Skip to content

Introduction to Agentic AI and AI Agents · Section 1 · Notes

Notes: AI Agents: From Your Idea to Work You Can Use

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

Arun Nagarathanam · 29 Oct 2026 Notes · 2 of 10 in this course

Lecture 1.0

What Could AI Agents Make Possible in Your Work?

Most of us have one idea that keeps getting postponed because nobody has time to make the first version. This lecture shows how an agent can bring you that first version, so your team has something real to react to.

Key ideas

  1. 01A postponed idea: New customers keep calling to ask how to begin. A getting-started page and a welcome email could answer them, but the idea stays in a meeting.
  2. 02A first version: A page or an email your team can open, judge and improve, which is easier than reacting to an idea.
  3. 03No coding needed: You do not need to code or know AI terms. The course follows one familiar job first.

Why the idea waits. Making that page and email takes research, writing, design and tools. Nobody has the time, so it stays an idea.

An agent can start from three things you give it:

  1. a goal
  2. the documents that matter
  3. permission to use the files and applications the job needs

From there it can choose steps, make files, work across tools and bring you a version you can inspect.

You respond, the work moves. Your team points to what is missing, decides what to keep and asks for the next version. Once the team has checked the page and email and made them available, a new customer can find the first steps without making the same call.

You will also see a real project, my website aruntastic.com, including how the agent tested the result in a browser.

Lecture 1.1

What Can an AI Agent Do Beyond Giving You an Answer?

A correct answer in a chat can still leave your whole follow-up undone. Here you see the same meeting handled twice, once by an answer-only chat and once by an agent that saves files and adds tasks.

Key ideas

  1. 01Answer-only chat: States the facts correctly, but nothing is saved, the slides still advertise Friday and the board has no tasks.
  2. 02Brief: A short written update for the manager, saved in the shared team folder.
  3. 03To be confirmed: The agent marks the new date this way, because the team has not chosen one yet.
  4. 04AI agent: In this course, an application making choices and taking actions toward a goal. The same application may answer a question in one task and do connected work in another.

The situation. Your team's customer webinar is next Friday. Nina manages the signup page and Omar prepares the invitation slides, which still advertise Friday and say registration is open. The signup form breaks. Nina can fix and test it by Thursday, but the team wants customers to have several days to register, so it postpones the webinar and will choose a new date once the page works.

Your follow-up has three parts:

  1. a short brief for the manager, saved in the shared team folder
  2. a draft of updated invitation slides for Omar to review
  3. Nina's and Omar's tasks on the online project board

What the chat gives you. Ask 'What changed?' and the chat states the facts correctly. Nothing has been saved, the slides still advertise Friday and the board has no tasks.

What the agent does. Give it the transcript, the current slides, access to the folder and a working board connection. It finds the decisions, marks the date 'to be confirmed' and removes the open-registration claim. It saves the brief and the slide draft, then adds both tasks.

Notice that the request names the results but not which slide or which part of the transcript to change. The agent works that out.

Make it yours

  • Freelancer or consultant
    Your input
    Notes from your last client call and the draft you sent them.
    Ask AI
    Save a short summary for my records and draft the follow-up email.
    Check
    Every date and promise matches the notes, and the file is where you expect it.
  • Small business owner
    Your input
    Your notes from a staff or supplier meeting and any announcement that changed.
    Ask AI
    Prepare the follow-up and update the announcement. Leave undecided details open.
    Check
    Undecided details are marked as open, not presented as final.
  • On a team at work
    Your input
    A meeting transcript your employer lets you use, plus the file that needs updating.
    Ask AI
    Save a short update for my manager and a draft of the changed file.
    Check
    The tasks on the board match the decisions in the transcript.

Lecture 1.2

What Does an Agent Need to Work on Your Job?

An agent can only be as useful as what you let it read and change. This lecture walks through each source and each connection, so you know what to supply and what it still cannot do.

