Three words used as synonyms, when they actually mean three different things. Confuse them and you buy the wrong tool.
It replies. It decides no path at all.
It runs a fixed, pre-programmed path.
It decides its path at run time.
Excellent for informing and guiding. Its limit is clear: it never touches your systems. If all it does is talk, it isn't an agent.
On a stable path it never gets it wrong and costs little. Its weakness: the moment a case falls outside the script, it stalls. Unbeatable as long as nothing changes.
It earns its place when the path isn't known in advance. That's its strength on variable tasks, and exactly why it needs safeguards: it takes initiative.
Stable task? Automation. Conversational? Chatbot. Judgement at every step, with action on your tools? Agent, with human approval.
saint-jean.co: scoping the right tool for your actual need.