In eleven days, Anthropic shipped a frontier model, exposed industrial-scale espionage, crashed the cybersecurity sector — and got banned by the US government for refusing to build autonomous weapons.
Read signal
Evidence first then a clear position and a useful next move
84% of humanity has never touched AI. You are the 0.04%. The S-curve hasn't bent yet — when it does, the speed will exceed everyone's expectations.
Musk's self-driving analogy for coding AI reveals the deeper signal: when capability converges, taste is the only moat.
The 'universal model' narrative is fracturing. Three companies stopped taking the same exam — and the fracture itself is the most important signal.
Start with a result you understand and open the technical detail only when it changes a decision
Let an AI coding agent inspect a project, explain it, and make a small reversible change.
Run a small language model on your own computer and complete one local conversation.
Put a browser interface in front of a model you control and complete one local chat.
Turn one understandable idea into public work that other people can inspect
Start from one understandable open project, then ship a version another ordinary person can see, try, and question.
Inspect the contract