Agents

An Agent is an AI teammate set up for a particular responsibility. You talk to it in ordinary language, review its work, and decide what it may use.

​

What an Agent is

Each Agent has its own:

• name and role;
• instructions;
• AI model;
• work environment;
• tools, files, and connections it can access;
• task history.

Keeping these separate makes it possible to have several focused Agents without giving every one of them the same information or access.

​

Create a useful Agent

Start with a responsibility you can describe in one sentence. Then write a short instruction that covers:

1. what result the Agent should produce;
2. what information it should use;
3. what “good” looks like;
4. what it must ask before doing.

For example: “Prepare a clear weekly project update from the latest notes. Highlight decisions, delays, owners, and questions. Ask before changing any source file.”

​

Give better tasks

For each task, tell the Agent:

• the outcome you want;
• the useful background;
• any deadline or limits;
• the format you prefer;
• where to save the result.

If the first answer is not right, explain what is missing. The same task keeps the conversation and attached context together.

​

Improve over time

Review the Agent’s results before increasing its responsibility. When you see a repeated correction, add it to the Agent’s instructions or turn the method into a Skill. Use a new task for a new goal. Keep long-running project files on a Dedicated cloud computer or in the File Library, depending on who needs them.

​

Add more when needed