Free guide · 7 pages

Organizing context for AI agents

How to structure the files your AI agent reads, so every task gets the right rules and reference material without loading your whole digital life.

PDF · Free to download, use, and share.

What this guide solves

AI agents do better work when they see the right context, not the most context.

Always loadedOne short entrypoint

Your global preferences, the workspace's hard rules, and an index telling the agent what to open next.

Loaded when neededEverything else

Projects, operating standards, research, templates, and history stay out of view until the task calls for them.

The guide adapts Tiago Forte's PARA method to a simple question: when should the agent read this?

Active efforts with a goal and a finish line. Load them when the task belongs to that effort.

Ongoing responsibilities and standards. Load them when a task touches that client, team, or duty.

Reference material, notes, research, and templates. Search them only when the work needs a fact.

Short enough to read now. Concrete enough to restructure a workspace from it.

01
Context, tokens, and context windows
The mental model you need before reorganizing anything.
02
The entrypoint
What belongs in CLAUDE.md or AGENTS.md, what does not, and why small wins.
03
PARA for agents
A decision tree for filing new information and keeping old context out of the way.
04
One task, traced
A real example showing exactly which files load and which files stay out.

Acme Co

AGENTS.md always

Projects current effort

Areas rules touched

Resources look-ups

Archive only if asked

Start with one short entrypoint, a few hard rules, and an index. The rest can grow when the work calls for it.