A chatbot answers a question. An agent is asked to do a job: gather context, choose a tool, take a step, look at the result, and decide whether to continue. The loop is what people mean by agentic. The tools are ordinary things, such as a database, a browser, a mailbox, or an API your company already trusts.
Autonomy is a setting, not a virtue. The more an agent can do, the more it needs a written job, a boundary, and a person who accepts the output. In commodities, logistics, healthcare, and financial services, that person is usually already accountable for the queue. The agent should make their week clearer, not move the accountability onto a vendor.
If you cannot name the job in one sentence, you are not ready to build the agent. Write the input, the decision, the system of record, and what wrong looks like. Then decide whether a person, a script, or an agent is the cheapest way to do that job well.