A chatbot answers. An agent acts.
A chatbot takes a message and returns a message. Everything it produces lands back in front of a person, who decides what happens next. Even a very capable chatbot is a conversation partner.
An agent is given a goal and a set of tools — search, a spreadsheet, your email, a calendar, an API — and decides which steps to take to reach that goal. It can read the result of one step and choose the next one without you typing anything in between.
The four parts every agent has
- A goal: a finished state, not a single question (“every new lead has a reply drafted and logged”).
- Tools: the actions it can take, each with clear limits.
- Memory: what it carries from one step, or one run, to the next.
- A loop: look at the situation, choose an action, check the result, repeat or stop.
When a chatbot is the better choice
If the work ends in words a human will read and act on — drafting, brainstorming, explaining, summarising — a chatbot is simpler, cheaper and easier to trust. You review every output by default.
Reach for an agent only when the value comes from steps happening without you: moving data between tools, following up on a schedule, or handling the same process dozens of times a week.
A quick test
If this worked perfectly, would I still copy its answer somewhere by hand? If yes, you need a chatbot and a good prompt. If no — if the useful part is the action — you are designing an agent.
What changes when you build an agent
Agents fail differently. A chatbot that gets something wrong shows you the mistake. An agent that gets something wrong may have already sent the email. That is why agent work is mostly about boundaries: which tools it can touch, what it must ask before doing, and how you will see what it did.
Before you call it done
- Write the goal as a finished state, not a question
- List every tool the agent may use — and what it may not do with each
- Decide which actions need a human yes first
- Make sure every action leaves a record you can read
Answer four questions about any task and get a clear verdict: automate it, assist it, or keep it manual.
A free master guide and resource pack for people starting with AI automation: the systems behind voice agents, lead workflows and a small automation practice.
Get it free on Gumroad