Chatbot versus agent
CHATBOT
Question → model → answer → done
AGENT
Goal → decide → act/tool → observe → decide again → finish- The feedback loop is the important difference.
- A chat interface alone does not make software agentic.
Agentic AI Engineering · Module 01 of 16
What Is an AI Agent?
A chatbot usually answers a message. An agent is an application that can work toward a goal by deciding what to do next, using bounded tools when needed, observing the result and continuing until it can stop safely.
Engineering focus: Agents, workflows and goal-directed AI
In plain English
What problem are we solving?
A chatbot usually answers a message. An agent is an application that can work toward a goal by deciding what to do next, using bounded tools when needed, observing the result and continuing until it can stop safely.
Framework names come later. First understand what the application is doing and where its control boundaries belong.
Running case · Operations Investigator
Across the course, imagine an assistant that investigates operational problems using narrow tools. The design progressively adds plans, specialist agents, approvals, reliability controls and MCP while keeping model output separate from execution authority.
Learning objectives
By the end of this module, you should be able to:
- Distinguish an LLM, chatbot, deterministic workflow and agent.
- Recognize when an agent adds value and when a workflow is safer.
- Describe agentic behaviour as bounded decisions plus actions plus feedback.
Start with the job, not the word “agent”
Many applications use an LLM without being agents. If the steps are already known, a deterministic workflow is usually easier to test, secure and operate.
Agentic behaviour becomes useful when the next action genuinely depends on what the system observes while working toward a goal.
Four terms that beginners should keep separate
An LLM produces or interprets language. A chatbot provides a conversational interface. A workflow follows predefined steps. An agent can choose among permitted actions as the task unfolds.
- LLM: language model capability
- Chatbot: conversational application
- Workflow: predefined control flow
- Agent: bounded dynamic control flow toward a goal
A simple decision rule
Ask whether you know the exact steps in advance. If yes, prefer a workflow. If not, ask whether bounded dynamic decisions are necessary. Only then consider an agent.
Worked examples
Choose the simpler architecture
Resize image → WORKFLOW
Summarize document → LLM TASK
Investigate intermittent 503 → POTENTIAL AGENT
Generate fixed invoice → WORKFLOW- Use dynamic decision-making only where it creates real value.
- Predictable tasks benefit from predictable control flow.
Try it · core lab · account-free
Classify agentic and non-agentic tasks
Goal: Practice deciding whether a task needs an LLM, a deterministic workflow or a potential agent.
Before you start
- Python 3. No API key, model download or GPU is required.
Download or open the lab
Use the supplied account-free Python exercise.
python3 public/labs/agentic-ai/01_classify_agent_or_workflow.pyClassify each scenario
Commit to LLM, WORKFLOW or POTENTIAL_AGENT before viewing the expected classification.
Explain one ambiguous case
State what observation would make the next action change.
Verify
- The script reports your score.
- You can explain why an LLM step inside fixed control flow is still a workflow.
Expected outcome
- You identify investigation tasks as potentially agentic.
- You prefer deterministic workflows when the exact steps are already known.
Break / fix · diagnose
- If everything looks like an agent, ask whether the next action really depends on an observation.
- If everything looks like a workflow, look for tasks whose next diagnostic step cannot be known before the previous result.
Production view
Teams often over-agentize simple automation. Choosing the simplest adequate control flow reduces cost, latency, security exposure and operational uncertainty.
Avoid these traps
Common mistakes
- Calling every application that uses an LLM an agent.
- Choosing an agent because the term is fashionable rather than because the task needs dynamic decisions.
- Assuming more autonomy automatically means a better system.
Knowledge check
Explain the control decision, not just the vocabulary.
Open every answer only after committing to your own answer. Opening all four answers records knowledge-check activity in My Learning.
1A program always runs extract → validate → save, but uses an LLM during extraction. What is it most accurately called?
A deterministic workflow with an AI step — Using an LLM in one predefined step does not make the whole system agentic.
2Which task most strongly benefits from bounded dynamic decision-making?
Investigate an intermittent 503 using several diagnostic sources — Investigation may require choosing the next action based on previous observations.
3What is the safest default when the exact steps are already known?
Use a deterministic workflow — Deterministic workflows are easier to reason about when the steps are known.
4What characteristic best separates an agent from simple text generation?
Dynamic action selection with feedback from results — Agentic behavior centers on choosing actions and incorporating observations.
My Learning
Record your progress for this module.
Progress is saved in this browser immediately. When you sign in to My Learning, it can also sync across devices.
Attempt the core lab and knowledge checks before marking the module complete.