Skip to content

Agent workflows

AI agents vs chatbots and workflows

Compare chatbots, AI-assisted workflows and tool-using agents with practical examples. Decide how much autonomy a task actually needs.

Jump to setup
Workstation showing conversation, sequence and branching diagrams

Imagine answering a visitor who asks about a syllabus. A chatbot can explain what is in an approved handout. A workflow can take a form submission through a fixed sequence. An agent may choose whether to look up a syllabus, check a catalogue or ask a follow-up question. These approaches can coexist in the same application.

Step-by-step setup

Compare the three approaches with one question

What you need A chat assistant, a note or spreadsheet, and optional access to the practice n8n workflow in the companion guide.

  1. Question
  2. Chat answer, fixed rule or tool lookup
  3. Compare outcomes
  1. Create a tiny practice catalogue

    Make one entry for Video Editing with “motion graphics” as a topic. Leave the batch date out.

  2. Try the chatbot version

    Give the assistant the one catalogue entry in the chat and ask both questions below.

  3. Try the fixed workflow version

    In a sheet or on paper, write a rule. If the question asks for the syllabus, return the topics. If it asks for a date, send it to admissions.

  4. Try the tool-using version

    Use the companion n8n setup guide. Give the agent a catalogue lookup tool and inspect whether it used that tool and what the tool returned.

Two questions to try

What does the Video Editing course cover?
When does the next batch begin?

How to tell it worked

Every version can mention motion graphics. Each should say the batch date is unknown or tell the learner where to confirm it.

Set up the n8n example

Use a chatbot for a conversation

The visitor enters a question and receives an answer. A useful course chatbot can cite the relevant course page and fall back to admissions when fees or dates are missing. Conversation describes the interface. A chatbot behind that interface may use tools, depending on how it was built.

Use a workflow when the sequence is known

A request comes in, the fields are checked, the message is summarized and a staff member reviews the draft. Ordinary rules control the sequence. A model can help with the summary while the workflow still decides when it runs and where its output goes.

Paper sketches comparing a conversation, a fixed sequence and a branching tree
A chatbot handles the conversation, a workflow runs a defined sequence, and an agent can choose from available tools.

Use an agent when tool choice can vary

A tool-using agent might look up a course for one request and search a list of documents for another. Give it a small set of clearly described tools. Set limits for the number of calls and require a person to review actions that change shared data or communicate externally.

Choose one task and measure the difference

Try a rules-only baseline before adding an agent. Compare how many questions get a useful answer, how many get forwarded and how often a reviewer needs to correct wrong information. Keep the simpler version if tool choice adds cost and delay without improving results.

Try it yourself

Sketch three ways to answer “What does the video editing curriculum include?” For each, note what starts the process, which data it needs and who checks the final answer. Then add a question for which the source has no answer and test the fallback.

Common questions

Can a workflow contain an agent?

Yes. A workflow can route a narrow task to an agent and validate the result before continuing.

Is an agent always more reliable?

Reliability depends on the task and how it is tested. A fixed rule is easier to verify for decisions such as checking whether a field is present.

Sources and further reading