AI Agents for Leaders Who Have No Time to Learn AI · Section 1 · Notes

Notes: Finish the meeting with less work left to carry

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

Arun Nagarathanam · 20 Oct 2026 Notes · 2 of 42 in this course

Lecture 1.1

What could an AI agent take off your working week?

I want the preparation behind your decisions to stop eating the attention your people need from you. This lecture shows which part of that preparation an AI agent can carry.

Key ideas

01Preparation: Before you can answer a client, you find last meeting's notes, chase your team for missing facts and work out a plan if the date slips.
02AI agent: Software that works through several steps on its own and saves one document for you to inspect.
03Chat versus agent: A quick chat reads what you attach and answers once. An agent keeps going until the finished document is saved.
04Your part: You know why the date matters and which decision is yours, so you examine, question and improve the plan before you speak.

Picture one question. A client asks whether your team can still start on the date you agreed. Answering well means reading the notes from your last project meeting, asking your team what is still missing, and drafting what changes if the date is at risk.

When I give an agent the project notes and explain the conversation I need to prepare for, it can work through the steps on its own:

read the notes
sort what is ready from what is still waiting
draft the messages
save one preparation document for me to check

Where I stay in charge. A polished draft can still miss what a leader needs. My knowledge of the work is what closes that gap, so I read the document before I talk to anyone.

The result I am after is simple. I walk into the conversation knowing what is ready, what is uncertain and which choice needs discussing. Then I can listen to the client instead of searching through notes.

Lecture 1.2

Can AI handle the work after your meeting?

One meeting record can leave you ready for the next decision, if the agent knows what that decision needs. Here is how Alex's purpose turns a transcript into a page worth reading.

Key ideas

01Role list: The client's list of who needs access and what each person may do. Priya's team needs it before Tuesday's account setup.
02Transcript: The written record of what people said in Friday's meeting.
03Blocker: Something preventing the work from continuing.
04Drafts stay drafts: Every message waits for Alex, because the person whose name is on it checks it.

The situation. In this prepared fictional example, Alex's team met on Friday, and Alex meets the Northstar client at 11 a.m. on Monday. Jordan, the client contact, promised the role list by 3 p.m., once their operations manager confirms it.

A standard summary could repeat everything Jordan said and still leave Alex chasing updates before Monday's call.

So Alex tells the agent what Monday needs, alongside the transcript. Alex wants to know three things:

is the role list ready?
if not, what is delaying confirmation?
would a decision from Alex help?

The agent prepares one page, called Northstar follow-through, with three parts:

meeting notes that separate what is confirmed from what is still waiting
follow-up drafts, including an email asking Jordan for the role list
a proposed reminder from 10:00 to 10:15 a.m. on Monday, so Alex reads the replies before the review

The sentence that does the work. The draft to Jordan asks, if confirmation is delayed, what remains open and whether a decision is needed. Jordan now has a useful way to reply even before the list is final.

Nothing goes out and nothing goes on the calendar until Alex approves it. That leaves Alex's attention for the client on Monday, instead of rebuilding Friday's meeting during the call.

Make it yours

Freelancer or consultant

Your input
Notes or a recording summary from your last client call.

Ask AI
What each person owes me before our next call, and any decision they need from me.

Check
Every date and promise in the drafts matches what was said.
Small business owner

Your input
Your notes from a supplier or staff meeting.

Ask AI
Draft one follow-up per person asking for their blocker before Friday.

Check
Nothing is sent or booked until you approve it.
On a team at work

Your input
A meeting transcript your employer lets you use.

Ask AI
Sort ready from waiting, and draft the follow-ups for my review.

Check
The waiting items are the ones that affect your next decision.

Lecture 1.3

Can you talk it through instead of writing a perfect prompt?

You do not have to know the exact help to ask for. I start by talking the problem through and let AI's questions find the job.

Key ideas

01Prompt: The instructions you give AI to start or guide a job. It can begin as an unfinished spoken explanation.
02Interview me first: Asking AI to question you before it prepares anything gives it a problem to investigate.
03A proposed plan: AI turns your answers into a plan you can accept or adjust before any drafting begins.

Start with the worry, not the request. In ChatGPT you can turn on the microphone and say what is bothering you. Alex says the team records its meetings but still spends time chasing people, and what really worries Alex is finding out on the call that someone needs a decision Alex has not prepared for.

Then Alex adds one instruction: interview me about this problem, suggest how you could help, and let's discuss the approach before you start the work.

In the prepared exchange, AI asks three questions:

What would you want to know before the review that you sometimes discover too late?
What do you already have that could help prepare that?
What should each person you follow up with send you before the review?

One answer changes everything. Alex says a list of completed tasks does not show whether something is blocked or whether the team needs a decision. That single answer explains why an ordinary summary would leave the problem unsolved.

From the three answers, AI proposes a plan: organize what is ready and what is waiting, prepare follow-ups asking for blockers and decisions, and bring everything back for review. Alex checks that it answers the real worry and uses Friday's transcript, then says yes.

