AI Agent Five-Element Test: Distinguishing Automation from True Agents

Greg Isenberg's AI Agent Five-Element Test provides a framework to evaluate whether a business workflow should become an AI agent. The five elements are: recurring trigger (high frequency), stable input (structured data), clear toolset (defined actions), measurable endpoint (verifiable completion), and the need for judgment (decision-making in the process). The first four determine if a task can be automated; the fifth distinguishes an agent from rule-based automation. An agent is required when the process involves gray-area decisions that a script cannot handle, which is where LLMs add value. The example of processing customer refund requests illustrates all five elements, with edge cases like late but reasonable requests or high-value customer exceptions demanding judgment.