#Guide

A working through-line for what changed in how we use ChatGPT, why the term "context engineering" is doing real work, and the practical moves that separate people who get good output from people who don't. Includes a 30-minute coached exercise you can run on yourself.

Anthropic's Amanda Askell says Claude slips into "criticism spirals" when users get sharp, hostile, or unpredictable. The finding is really about prompt tone as a workforce skill, and what it means for AI training programs that skip right past it.

The conversation around Claude skills is valid, but it misses the bigger picture. The real unlock in Cowork is building modular context as you work, something no web-based AI chat can do.