminate the right to take renewal for granted.”

From Practicing Wisdom, Issue 18.

Enduring demand for a useful activity does not guarantee that its familiar providers will retain their share of the economics.