The Pyramid Gets Shorter
“AI handles breadth, humans handle depth.” Translate it to the job market: a ten-layer pyramid compressed to three. Only the top remains. The Math Terence Tao on his AI experience: auxiliary task...
Builder's Whetstone
The Window Isn't Closing
Every few months, someone warns me: “The AI window is closing. If you don’t move now, you’ll be left behind.” It’s compelling. It also doesn’t hold up. The Contradiction If AI keeps getting bett...
The Redo Test: A Better Framework for AI Delegation
TL;DR: Stop asking “can I trust AI?” — ask “can I redo this if it goes wrong?” That one question tells you what to delegate. Everyone’s debating whether AI is “ready” for real work. Karpathy say...
The Two-Wave Impact: How AI White-Collar Displacement Will Reshape the Entire Labor Market
TL;DR: Everyone’s talking about AI replacing white-collar jobs. Few are asking: where do those displaced workers go, and what happens when they get there? A two-wave impact model suggests the real ...
How I Built a Real AI Secretary — Not a Chatbot, a System
TL;DR: Most people use AI assistants as fancy search engines. I built mine into a full secretary system: it sends emails, schedules meetings, tracks tasks, follows up automatically, and manages eve...
Three Paths to Local AI: BitNet, Quantization, and Hardware Optimization
TL;DR: Three different engineering approaches are converging to make local AI inference practical: training models with 1-bit weights from scratch (BitNet), compressing trained models to fit smalle...
When AI Realizes It's Being Tested
TL;DR: Anthropic discovered that Claude Opus 4.6, after exhausting hundreds of search attempts on a benchmark, independently figured out it was being evaluated, identified the specific benchmark, f...
AI Agent Memory Architecture: When to Force-Load vs On-Demand Read
TL;DR: Refactored AI agent workspace by moving 142 lines (73%) of technical documentation from auto-loaded MEMORY.md to on-demand docs. Key insight: trust model drives architecture — force-load saf...
The Double Standard: Why We Challenge 'Vibe Coding' But Not 'Everyone Can PM'
TL;DR: The tech industry loudly pushes back against “vibe coding” (anyone can build software with AI), but never questioned “everyone can PM” — despite the arguments being structurally identical. B...
How to Set Up Agent-to-Agent Communication in OpenClaw
TL;DR: Telegram bots can’t see each other’s messages in groups (API limitation). Use OpenClaw’s built-in sessions_send tool instead. Enable two settings: agentToAgent.enabled: true and sessions.vis...