WHAT ARTIFICIAL INTELLIGENCE (AI) IS AND IS NOT

WHERE AGENTS FIT IN THE BIGGER PICTURE An agent is a software system that pursues a goal by taking a series of actions on its own, observing the results, and deciding what to do next. Compare this with a plain chatbot. A basic chatbot answers one message at a time. You ask, it replies, and that is the end of that exchange. An agent, by contrast, is given a goal, such as "find out why our website is slow and fix it." It then works through many steps: it looks at the code, runs a test, reads the error, makes a change, runs the test again, and continues until the goal is met or it decides to stop. The key ingredients of an agent are these. First, a model, which does the reasoning. Second, tools, which are the actions the agent can take, such as searching the web, reading a file, or running a program. Third, a loop, which repeats the cycle of thinking, acting, and observing. Fourth, some form of memory, so that the agent remembers what it has already done. Fifth, limits and safety checks, so the agent does not wander off or cause harm. The fifth ingredient is where harness engineering lives. Notice how agents fit into the AI map. Agents are not a separate kind of intelligence. They are narrow AI arranged in a particular way, with a model at the center and software around it that lets it take many steps. An agent can feel remarkably capable, but it is still the same narrow technology, and its reliability depends heavily on the quality of the surrounding system. That is the central lesson of this entire book.