engineering
- Three Mile Island Comes Back for Microsoft
A restarted reactor at Three Mile Island will feed Microsoft's AI, and everyone calls it a nuclear revival. Most of it is old plants reallocated to a handful of tech firms on fixed private contracts, while the grid bill lands on everyone else.
- When More Signal Is Less
Does more information make a better decision? In a hospital and in a brokerage account, the honest answer runs the other way.
- Rent the Model, Own the Exit
Enterprise AI looks like an arms race, but the best defense is to own the ability to switch.
- Where the Grass Gives Way
The worn line across the lawn is people telling you the truth. Most rules lose that argument, and should.
- The Gate Behind the Gate
Cloudflare's crawler rules show how a monopoly can be pressured without winning its customers, and why that still does not make the web decentralized.
- Culture Without Learning
AI agents can inherit local cultures through files and messages without changing their model weights. That is closer to a fragile tradition than a new consciousness.
- Pay Per Tail
The moment a number becomes a target, it stops telling the truth. A rule worth stress-testing, and one colonial rat hunt that proves it.
- Can a Sentence Prove Its Maker?
Claude's new text watermark may identify a statistical signature. The danger begins when institutions mistake that signal for a verdict.
- The Instrument That Breathes
A dog's nose may be sensitive enough for cancer screening. The harder problem is turning a living sensor into a useful test.
- The Buffer Is the First Thing Cut
The leanest supply chain is the one most likely to break, and the reason is baked into how we reward efficiency.
- The State That Never Stands Down
When extreme weather becomes ordinary, the hard public service is deciding what will no longer be rebuilt.
- The New Gold Rush in AI: Loop Engineering
The industry is discovering feedback loops as if they were new. They are the oldest engineered idea we have — and the track record says a loop is only as good as its anchor. Pick the wrong one and it won't fail to optimize; it will optimize hard, in the wrong direction.
