Varun Vemulapalli

Day 35 of 100 · Phase 5: How AI agents work

Chatbot vs workflow vs agent

2 min readBy Varun Vemulapalli

Not everything with AI inside is an "agent".

A chatbot answers. A workflow follows your steps. An agent picks its own steps. Mixing them up is how projects get over-engineered, or under-built.

Video explainer · 1:54
Key takeaway

Use a chatbot to answer, a workflow when the steps are known, and an agent only when the steps depend on what it discovers along the way.

Day 35 visual: Chatbot vs workflow vs agent
The visual from the original post.

The concept

Three ways to put an AI model to work:

  • Chatbot: you ask, it answers. One question, one reply, often grounded in documents with RAG (Day 31). It doesn't take actions.
  • Workflow: a fixed path that people designed in advance. The AI does some steps (read a receipt, classify it), but code decides what happens next. Same input, same route, every time.
  • Agent: you give it a goal and some tools. The model decides which step to take, looks at the result, and decides the next one, until the goal is met or it hands off to a person.

More freedom = more flexibility, but also more cost, less predictability and more testing.

Walkthrough: travel expense reimbursement

Imagine the finance team at a mid-size company, three requests on one Monday.

1. Chatbot: "What's the meal limit for a client dinner?"

  • It retrieves the travel policy and replies: say, "$75 per person, receipt required. Source: Travel Policy, section 3."
  • Done. Nothing in any system changed.

2. Workflow: an employee submits a $62 dinner receipt.

  • Step 1: AI reads the photo and pulls out date, vendor and amount (Day 26, Day 27).
  • Step 2: code checks $62 against the $75 limit. Pass.
  • Step 3: route to the manager for approval, then to payroll.
  • Every claim follows the same three steps. Easy to audit, easy to test.

3. Agent: "My reimbursement from three weeks ago is stuck. Why?"

  • No fixed script fits this. The agent decides to look up the claim: status "pending approval".
  • It checks the approver: that manager is on leave.
  • It checks the policy for a backup approver and finds one.
  • It tells the employee what happened and drafts a re-route request for a person to confirm.

Notice: the agent chose its steps based on what it found. The workflow never would.

Where you see it at work

  • Banking: chatbots for product FAQs, workflows for account opening, agents for investigating an unusual dispute.
  • Insurance: a workflow routes simple claims; an agent helps an adjuster dig into a messy one.
  • Business analysts: if you can draw the process as a flowchart, start with a workflow. Use an agent only where the path can't be drawn in advance.

In one sentence

Use a chatbot to answer, a workflow when the steps are known, and an agent only when the steps depend on what it discovers along the way.

Your turn: name one process at your work. Is it a chatbot, a workflow or an agent job? 👇

Join the discussion on LinkedIn