Key ideas

  1. 01Sources: The transcript records the decisions. The slides show what customers would see. You choose both, because you know which meeting and event the job is about.
  2. 02Read versus change: Reading a file and changing it are different. Saving the brief and the slide draft needs permission to change files in the team folder.
  3. 03Its own connection: The project board needs its own working connection and permission to add tasks.
  4. 04Approval stays with you: A draft email to registered customers can be reviewed now. Sending it waits for your team's approval.

Four things decide how much the agent can do on this job:

  1. The transcript. It records why Friday was postponed, Nina's Thursday repair and the decision to choose a date after registration works. Without it the update can be polished but miss the decisions.
  2. The slides. They show the Friday date and the open-registration claim customers would see. Without them the invitation stays unchanged.
  3. Permission to change files. Saving the brief and the slide draft needs it.
  4. A board connection. Without it, the agent can still write the task descriptions for you to add.

Check the board. Open it to see which result you actually got, the tasks added or only the descriptions.

No connection can supply a decision the team has not made, so the new date stays marked as undecided.

On the website project. The agent used the site's page files and notes and changed a test copy that I could open in a browser, while visitors kept using the public site.

Lecture 1.3

Can You Guide an Agent Before You Know the Final Result?

You do not need the perfect request before you start. A rough request gives you a version to judge, and one round of specific feedback makes it useful to a busy manager.

Key ideas

  1. 01Rough request: 'Turn the planning meeting into a short update for the manager.' It still gives you a version to judge.
  2. 02Order matters: Every fact can be correct while the decision is buried at the end, because the order follows the meeting.
  3. 03Specific feedback: Name what goes first. The same facts become more useful because you changed their order.
  4. 04Separate decisions: Approving a test copy and making a site public are two decisions, made one after the other.

If your manager reads two sentences. Those two sentences should hold the decision and who is responsible. A rough request gives a brief where the manager reads the page problem and the repair before finding out the webinar is postponed.

So the feedback names the order:

  1. the postponed webinar and the undecided date
  2. Nina's Thursday page repair and Omar's review and final revisions to the invitation slides
  3. the registration problem and the reason, after those decisions

The facts did not change. Clear direction helps from the start, and rough direction still gives you something to react to.

On the website. Early visual directions looked dated to me, so this is how I steered:

  • I gave the agent website-design videos showing styles I preferred
  • I chose and refined a later direction
  • I approved a test copy
  • I then made a separate decision to make the site public

Make it yours

  • Freelancer or consultant
    Your input
    A first draft the agent made for a client: a proposal, a summary or an email.
    Ask AI
    Put the decision and who owns it in the first two sentences. Move the background after.
    Check
    Reading only the first two sentences, the client knows what to do.
  • Small business owner
    Your input
    A draft update for your staff or customers.
    Ask AI
    Lead with what changed and what is still undecided. Put the reason last.
    Check
    Nothing undecided reads as final.
  • On a team at work
    Your input
    A draft status note for your manager.
    Ask AI
    Lead with the decision and the people responsible, then the detail.
    Check
    Your manager can act on the first two sentences alone.

Lecture 1.4

How Can an Agent Check the Work It Has Done?

Before you hand work over, ask the agent to check it against the source. Then you spot-check one fact that matters, so you are not hunting for every difference yourself.

Key ideas

  1. 01Check against the source: Ask the agent to compare its drafts with the transcript and report any mismatch, missing decision or undecided detail shown as confirmed, with the source passages.
  2. 02Spot-check: Read the report, then verify a fact that matters yourself, such as Nina's Thursday deadline. You decide whether the result is ready.
  3. 03Test like a visitor: With a browser, the agent chooses which links to follow and which buttons to try, so you do not describe every click.
  4. 04Fix, then repeat: If a test finds a problem, ask for the fix and repeat the same check.

Drafts first. If a draft announces a new webinar date, the transcript shows the team has not chosen one. The agent's report gives you a specific correction to review.

A good check loop has three steps:

  1. ask the agent to compare the drafts with the source and list mismatches
  2. read the report and spot-check one fact that matters
  3. ask for the fix, then repeat the same check

A website needs more than words. You also need to know what a visitor can do. In the recording, the agent tests the student resource library for my SEO course. Choosing Toolkit leaves only toolkit resources, and the agent opens the main toolkit on its own.

