Skip to content

Introduction to Agentic AI and AI Agents · Section 3 · Notes

Notes: Tell Agentic AI, Chatbots and Workflows Apart

Read a lecture instead of watching it, then make each idea yours.

Arun Nagarathanam · 29 Oct 2026 Notes · 7 of 10 in this course

Lecture 3.0

Is Everything Called an Agent Really One?

By the end of this section, when a colleague calls a product an assistant, an agent or an automation tool, you can say what it actually does. It starts with one question about steps.

Key ideas

  1. 01The one question: How are this product's steps chosen? Does it follow a sequence somebody prepared, or can the model choose another action after reading a result?
  2. 02Name versus work: The name on a product does not tell you what it does. The way it chooses its steps does.
  3. 03ChatGPT as the test case: ChatGPT can answer a question, search a website or carry out several actions, so it needs an explanation that fits the work in front of you.

Why one product has many names. ChatGPT sometimes just answers, sometimes searches a site, and sometimes carries out several actions in a row. A single label cannot cover all three, so the section looks at the work instead of the label.

Before the next lecture, try this on a product you know:

  1. Think of one AI product someone at work has called an agent.
  2. Write down what you have actually seen it do.
  3. Keep that note. You will test it against the questions in the next lectures.

Lecture 3.1

Is ChatGPT an AI Agent?

ChatGPT is not always an agent. It acts as one when it chooses its next step from what it finds, and you can check that yourself on any price question.

Key ideas

  1. 01Agentic work: The model chooses a next action from what it finds and uses tools to continue toward your result. The system doing it is an agent.
  2. 02Workflow: Finding the pages, reading the prices and preparing the answer. When the model chooses and takes actions inside it, it is an agentic workflow.
  3. 03Training versus search: Asked for a current price, ChatGPT can answer from what the model learned in training, which may be an older price. With web tools it can search and give you the source page.
  4. 04Read the source: Notion's pricing page opens on yearly billing, so it shows $20 beside ChatGPT's $24 monthly answer. Open the source and read which billing option it shows.

Two ways to answer. Without checking the pricing page, ChatGPT may give you a price from training. With web tools available, it can search for the price instead and give you the source page, such as the Notion pricing page in the recording.

The two numbers are both right. The $24 is a monthly price and the $20 is what the page shows on yearly billing. The check is to open the source and see which billing period it shows.

Judge the work. Look at the task it performs, the tools it uses and the choices it makes, not the product name.

Make it yours

  • Freelancer or consultant
    Your input
    A tool you pay for, such as a design or writing app.
    Ask AI
    Ask ChatGPT, with web search on, for its current monthly price.
    Check
    Open the source page and compare the price and the billing period.
  • Small business owner
    Your input
    A supplier subscription your business uses.
    Ask AI
    Ask for the current monthly price with the source link.
    Check
    Note whether it searched, and whether it chose another page after the first one.
  • On a team at work
    Your input
    A software tool your team subscribes to.
    Ask AI
    Ask for the current monthly price for one plan, with web search on.
    Check
    Confirm the billing period on the source page before you quote the figure.

Lecture 3.2

How Do Chatbots, Fixed Workflows and AI Agents Fit Together?

A chat window shows how you talk to software, not how it chooses its steps. Two questions about the steps tell you more than the window ever will.

Key ideas

  1. 01Fixed workflow: Opens each saved page, reads the price, compares it with last week's price and writes the report. Someone prepared those steps before the job began.
  2. 02Agent: Works on the same job with more freedom. It can choose a different next action when something changes, such as a missing price.
  3. 03Working together: An agent finds the prices, then a fixed calculation works out how much each price changed.
  4. 04Freedom has a cost: More freedom helps when you cannot know every step in advance, and it gives the agent more choices to get wrong.

Same job, two systems. In the supplier-price job, a fixed workflow marks a missing price as missing. In the recording, the agent reported Not confirmed for a saved page that had lost its price, and the activity log showed it chose to search the supplier's website.

What a prompt cannot tell you. A long list of steps in a prompt does not show which kind of system carries them out. Both kinds can use web tools, and either can start from a chat.

Check both. Prepared steps make mistakes too. Before using any report, check that each price belongs to the right product and that each calculated change is correct.

Make it yours

  • Freelancer or consultant
    Your input
    A repeated job, such as checking client rates or collecting figures.
    Ask AI
    Write the steps you could prepare in advance.
    Check
    Mark where a different next action might be needed.
  • Small business owner
    Your input
    A weekly price or stock check for your shop.
    Ask AI
    Split the job into prepared steps and points where the agent chooses.
    Check
    Each price belongs to the right product, and each calculated change is correct.
  • On a team at work
    Your input
    A report your team collects every week.
    Ask AI
    List which steps are fixed and which depend on what the agent finds.
    Check
    A person reviews the figures before the report is used.

Lecture 3.3

What Starts an AI Agent?

A job's start tells you when work begins, never how its steps are chosen or whether it finished. Keeping those three apart clears up most of the confusion.

Key ideas

  1. 01Manual start: You open an application and ask, so your request starts the job.
  2. 02Scheduled start: A system that supports scheduled work runs the comparison at a set time, such as every Monday at 9 a.m.
  3. 03Event start: A service configured to detect invoice emails starts the work as each one arrives.
  4. 04Trigger: The request, time or event that starts a job. It tells you when work begins, not how the steps are chosen.

A job can start in three ways:

  • Manual: you ask.
  • Scheduled: the clock reaches the time the system was set to run.
  • Event: something happens, such as an invoice email arriving.

