Ask who chooses the next step. With an assistant, you do. With workflow automation, a fixed route does. With an agent, the model chooses its next step within the tools it’s allowed to use.
Aren’t these just labels?
Often, and they overlap. The label doesn’t tell you how the work runs, so ask how it works instead.
The same request, three ways.
A customer asks, “Where is my order?”
An assistant. You give it the order details and ask for a draft reply. You review it and decide what happens next. You direct the work.
Workflow automation. Configured steps look up the order, check the tracking and prepare a reply. A person reviews it before it goes out. AI can draft a step without controlling the route.
An agent. Within the tools it’s allowed to use, the model decides whether to check the order, look at the tracking or ask for missing information, and uses each result to choose what to do next. A person approves the reply.
They combine.
An assistant can be the way people talk to it. A workflow can control the steps that must happen. An agent can handle the part that needs judgment. Running on a schedule doesn’t, by itself, make something an agent.
Four questions before you buy or build.
- Start
- What starts the task?
- Next step
- Who chooses the next step?
- Reach
- What can it read or change?
- Stop
- When must it stop or ask a person?
Practical check: Choose the simplest setup that does the job.
Read more: What AI Agents Actually Do · AI Assistant, Agent or Automation? Three Kinds of AI Help in HR
Examples are illustrative. This is general guidance, not advice for your situation.