What it caught. 'Fix the Technical SEO Problems That Google Hates' has a description that calls it a toolkit, but it is tagged Guide, so the Toolkit filter hides it.

The agent finds the problem, and you decide what to do about it.

Make it yours

  • Freelancer or consultant
    Your input
    A report or proposal the agent drafted, and the notes it came from.
    Ask AI
    Compare the draft with my notes and list any mismatch or detail shown as confirmed that was not.
    Check
    One figure or date that matters, against your own notes.
  • Small business owner
    Your input
    A page on your website that the agent built or changed.
    Ask AI
    Test my website like a visitor. Tell me what you checked, what worked and what needs fixing.
    Check
    Open the one page or button the agent flagged and see for yourself.
  • On a team at work
    Your input
    A draft and the meeting transcript it came from.
    Ask AI
    Show the source passage behind each mismatch you find.
    Check
    The decision your manager will act on is stated as the team decided it.

Lecture 1.5

What Could You Put an Agent to Work On?

A bigger job is a small number of jobs you can name. This lecture breaks the customer-help project into four, and shows how a familiar job repeats with the next meeting.

Key ideas

  1. 01Start from what you have: Your company's existing instructions and the questions customers keep asking.
  2. 02Plain questions: Does the guide answer those customer questions? Does it put the first action first? If it leaves out how to get help, point to the support information.
  3. 03Team reviews: Your team reviews the page and the welcome email before anything is published or sent.
  4. 04A familiar job repeats: After the next webinar meeting, give the agent the new transcript and current slides.

The customer-help project becomes four jobs you can name:

  1. write the guide
  2. check its instructions
  3. make the page
  4. prepare the email

The guide first. Ask for a one-page guide to the first steps, saved in the shared folder. Judge it with the plain questions above, then ask the agent to compare the revised guide with the company instructions and check the answers customers are most likely to rely on.

Then the page and the email. Once you are happy with the guide, the agent prepares the website page and a welcome email that tells new customers where to find it.

The website redesign. The agent changed the home page, course pages and articles and helped create and place 211 refreshed article images. I compared designs, chose how to present my courses and tools, asked for a visitor test and approved making the site public.

The next meeting. If a date has now been chosen, the update and slides should show it.

Make it yours

  • Freelancer or consultant
    Your input
    Your own instructions for clients and the questions they keep asking.
    Ask AI
    Write a one-page getting-started guide, then draft the welcome email that points to it.
    Check
    The guide answers the questions clients really ask, with the first action first.
  • Small business owner
    Your input
    Your product or service notes and the questions customers call about.
    Ask AI
    Draft the getting-started page and welcome email. I will review before anything goes out.
    Check
    How to get help is included, and nothing is published until you say so.
  • On a team at work
    Your input
    Your company's instructions for new customers, from your own team's files.
    Ask AI
    Write the guide, check its instructions against ours, then prepare the page and email for review.
    Check
    The instructions match the company document.

Quick reference

What the agent needs

TranscriptRecords the decisions. Without it the update can be polished but miss them.
Current slidesShow what customers would see. Without them the invitation stays unchanged.
Permission to change filesSaves the brief and the slide draft in the team folder.
Board connectionAdds the tasks. Without it, the agent writes the descriptions for you to add.
Team approvalSending an email to customers waits for it.

Words used in this section

AI agentAn application making choices and taking actions toward a goal.
BriefA short written update for the manager.
To be confirmedHow the agent marks a detail the team has not decided.
Spot-checkVerifying one fact that matters yourself, after the agent's report.

Requests from the lectures

Rough first requestTurn the planning meeting into a short update for the manager.
Reorder the briefLead with the postponed webinar and the undecided date, then Nina's repair and Omar's slides, then the reason.
Check the draftsCompare the brief and slides with the transcript. Report any mismatch, missing decision or undecided detail shown as confirmed, with source passages.
Test a websiteTest my website like a visitor. Go through the main pages, follow links and try the buttons. Tell me what you checked, what worked and what needs fixing.

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