Back
1 / 0
→ next ← back F full ? help
Take a break
How long?
Break
00:00

Keyboard

Next slide→
Previous slide←
Full screenF
Black the screenB
First / last slideHome End
Close anythingEsc
Day 1 · Tuesday 8 September 2026 · 09:00 to 14:00

AI Essentials
in the Workplace

Foundations, responsible use, and prompting. Today you learn what this tool really does, and how to ask it properly.
CODED Campus, KuwaitTool: ClaudeDay 1 of 3
Today

Three things you
leave with.

01
A document you improved
One of your own. Real, not a sample.
02
A research summary you can defend
Every fact checked back to its source.
03
Five prompts you will reuse
Saved, named, and ready for tomorrow.
Before we start

First, a short test.

Pre-program MAP test
8 minutes, silent
This is your baseline. There is no pass mark, and your score is not shared with anyone. You take the same test again at the end of Day 3.
Link to be added
Question

What is one task or process
on your desk you dread the most?

Say it out loud. We are writing them all down.
Question

What is AI?

No wrong answers. Shout.
🤖

What is AI, really?

Software that can understand, learn, and generate content that used to need a person to do it.
📖Reads and understands text such as a contract, a report, or an email thread
✍️Writes content such as memos, notices, summaries and replies
💬Answers questions about anything you paste into it
🧩Helps you solve problems by laying out options, trade-offs and next steps
🔋Never gets tired and never forgets at 8 AM or 8 PM, the same quality
Think of it as a fast assistant, not a replacement for your judgment.
Section 1 1.1

You already use it
every day.

Your inbox
The spam filter learned what junk looks like. Nobody wrote a rule for each message.
Your maps app
It predicts the traffic in twenty minutes, then routes you around it.
Your bank
A card payment in another country gets flagged in under a second.
Your phone
Face unlock. Photo search for "beach". Autocorrect.
None of that felt like AI. It just felt like the thing working.
Question

So what is a
large language model?

Section 1 1.2

Step one:
it read.

Books, websites, code, manuals, forum posts, transcripts. More text than any person could read in a thousand lifetimes.
That is the "large" part. It is about the reading, not the file size.
Section 1 1.2

Step two: it kept
the patterns.

The important part
It did not keep the documents. It kept how words tend to follow other words. The library is gone. Only the habits are left.
This is why it cannot quote your company handbook. It never stored a copy of anything.
Section 1 1.2

You have seen this
a thousand times.

Your phone keyboard
Your keyboard read your messages and learned what usually comes next. Claude is the same idea, at an enormous scale.
Section 1 1.2

Therefore:
it predicts.

The one idea the whole day rests on
A large language model does one thing. It looks at everything so far, and predicts the most likely next piece of text. Then it does it again. That is the entire trick.
Everything good and everything bad about this tool comes out of that one sentence.
Question

Which one should
you actually open?

Section 1 1.3

The four you will
hear about.

Anthropic
Claude
Long documents, careful writing, analysis you need to trust. Our tool for three days.
OpenAI
ChatGPT
The one everybody knows. Broad, fast, huge add-on ecosystem.
Google
Gemini
Sits inside Google Workspace. Docs, Sheets, Gmail.
Microsoft
Copilot
Sits inside Office. Word, Excel, Teams, your company files.
They are closer to each other than the marketing suggests.Pick by where your work already lives, not by the brand.

Why should you care?

Not because it is new. Because of what it gives back to your week.
⏱️
Save time
Automate repetitive writing. Shorten long documents. Get a first draft in minutes instead of an hour.
✅
Improve quality
Catch errors, tighten unclear writing, and sound consistently professional.
💡
Think better
Weigh up options, organize messy information, and see a problem from angles you had not considered.
📈
Work smarter
Read data faster, prepare for meetings quicker, and spend your hours on judgment instead of typing.
None of this replaces you. It removes the part of the job that was never the point.
Section 1 1.4

Three things change
for one person.

Change 1
The blank page is gone
You never start from nothing. You start by fixing a draft, which is far easier.
Change 2
You get a second opinion
At 11pm. On anything. Before you send it to your manager.
Change 3
The boring part shrinks
The formatting, the first summary, the rewrite for a different reader.
Nobody is going to do this for you.IT can give you the licence. Only you know which of your tasks is worth it.
Section 1 1.4

One thing does
not change.

Your name goes
on the output.

If it invents a number and you send it, you sent it. Not the tool. That rule holds for every hour of the next three days.
Section 1 1.5

Strong here.
Weak there.

