
Three Kinds of Information AI Agents Need
Organizing global, local, and ambient context helps AI agents understand your work and makes prompting easier.
An AI agent needs information about your work to give useful answers. A prompt alone may not explain your needs, business goals, or brand. Nielsen Norman Group studied Claude power users who used the agent daily for professional work. Every participant had a context library. This is a collection of files, databases, and live connections available to the agent. The information in these libraries played three roles: global, local, and ambient context.
Global context guides the agent across different tasks and usually changes slowly. It helps the agent understand lasting priorities and ways of working. A file about your writing style can play this role across projects. You do not need to explain the same preferences in every conversation. Global context should be chosen carefully and stored where the agent can find it regularly.
Local context supports the work you are doing right now. It may include research data, a task list, or project decisions. This information helps the agent understand the project's priorities and progress. It can live in a file or a database. The agent may also update some of it, such as decision logs. For example, one participant gave Claude a markdown plan for a dietary app. A short request was enough because the plan supplied the details.
Ambient context comes from ongoing information around your work. Emails, meeting transcripts, and chat messages are examples. These streams are often large, messy, and not organized for one task. They may reach the agent through MCP connections to other tools. Participants valued this information but often had no time to organize it. More information, however, is not always better.
The same information can change roles when its use changes. A Slack message may begin as ambient context. It becomes local context when you select it for a task. A useful report may later become a reference for future projects. The main question is how the information should support the agent. Its role helps you decide where to store it and how to keep it useful.
핵심 표현
have no time to
~할 시간이 없다
I have no time to organize every email.
keep it useful
계속 유용하게 유지하다
Update the plan to keep it useful.
find it regularly
정기적으로 찾아볼 수 있다
Save the file where the agent can find it regularly.
AI를 사용해 생성되었으며, AI는 실수를 할 수 있습니다.