The word agent gets attached to everything from a scripted chatbot to a system that books travel unsupervised, so precision matters before design begins. This lesson draws the line that actually matters: who decides what happens next.
A Chatbot Responds an Agent Pursues
A conventional assistant produces one response per request; the human decides every next step. An agent is given a goal rather than an instruction, and it decides for itself which steps to take, in what order, using which tools, and when the goal is satisfied. That shift, from human-driven turns to model-driven steps, is the defining change. Everything else, memory, tool access, orchestration, exists to support the model making those decisions repeatedly without a person in between.
Autonomy Is a Dial Not a Switch
In practice, systems sit on a spectrum. At one end, a workflow with fixed steps calls a model for one classification. In the middle, a model chooses among a few allowed paths with checkpoints. At the far end, an open-ended agent loops freely until it declares completion. Each notch of autonomy trades human control for coverage of unanticipated situations, and each notch multiplies the ways things can go wrong unattended.
Why the Distinction Drives Design
The moment the model chooses next steps, you inherit new obligations: constraining what it may touch, observing what it did, and catching it when its judgment fails. Teams that treat an agent like a chatbot with extra features skip those obligations and discover them in production.
Action Step
Take one automation idea you care about and write two versions of it: as a fixed workflow with a model doing one bounded task, and as a goal-driven agent. List what could go wrong in each version. Keep both sketches; later modules will help you choose between them.
Educational content only. AI tools change quickly and can produce confident errors — verify important outputs independently. No productivity or income outcome is promised, and this is not business or legal advice.