Context-maxxing
- Maxxing
- Research term
Context-maxxing is an informal name for building and maintaining sources, instructions, examples, tool descriptions, and test cases that an AI workflow can reuse. The aim is to improve the material the workflow works from over time.
Work-specific facts and corrections help only when the workflow makes them available. Keeping them in maintained files lets later requests reuse them, provided the app loads the right version. Updating a file does not update requests that never read it.
Where you’ll meet it
For weeks you've used one chat to draft replies to customers. You've corrected the tone, explained your refund rules, and pasted in replies you liked, and the drafts are finally right. Then you open a new chat for a different customer, and the first draft is stiff and generic again.
What to do
Ask the assistant in the old chat to draft reusable instructions from your rules and preferences. Check them against what you said. Have it save approved guidance and chosen replies, then connect later tasks to that document. Give it responsibility for maintaining this guidance in the project instructions.
Easy to mix up
Context engineering assembles the information for one run of a task. Context-maxxing improves the material every run draws on, such as sources, examples, and test cases, over time.
Check yourself
You're building a reference file so your assistant can answer coworkers' questions about their current benefits. Which collection is better suited to that task?
Go deeperReusable material still needs checking after a model change
The suffix comes from internet slang, where “-maxxing” means pushing one thing as far as it will go. Jacob Taylor and Kershlin Krishna used the term in a Brookings paper, “Context-maxxing: A path to cognitive agency with generative AI.”
Some instructions compensate for a particular model's habits and may become unnecessary after a change. An agent's record of why each instruction was added helps you decide what to keep. Have it rerun representative test requests after switching models, then judge the results before approving changes to the guidance.
Markdown documents are plain-text files that many tools can read. A tool description explains what a tool does and which inputs it accepts. Publishing tools through the Model Context Protocol (MCP), an open connection standard, can reduce custom integration work; compatibility still depends on the app's supported features and permissions.
Sources