The AI Agent Cycle

This is probably obvious to most at this point, but I wanted to put this down on paper. Any AI agent is an agent because it has agency, meaning that it can take actions on its own given specific inputs.

In order for an agent to be effective, it must follow these four steps in a loop until the original goal is met:

1. Understand the Intent
2. Create a Plan
3. Execute the Plan using tools
4. Analyze the result

If step 4 concludes that the original goal or intent wasn’t met, then start over again, but perhaps with more context from what it’s already learned.

This is quite effective whether used by an AI agent or by a human.