Writing a schedule does not set one. Writing every Monday at 9 a.m. in a message does not set up a schedule. The system has to support scheduled work.

Same trigger, different work. Two Monday price checks share a scheduled trigger. The one where the model chooses another page to search is the agentic one.

Notification is not completion. A message saying A new invoice arrived is a notification. The work is done only when the invoice details are in your spreadsheet. Describe a job with both its start and the result you expect: when an invoice email arrives, add its details to this spreadsheet.

Make it yours

  • Freelancer or consultant
    Your input
    A recurring task, such as sending invoices or reports.
    Ask AI
    Describe it as a start plus a result.
    Check
    Which of the three starts it uses, and whether the system supports it.
  • Small business owner
    Your input
    Invoice emails that reach your inbox.
    Ask AI
    When an invoice email arrives, add its details to this spreadsheet.
    Check
    The details are in the spreadsheet, not only a notification.
  • On a team at work
    Your input
    A weekly job your team repeats by hand.
    Ask AI
    Name the trigger and the finished result.
    Check
    The schedule is set up in a system, not only written in a message.

Lecture 3.4

How Do AI Agents, Agentic AI, Assistants and Automation Differ?

One product can honestly be called an assistant, an agent, agentic AI and automation at the same time. A customer email to an online shop shows each word in turn.

Key ideas

  1. 01Assistant: An AI product that helps you prepare the reply, because it helps you with your work.
  2. 02Agent: The system that looks up the order, finds the delivery is late, and chooses to check the latest tracking update before writing.
  3. 03Agentic AI: AI choosing and taking actions toward a goal.
  4. 04Automation: Software that prepares a reply whenever a customer email arrives. An automated process can follow fixed steps or use an agent.

The shop email. A customer emails your online shop: Where is my order? The assistant helps you reply. When the model finds the delivery is late and chooses to check the delivery company's tracking before writing, that system is an agent and its way of working is agentic.

Three separate actions sit inside that one reply:

  1. Reading the order.
  2. Preparing a message.
  3. Sending it.

Permission follows the action. You can allow the product to prepare replies and still require your approval before it sends one.

Make it yours

  • Freelancer or consultant
    Your input
    A client email you answer often.
    Ask AI
    Have the assistant draft a reply you can edit.
    Check
    You approve before anything is sent.
  • Small business owner
    Your input
    A where-is-my-order email from a customer.
    Ask AI
    Prepare a reply after checking the latest delivery update.
    Check
    Which action is still yours: the send.
  • On a team at work
    Your input
    A shared support inbox.
    Ask AI
    Decide which actions the product may do alone.
    Check
    Reading, preparing and sending each have their own permission.

Lecture 3.4A

How Do You Talk About AI Agents in a Meeting?

In a meeting you rarely need a definition. You need a sentence that helps people decide what an AI should do, or why a result needs attention.

Key ideas

  1. 01Agentic: Most useful when you give the action and the reason: it chose another search because the first page did not answer the question.
  2. 02Agent-washing: The gap between what an advertisement promises and what the demonstration shows.
  3. 03Connectors, tools and APIs: Ask what a connection can actually do, such as create a calendar event or accept a cancellation request.
  4. 04Steps and start: Separate how the work proceeds (prepared steps or model choices) from what starts it (which trigger).

Twelve terms, one sentence each. The Meeting vocabulary companion gives twelve terms, each with a sentence you can use in a meeting and the lecture where it was explained.

Adapt, keep the meaning. Change the example to your own work, and keep the meaning of the sentence the same.

Lecture 3.6

Which Shape Was That?

Back to the shop email, with every word from this section in its place. The trigger tells you when the job begins, and the later choices tell you how it works.

Key ideas

  1. 01Prepared workflow: Looks up the order and drafts a reply with fixed steps.
  2. 02Agentic version: The model found the delivery was late and chose an extra tracking check before writing.
  3. 03Tool and permission: A tool supplied the current information. Sending stayed a separate action that needs permission.
  4. 04Trigger: The email's arrival could start the process automatically, or a person could start it.

For the shop email, name each part in ordinary language:

  • When the job begins: the trigger.
  • How it works: fixed steps, or the model choosing the next action.
  • What is still yours: approving the send.

Neither the word assistant nor a chat window settles those questions.

Quick reference

How the steps are chosen

Fixed workflowSteps prepared before the job begins, such as open the page, read the price, compare, write the report.
AgentChooses a next action from what it finds and uses tools to continue toward your result.
Agentic workflowA workflow in which the model chooses and takes actions.
Agentic AIAI choosing and taking actions toward a goal.

What starts the work

Manual startYou open an application and ask.
Scheduled startA system that supports scheduled work runs at a set time, such as every Monday at 9 a.m.
Event startA configured service starts the work when something happens, such as an invoice email arriving.
TriggerThe request, time or event that starts a job.

Words for meetings

AssistantAn AI product that helps you with your work.
AutomationSoftware that does a job whenever its start happens. It can follow fixed steps or use an agent.
Agent-washingThe gap between what an advertisement promises and what the demonstration shows.
NotificationA message that something happened. The work is done only when the result exists.

More in this section

Keep going

Keep building

More from Aruntastic

Practical courses for people who want to build useful things with AI.

Toolkits and programme

Go deeper

Claude certification practice

Preparing for a Claude certification?

Practice exams written by Arun: 600 questions across 4 mock exams and 2 drill banks. Unofficial preparation, not affiliated with Anthropic.

Keep going

This resource is part of an 11-section course.

Follow the lectures on Udemy, then come back here whenever you want the files, requests and notes for a section.

Open the course Introduction to Agentic AI and AI Agents · Udemy