GuidesAI agents
AI agent or automated workflow: the difference
Two logics
Workflow
For cases that look alike.
Agent
For cases that vary a lot.
How to choose
| Situation | Rather |
|---|---|
| An invoice arrives, three fields are extracted and filed | Workflow |
| A customer writes; you must understand, search, then act | Agent |
| A mistake is costly and must be avoided at all costs | Workflow, with human approval |
| Cases are too varied to all be described | Agent, with limited tools |
The right mix
In most projects, a workflow carries the skeleton (trigger, checks, storage) and hands an agent the one step that needs judgement.
For the tools: n8n, Make or custom code. For the mechanism: tool use.
Frequently asked questions
- Is a workflow with a ChatGPT call an agent?
- No. The model does one step (summarise, classify), but the order of steps is still written in advance. It is an agent when the model decides what comes next.
- Which one costs less?
- The workflow: a number of calls known in advance. An agent may make more calls depending on the case, hence the need for a cap.
- Can you start with a workflow and move to an agent?
- Yes, it is even the safest path. The workflow shows where cases vary too much; that is where an agent adds something.