When I already know the exact output, I prompt directly. When the approach is still unclear, a short spoken interview finds it.

Make it yours

Freelancer or consultant

Your input
A spoken description of what worries you before a client review.

Ask AI
Interview me about this problem, then suggest how you could help.

Check
The proposed plan answers the worry you actually said.
Small business owner

Your input
A few sentences about a meeting that keeps leaving loose ends.

Ask AI
Ask me questions first, then propose one way to help.

Check
The plan uses material you already have.
On a team at work

Your input
What you are dreading about your next review, said out loud.

Ask AI
Interview me, then let's agree the approach before you start.

Check
You can accept or adjust the plan before any drafting begins.

Lecture 1.4

What can you now see AI taking off your week?

A draft that sounds professional can still leave you a message short of the answer you need. The test I use is what the reply lets me learn before my next decision.

Key ideas

01The real test: Does the draft merely sound finished, or does it set up the next decision you have to make?
02Generic request: "Please send me a progress update" can bring back a list of finished tasks and nothing about a decision.
03Purposeful request: Asking what is preventing progress and whether a decision is needed gets an answer while there is still time to act.

Two requests about the same project. If Alex sends Jordan a plain request for a progress update, Jordan could reasonably list the completed tasks and mention that the role list is still under review. Alex would then need another message to find out whether the delay needs a decision before Monday.

Alex's purposeful request asks what is preventing progress and whether a decision is needed before Monday's morning review. The reply to that one tells Alex what is stuck while Tuesday's options are still open.

Here is how the work splits between Alex and the agent:

the agent reads Friday's transcript, sorts what is ready from what is waiting, and writes the first draft of every follow-up
Alex supplies the purpose of Monday's conversation and judges whether the drafts ask for the right information

That partnership is the first real insight of the course, and it needs no technical vocabulary. Whenever I inspect an AI draft, I ask whether it sets up my next decision.

Make it yours

Freelancer or consultant

Your input
A status email you were about to send a client.

Ask AI
Rewrite this so the reply tells me what is blocked and whether I must decide anything.

Check
The client can tell exactly what to send back, and by when.
Small business owner

Your input
A message chasing a supplier for an update.

Ask AI
Make the request clear: what I need, by when and why it matters.

Check
No new deadline was invented.
On a team at work

Your input
A request for an update from a colleague.

Ask AI
Turn this into a request for their blocker and any decision they need from me.

Check
The reply would arrive in time for your decision.

Lecture 1.5

How does AI do the work, not just answer?

You direct an agent with more confidence once you understand what happens between handing over the transcript and getting the page back. Here are the pieces, all on Alex's one job.

Key ideas

01Generative AI: The ability to create new wording, like the sentences in Jordan's draft.
02Model: The part that reads, reasons and writes. In ChatGPT, one of OpenAI's GPT models.
03Tool: An ability to take an action, such as reading a file or saving one.
04Agentic AI: How the agent works: it chooses and adjusts its own next steps toward the goal you set.

Writing is only one part of the job. The model can only write text. Friday's transcript sits in a document file, and Alex wants the finished page saved, so the software needs a tool to read files and another to save them. Even a quick chat that reads a file you attach is using a tool.

On Alex's job, the pieces connect in this order:

a reading tool brings in the transcript
the model interprets the discussion and writes the notes and drafts
a saving tool stores the finished page for Alex to inspect

An agent can only take the actions its tools allow and that you have given it permission to take. In ChatGPT, this kind of work happens in a mode called Work, which Section 2 opens.

Why it is called agentic. Suppose the agent reads that the role list still needs the operations manager's confirmation. Nobody told it to check what depends on that list, but it searches the rest of the transcript, finds that Priya's team needs it before Tuesday, and adds it to the waiting items. When the purpose is missing, it stops and asks Alex.

Rule and skill. A rule is one short instruction, such as asking the leader when the purpose is missing. A skill is a saved set of rules with examples or supporting files, so I do not invent the same instructions after every meeting. Saving one gives the agent a method to reuse, not another piece of software to look after.

Lecture 1.6

Which task should you hand over first?

Handing an agent everything at once tends to create a second job: correcting it. I start with one contained job I can judge at a glance.

Key ideas

01One review first: Alex already has the transcript, knows what Monday must settle and can judge the drafts quickly.
02Brief: A short description of the job, its inputs and its return point. AI can draft it from your conversation.
03Workflow: A repeatable sequence of steps for a bounded job. It can use one skill or several.

Agendas, tracking every discussion, following up with everyone and scheduling meetings involve different people, files and judgment calls. When a week-wide job disappoints, I cannot tell whether the agent lacked background, picked the wrong meeting or took an action I wanted to keep.

So Alex asks AI to turn the plan agreed in the interview into a short job description. AI writes back three sentences: prepare the notes and follow-up drafts, request blockers and decisions before the review, and return everything for Alex to review.

Before anything runs, Alex makes 3 checks on that brief:

the inputs already exist, in Friday's transcript and Alex's explanation
the output is a concrete set of notes and drafts
it stops and returns for review instead of sending anything

Those checks make any correction obvious. If the draft to Jordan forgets to ask whether a decision is needed, Alex points straight to the missing line.

