AI Field Notes

GPT 5.4 dominates every benchmark. But when I gave both models the same complex product strategy prompt, the gap between benchmark scores and real-world output was staggering. Here's what actually happened.

I switched to API billing on Claude Code and finally saw the real cost of my sessions. Thousands of dollars on my laptop in 14 days. Thousands more on my cloud devbox. The Max plan is a flat monthly fee. The math is absurd.

AI doesn't lack execution power. It lacks people who know what to execute. The future belongs to systematic thinkers who can decompose problems from first principles and orchestrate armies of AI agents. This is the era of the super individual.

The same GitHub project is a quiet dev tool in the US and a nationwide mania in China. This isn't a knowledge gap — it's the product of WeChat's trust networks, Douyin's algorithm, A-share retail investors, an anxiety monetization industry, and government subsidies. None of these amplifiers exist in the US.

A year ago, two essays predicted AI would create permanent wealth lock-in and make humans economically irrelevant. Twelve months later, the scorecard is in: entry-level jobs are collapsing, capital concentration exceeded predictions, but open source and solo founders are the counter-force nobody saw coming.

Two reports dropped the same week — one says 6.8M digital workers today scaling to 720 billion by 2035, the other says 'no systematic unemployment yet.' Both are right. People are panicking about the wrong timeline. Short-term fear is overblown, long-term impact is wildly underestimated, and the right move is to become the person who orchestrates AI teams.

A former Google TPU engineer who worked on V7 and V8 revealed how TPU actually competes with Nvidia. The answer isn't about chip specs — it's about system-level design, software co-optimization, and a fundamentally different philosophy. Apple, Anthropic, and Meta are all using TPU now. Here's what that means.

Three demand curves, one deflation trend, and why compute is the new oil

How I built a hybrid read+write WeChat automation pipeline on macOS — the tools, the architecture, and the 6 pitfalls that'll save you hours

First principles on AI: every trend, traced to its root, is about whether there's enough compute — and whether it's good enough

I wanted a voice-to-notes pipeline. The internet said build a watchOS app. A friend who tried said don't. Thirty minutes later — with Claude Code doing the heavy lifting — I had a zero-cost, zero-dependency solution using nothing but Voice Memos, iCloud, and one Gemini API call.

A 15KB JavaScript library just solved one of the oldest architectural debts in web development — knowing how tall text will be without asking the browser to reflow the entire page.