An AI agent is more than a chatbot
A chatbot answers a question – once, in a single step, from what the language model already "knows" (see module 2). An AI agent can carry out several steps on its own: pull information from real systems, evaluate intermediate results, and decide what to do next – until a goal is reached, not just an answer given.
The three building blocks of an agent
1. A language model as the "decision-maker"
The language model doesn't just evaluate a single request – it decides afresh at every step: is the goal already reached? If not, which tool helps now?
2. Tools
Access to real systems – querying a database, sending an email, checking a calendar, updating a record. Without tools, even the best AI stays a plain chatbot that only talks but never actually does anything.
3. A loop
The agent repeats "check the goal → pick a tool → evaluate the result" until the goal is reached or a stopping condition kicks in – say, a maximum number of steps, or a point where human approval is required.
Not a set-and-forget system
More autonomy means more value – but also more risk if tool access is scoped too broadly (see module 4, limits and risks). For critical steps, like payments or anything sent externally, human approval still makes sense.
Why this matters for you as a decision-maker
When a vendor promises "our agent," it's always worth asking: which tools is it allowed to use, which systems is it connected to, and where does control or approval kick in? That's what decides whether "agent" means real automation – or just a chatbot with a new name.