The AI Stack
The models, harnesses, platforms, infrastructure, markets, and economics reshaping software delivery.
← All dispatchesDeepSeek Did Not Make Local AI Inevitable
Open weights made cheaper AI possible. They did not make hybrid AI easy. The missing product is the routing and operations layer that lets an ordinary enterprise use more than one model safely.
One User Can Spend Your AI Budget
The moment one employee can consume half a team's AI allocation in a day, AI stops being a productivity subscription and becomes infrastructure that needs governance.
What AI Companies Can Learn From the Auto Industry
Frontier labs do not need one model for every job. They need a product ladder that gives customers Toyota economics, Porsche capability, and Koenigsegg performance when the work earns it.
The Prediction Is Already Becoming Testable
Eighteen days after I predicted Anthropic would be the first major frontier-model provider to fail, the mechanism behind that prediction has started to become testable.
The Mainframe Era of AI Is Ending
AI began in centralized data centers, but hardware, local models, and enterprise sovereignty are starting to move inference closer to the user.
Anthropic Will Be the First Major AI Company to Fail
Anthropic leads the business AI coding market. I predict that its lack of control over compute will also make it the first frontier-model provider to fall.
The Harness Wars Have Started. We've Seen This Before.
Claude has already won. Copilot is dead in the water. Cursor is Elon's fantasy play. Every one of those takes is pontification dressed in the robes of an oracle.
All Your (Code)Bases Belong to Us
AI companies are scraping public repos and shipping community innovations as features. If your code enhances an agentic platform, think twice before going public.