All sharing
The homepage keeps six fixed highlights. This archive keeps every public essay, talk, podcast, and slide deck.
Chat Is a Synchronous Task. A Task Is an Asynchronous Chat.
Chat synchronously clarifies intent; Tasks keep that intent running asynchronously. A Task cannot be just a longer Chat.
Loop Engineering Is Hot. It Will Be Outdated Too.
From prompt, context, and harness to loop: new terms only mark where the bottleneck is. The end state is giving the system a goal.
Don’t Tokenmax. Optimize for Interruption Minimizing.
The maturity of an AI-native system should not be measured by token burn, but by IPB: interruptions per billion tokens.
AI Is Not a Tool, Nor a Coworker. AI Is a Hydropower Station
A new metaphor for human-AI collaboration: AI is a cognitive production system. The core of an AI-native organization is not usage or management, but control system design.
Reconstructing Organizations: From Human-Native to AI-Native
Tencent Academy slides on how AI-native companies should operate across workflow, talent, and organization design.
How Should Humans and Agents Collaborate? Don’t Manage a Bunch of AI. Train a System.
Humans should face one master agent, delegate asynchronously, and turn every artifact into system capability.
Claude Code Is Already Starting to Feel Outdated. The Next AI Paradigm Is Hiding in the Task Board
AI coding is moving from humans operating agents to agents managing agents. The humble task board is becoming the control plane for AI-native organizations.
Organizational Capability Is the Real Moat for AI Companies: Six-Month Review
A six-month review of the AI-native organization thesis: what held up, what needs correction, and where workflow, talent, and organization design go next.
An AI Team Built from the World's Best Minds
One human + an AI-only team. Meet Elon (CEO), Jobs (product), Linus (engineering), Turing (verification), and Bezos (customer success). Custom software at 1/10th the cost, delivered in days.
OpenClaw Triggered an Inflection Point in AI Evolution
AI-only systems are becoming practical: agents managing agents, file systems replacing chat, Agent = Model x Harness, and human-facing software becoming highly customized.