Keep it or change it. Right after Monday's review, Alex decides. If the drafts needed only small corrections, the workflow is worth keeping. If Alex still had to chase people, the brief needs changing first.

Four sizes of instruction an agent can keep:

a rule is one short instruction
a brief describes one job
a skill saves a method, with its rules and examples
a workflow is the repeatable sequence of steps for the whole job

Make it yours

Freelancer or consultant

Your input
The plan you agreed with AI for one client follow-up.

Ask AI
Turn the plan we agreed into a short description of the job.

Check
Inputs exist, output is concrete, it returns for review.
Small business owner

Your input
One recurring meeting, such as the Monday staff huddle.

Ask AI
Write a brief for preparing just this meeting's follow-ups.

Check
Nothing in the brief sends or books anything.
On a team at work

Your input
A single recurring review in your week.

Ask AI
Describe the job, its inputs and where it stops for my review.

Check
After the meeting: small corrections, or still chasing people?

Lecture 1.7

How to Get the Most Out of This Course

I built this course for leaders who want to understand a strategy today and try it when the right meeting arrives. Here is how to follow now and come back later.

Key ideas

01Follow first: The lessons follow one Northstar meeting from the finished page to the brief, with no homework in between.
02Return by problem: Each lesson solves one job, so you can go straight back to the one you need.
03Fictional files: Practise on the prepared Northstar case instead of confidential work documents.

When a real need comes up, go back by the problem you have:

work to chase and unsure what to ask AI for: the interview lesson, where Alex answered 3 questions
tempted to hand over too much: the first-workflow lesson, where Alex made 3 checks on a brief

The Section 1 starter zip holds:

the Northstar meeting transcript and notes about Alex's team
the worked follow-through page
the copyable requests
the optional progress-update comparison with its sample answer
the meeting workflow and the skills it uses

When you apply a method to your own work, use the documents and AI products your employer permits.

Ask rough questions. If an explanation feels unclear, post in the course Q&A. If an attempt gives an unexpected result, share the prompt you used and a non-confidential part of the draft. For example, you wanted it to ask whether a decision was needed, and it only asked for a status update.

Lecture 1.8

What should you keep from this meeting?

The habit that lets your judgment build up over time is keeping the reason a draft worked apart from the dates it worked on.

Key ideas

01Preference: One short rule written with its reason, such as making the action and its purpose clear.
02Project fact: Northstar's Monday, October 5, 11 a.m. review belongs in Northstar's own notes.
03Travel test: Apply the preference to a different task and see whether it still works.

Let the agent write it. At the end of the conversation, Alex asks the agent to capture what made the follow-up useful and why, keeping this meeting's facts separate.

The agent proposes three sentences. Make clear what the recipient needs to do and why the response matters. Use the supplied context, and ask when the purpose is missing. A clear action and reason help the recipient understand what response will be useful. Alex's job is to check that it keeps the reason the follow-up worked.

Why the date stays out. If Monday's date sat inside a general follow-up instruction, the next meeting would inherit a deadline from the wrong project.

The travel test. A leader asks a colleague for next month's support cost so it can be compared with the budget before deciding. The subject has changed and the Northstar date is gone, yet the same structure makes the request work.

Right now those three sentences live only in Friday's conversation and on Friday's page. A fresh chat next week will not bring them along, and Section 2 is how you fix that.

Quick reference

Words used in this section

AI agent	Software that works through several steps on its own and returns the finished work for you to check.
Generative AI	The ability to create new wording, such as a message or a summary.
Model	The part that reads, reasons and writes.
Tool	An ability to take an action, such as reading or saving a file.
Agentic AI	Choosing and adjusting its own next steps toward the goal you set.
Prompt	The instructions you give AI to start or guide a job.
Transcript	The written record of what people said in a meeting.
Blocker	Something preventing the work from continuing.

Four sizes of instruction

Rule	One short instruction, such as asking when the purpose is missing.
Brief	One job: what it is, its inputs and where it returns for review.
Skill	A saved method, with its rules and examples, reused on the next job.
Workflow	The repeatable sequence of steps for the whole job. It can use one skill or several.
Preference	A short rule with its reason that you want in every similar piece of work.

Use X when

Prompt directly	When you already know the exact output you want.
Spoken interview	When the best approach is still unclear.
One contained job	When you are tempted to hand over everything touching your meetings at once.

Check before you rely on it

A draft	Does it set up your next decision, or does it only sound finished?
A brief	Inputs exist, output is concrete, it returns for review without sending anything.
A workflow	After the meeting: small corrections means keep it; still chasing people means change the brief.
A preference	It keeps the reason the follow-up worked, and no dates or names from this project.

Numbers from the Northstar example

Friday, October 2	The team meeting. Jordan promises the role list by 3 p.m.
Monday, October 5, 11 a.m.	The client review.
10:00 to 10:15 a.m. Monday	The proposed reminder to read replies. Proposed only.
Tuesday, October 6	The planned account-setup start.