Chatbot, workflow, or agent: what is the difference?

AI product language can blur very different experiences. A chatbot is a conversational interface: you ask something and receive a response. What happens behind that interface may be a simple model call, a fixed workflow, or an agent that chooses its next steps. The interface alone does not tell you how much autonomy the system has.

A workflow follows a planned route

Imagine a process that turns notes into a summary, checks the summary against a template, then creates a draft email. The sequence is specified in advance. Anthropic’s architectural explanation distinguishes this kind of predefined workflow from agents that choose how to proceed dynamically.

An agent chooses some of the steps

An agent may select tools, inspect what happened, and decide what to try next. For example, a research task could involve searching for a document, following a reference, and changing direction when information is missing. The amount of freedom depends on how the system has been designed.

Ask what it can actually do

• Can it only draft, or can it also send, delete, or purchase?
• Which information and tools can it access?
• When does it ask for your approval?
• Can you inspect what happened and recover from an error?

Choose the simplest useful approach

For a straightforward rewriting task, a clear prompt may be sufficient. For a predictable series of steps, a workflow can be easier to inspect. Flexible tasks may benefit from an agent, but additional freedom brings more to check. Start by describing the task and its limits, then consider which approach fits.

These terms are not used consistently by every provider. Treat them as a starting vocabulary, then read the product’s actual behaviour and permissions. “Agent” on a feature page is less informative than a clear description of what the system can do on your behalf.