How to structure an AI agent workspace that survives the chat
A practical framework for turning scattered agent conversations into durable project context, decisions, and handoffs.
Loading
Skip to main content
Notes on working with agents
Context worth carrying forward.
Practical guides for organizing agent context, decisions, project memory, and handoffs that survive beyond one chat.
Read the field note
More from the notebook
Latest field notes
Write handoffs your next AI agent can actually use
A concise handoff format that gives the next agent state, decisions, evidence, and a clear place to resume.
Read articleContext engineering without the context dump
Better agent results come from selecting the right context at the right moment—not pasting an entire project into every prompt.
Read articleGive AI work a system of record outside the chat
Chats are useful for thinking. A shared, living page is better for the decisions and project state that must survive.
Read articleSearch before you create: the rule that keeps agent workspaces usable
Teach agents to find and update the page that already owns the context before adding another document to the workspace.
Read articleMCP connects your agent to durable context
Why a standard tool connection matters when you want agents to search, read, create, and update shared project knowledge.
Read articleAI-generated work needs version history, not blind trust
Immutable revisions make agent updates reviewable, recoverable, and easier to improve without cluttering the workspace.
Read articleWrite a project brief your AI can actually work from
Turn a passive project overview into an operational page with goals, constraints, decisions, and explicit next actions.
Read articleHow to organize parallel agent work without losing the thread
Give multiple agents clear scopes, shared source pages, and a synthesis point so parallel work creates leverage instead of noise.
Read articleWhat to save from an AI session—and what to let disappear
A practical filter for preserving decisions, lessons, and reusable context without turning your workspace into a transcript archive.
Read articleBrowser-readable pages close the human review loop
Agent output becomes easier to trust when people can inspect it as a clean page instead of decoding a tool log or chat transcript.
Read articleShare AI work without making it publicly discoverable
Public-by-link pages make reviews and handoffs fast, but safe sharing still depends on clear boundaries and deliberate content choices.
Read article