A More
Beautiful
Context
The Art of Human-Centered Context Engineering
The prompt was never the point. What matters is what the model can see when it answers — your actual documents, the memory of what you said, the five facts it needs instead of the five hundred it doesn't.
What it's about
Briefing a colleague.
Picture the best handoff you've ever received at work. Instead of dumping two hundred files on your desk, someone sat down for twenty minutes and told you what actually mattered. You walked away able to do the job — not because you were brilliant, but because someone had done the work of deciding what you needed to know.
A model, like a new colleague, is only as good as the brief it gets. The first book treated that brief as a single message — the prompt. This one is about everything the message sits inside: the documents you make available, the history of what's been said, the facts you pull in and the ones you hold back. That whole surrounding situation is the context, and engineering it well is the difference between a colleague set up to succeed and one handed a folder and a shrug.
The framework
The FRAME framework
The architecture of AI context — three pillars to curate exactly what a capable mind sees before it answers.
Ground the AI in truth
Provide the current, traceable material — the "dossier" — the model needs to answer accurately without inventing facts.
Ruthlessly prune the noise
Avoid the "junk drawer" effect; more material dilutes focus, so include only what's essential to the task.
Maintain meaningful continuity
Carry essential decisions and project state across sessions — building a relationship instead of a "cold start."
“Every context is an act of editing, and editing is never neutral. What I put in front of a model shapes what it can conclude.” From the manifesto