Roundup: Cheap Thought, Costly Consequences

Various thesis-bending topics that we have been thinking about; the most important one first. And thanks for the overwhelming number of notes on last week's appearance on the Ezra Klein show.

What's ahead:

1. Agents Create New Markets
2. Intelligence Is Becoming Too Cheap to Meter
3. Banking Is Hollowing Out Its Training System
4. Productivity Can Make Everyone Poorer
5. China’s AI Build-Out Has a Local Extraction Problem
6. China May Be Japanifying—With Factories
7. Scott Bessent Falls for the Smart People Trap
8. Physical AI Moves Failure Into the Real World
9. Blackstone's Adventures in Post-ZIRP World

1. Agents Create New Markets in Plumbing

After puttering along for years, password manager 1Password has passed $500 million in recurring revenue, with revenue growth accelerating sharply. Why? Agentic traffic is driving demand for automated identity, permissions, payments, monitoring, and revocation. These are all explosively larger markets when software begins acting rather than merely answering. Agents change what software companies can sell, and the plumbing beneath it. We talked about this previously, but this is further evidence.

Source: 1Password Surpasses $500 Million in Revenue as AI Agents Drive Growth

2. Intelligence Is Becoming Too Cheap to Meter

Epoch AI estimates that the cost of reaching a fixed level of AI performance has fallen 47 percent per quarter since 2023, or roughly thirteenfold annually. Standing still, to put it another way, requires token growth to double every 90 days or so. Scarcity has exited cognition toward everything required to apply it: proprietary data, trusted access, physical infrastructure, distribution, and, yes, the scarcest resource of all, managerial competence.

Source: The Plunging Price of Thought

3. Banking Is Hollowing Out Its Training

Junior banking employment is 4.7 percent below its late-2022 level, while senior and technical roles continue to grow. Banks are shrinking the apprenticeship layer from which they traditionally produced experienced senior employees. Granted, larceny doesn't require a long apprenticeship, but AI is bending the curve away from new hires.

Source: Banks Are Losing Interest in Junior Talent in the AI Era

4. Productivity Can Make Everyone Poorer

Machines can increase production while reducing the labor income needed to purchase that production. Under some conditions, supply destroys its own demand, producing “immiserating growth.” Falling AI costs strengthen the returns to ownership while weakening the bargaining position of replaceable labor.

Source: Robots Are Us: Some Economics of Human Replacement

5. China’s AI Build-Out Has a Local Extraction Problem

Shaoguan plans data-center investments equal to roughly 65 percent of its annual economic output, which is bananas. The construction boom raises electricity use and measured GDP, but the better jobs often go to outsiders while villagers get dust, low-paid service work, and uncompensated land loss. National AI capacity can look like economic development from Beijing and resource extraction from the village.

Source: China’s Villages Pay the Price for Beijing’s AI Ambitions