1.1. From Chatbots to Agents
💡 First Principle: A chatbot answers; an agent acts. The moment software can decide for itself which of several capabilities to use in pursuit of a goal — look something up, call an API, hand off to a specialist — it has crossed the line from scripted conversation to agency.
Why does this distinction matter for the exam? Because AB-620 is not a chatbot exam. Nearly every question assumes an agent that autonomously chooses among topics, tools, knowledge, and other agents. If you reason from "the agent follows the script I wrote," you will pick wrong answers; if you reason from "the orchestrator decides, and I shape its decisions," the correct option usually stands out.
Think of the difference like a phone tree versus a concierge. A phone tree ("press 1 for billing") only ever does what its menu allows, in the order the menu allows it. A concierge hears your goal — "I need to change my flight and find a hotel near the conference" — decomposes it, and decides which desks to visit on your behalf. Copilot Studio lets you build both, but the exam rewards you for knowing how to build, govern, and extend the concierge.
⚠️ Common Misconception: An AI agent is just a chatbot with an LLM bolted on. In reality, an agent perceives context, reasons over instructions, and autonomously selects actions — tools, knowledge lookups, delegation to other agents — to accomplish a goal. The LLM is the reasoning engine, but the agent is the whole system around it: instructions, knowledge, tools, triggers, and orchestration.
