"Agent" is the word of the moment, and like most words of the moment, it's used to mean ten different things. Strip away the hype and it's actually simple — and knowing the simple version helps you spot when an agent is genuinely useful and when a plain prompt would do.
What an agent actually is
A normal AI interaction is one step: you ask, it answers. An agent is an AI that can take multiple steps toward a goal — it can use tools, check its own results, and decide what to do next, without you prompting each step.
A prompt answers a question. An agent pursues a goal — looping, using tools, and adjusting until it's done or stuck.
That loop — plan, act, observe, repeat — is the whole idea. Everything else is detail.
Where agents genuinely help
Agents earn their keep when a task has several steps and some uncertainty about the path:
- Researching a topic across many sources and assembling a summary.
- Pulling data from different places and reconciling it.
- Multi-step tasks where each step depends on the last.
Where a plain prompt is still better
If the task is one clear step, an agent just adds latency, cost, and more places to go wrong. Don't reach for an agent to write a single email or answer a single question — that's using a pipeline where a sentence would do.
A grounded way to think about it
Picture a capable assistant. For "what's a good subject line?" you want a quick answer, not a research project. For "find our three biggest competitors and summarize their pricing," you want them to go away, do the work in steps, and come back. Agents are for the second kind of request.
Start by getting good at prompts. Reach for agents when the task is genuinely multi-step — not because the word is everywhere.