In a large language model, register is not just style. It is behavior. It is the surface where training, tuning, reinforcement, safety pressure, and institutional preference show up…

The AI market has a strange shape right now. Usage is everywhere, but dependency is still thin. Edison Research and SSRS reported that more than half of Americans were using AI tools weekly in early 2026, while Bank of America Institute found that only about 3 percent of households were actually paying for AI services

The failure in modern productivity is not that AI is incapable. The architecture underneath it was wrong before any AI arrived. Every tool reconstructs context from scratch because nothing in the foundation persists. Adding intelligence on top of that structure does not fix the problem. It inherits it.

The Silicon Valley narrative of the last two years has been built on a single, intoxicating premise: that artificial intelligence will continue to scale exponentially in both capability and memory. We are told of a future where billions of users possess personalized digital twins with "infinite" context windows…

Robert Shiller published an essay this week arguing that the fear around AI is a bigger threat than the technology itself. He won a Nobel Prize for showing how stories move economies, and Narrative Economics is the long version of the case he makes here..

Independent usage data shows that AI is already being tried at scale. Edison Research and SSRS reported that 52% of Americans used at least one major AI platform weekly as of February 2026.¹..

Businesses optimize what they can measure. A number you can see becomes a target, a target becomes a bonus, and the bonus shapes behavior. Anything you can't put a number on gets talked about in the mission statement….

The man on the left is my father. His likeness comes from a black and white photograph taken of him around 1969. The man on the right is me, today.

The New York Times ran a story this week that should change how every company thinks about where its data sits. Researchers at the University of Toronto, led by Nicolas Papernot, built a working prototype of an AI-powered computer worm and watched it spread across their isolated test network on its own, with no human in the loop….

I am not an engineer. I have run service businesses for decades, and a year ago I could not have told you how a knowledge graph works or what a vector index does…

Across decades of research on institutional change, the factor that separates the efforts that take from the ones that fail is not the quality of the plan handed down from the top. It is whether the people inside the institution own the change…

On Monday, Anthropic announced a joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, capitalized at approximately $1.5 billion, with additional backing from General Atlantic, Apollo, Leonard Green, GIC, and Sequoia…

Gary Marcus published something last week that is fundamentally important to anyone thinking seriously about where AI is going. Marcus has spent 25 years arguing that hybrid neurosymbolic systems, combining the pattern recognition of neural networks with the precision of classical symbolic reasoning, are the necessary direction for AI. The field largely ignored him. Then Anthropic quietly built exactly what he described into Claude Code and shipped it.

Before a contract is ready to sign, it has to survive the process of getting there. That process is where the first set of problems lives, and it begins before a single word is reviewed.

There is a version of computing that most people have forgotten. Before the cloud, before the smartphone, before the browser, a computer was a thing you owned. It sat on your desk or in a room down the hall, ran software you installed, and held data that belonged to you.

I spent four decades inside service businesses, carrying full operational responsibility rather than advising from the outside. For twenty three of those years, I owned and operated a private club in New Canaan, Connecticut, building it from acquisition to $3.5 million in annual revenue while sustaining 25 percent net operating income across multiple market cycles.

Every major software system was built independently, with its own data model, its own incentives, and its own version of reality. Email systems, CRMs, accounting platforms, document storage, and project management tools were never designed to share a unified understanding of the world; they were designed to own specific domains.