It is strong at
  • Turning rough notes into clean writing
  • Shortening something long
  • Changing the tone for a different reader
  • Translating between English and Arabic
  • Finding the pattern in text you give it
  • Structure: agendas, outlines, templates
It is weak at
  • Arithmetic, unless it writes the formula
  • Today's facts, prices, and news
  • Anything inside your company it was never given
  • Knowing when it is wrong
  • Saying "I do not know"
  • Your judgement about your own business
Section 1 1.5

A fast, confident junior
who has read everything
and remembers
nothing exactly.

You would check that person's numbers. Check these too.
Task 1 In Claude

First contact.

Task 18 minutesIn Claude
Your first useful result, on your own laptop
  1. Open Claude and start a new chat.
  2. Pick one starter from the menu and run it. There is an Arabic one too.
  3. Read what comes back. Would you actually use it? That is the bar.
  4. Turn to the person next to you and name one thing that surprised you.
Open task 1 →
Everything today runs on sample data we supply. Do not paste anything real from your own work. The rules for that come at 13:25.
Break

Section 1 done.
Take a break.

You know what it is, what it does, and where it fails. Prompting is next.
Pick a length. The countdown goes full screen and shows the room when to be back.
Section 2 2.1

Two prompts.
One task.

What most people type
Write me a report about last quarter.
The same task, written properly
You are a finance analyst writing for a busy director. Context: Q3 sales were 8% below target. The main gap was in the retail channel. I have the figures below. Task: Write the summary section of the quarterly report. Format: 200 words. Three short paragraphs. No bullet points. Tone: Direct and factual. No hedging. Constraints: Use only the figures I give you. If something is missing, ask me instead of estimating.
Same tool. Same task. The only thing that changed was the person typing.

You are
the variable.

Everyone in this room has the same model. The difference in what you get out is entirely the difference in what you put in.
Question

What actually makes
a prompt work?

Section 2 2.1

C T F T

Context
What is going on. Who the reader is. What already happened.
Task
The one thing you want done. One verb.
Format
How long. What shape. Paragraphs, table, bullet list, email.
Tone
Formal, direct, warm, technical. Say it, do not hope for it.
Most weak prompts have the Task and nothing else.
Section 2 2.1

Then two upgrades.

Upgrade 1
Role
Tell it who to be. "You are a procurement officer reviewing a supplier bid." This changes the vocabulary, the priorities, and what it thinks matters.
Upgrade 2
Constraints
Tell it what not to do. "Use only the figures I give you. Do not estimate. If something is missing, ask." This is your main defence against hallucination.
Constraints are the half people skip. They are the half that keeps you safe.
Task 2 In Claude

Build a
full prompt.

Task 214 minutesIn Claude
Six parts, one stand-in document, one real result
  1. Pick a scenario: a staff survey result, or a budget overrun. Both are sample data.
  2. Read the stand-in document, then fill all six fields in the builder.
  3. Press Assemble, press Copy, then run it in Claude.
  4. Compare the answer against the document. Did it miss anything? Did it invent anything?
Open task 2 →
Boss tier: drop one part on purpose and see which omission does the most damage.
Question

When is describing
not enough?

Section 2 2.2

Tell it, or
show it.

Zero-shot
You describe
"Write it in our house style: short, formal, no jargon." Fine when the style is easy to put into words.
Few-shot
You show
"Here are two emails we sent last month. Write the third one like these." The examples do the explaining for you.
Two good examples beat a paragraph of description. Every time.
Section 2 2.2

Use few-shot when the style
is real but hard to explain.

Your team has a way of writing customer replies. Nobody has ever written it down, and everybody knows when it is wrong. That is exactly the case for showing two examples instead of describing it.
Task 3 In Claude

Show, do
not describe.

Task 310 minutesIn Claude
Zero-shot against few-shot, same task
  1. Two sample customer replies on the page are the house style.
  2. Round one: describe the style in words. Ask for a third reply. Do not paste the examples.
  3. Round two, new chat: paste both examples, then ask for the third.
  4. Put the two side by side. Which one sounds like the team?
Open task 3 →
Break

Section 2 done.
Take a break.

You can write a prompt properly now. Next we point it at real writing.
Pick a length. The countdown goes full screen and shows the room when to be back.
Question

The first draft
is weak. Now what?

Section 3

Three repair moves.

Move 1
Cut it
"Half the length. Keep every fact." The fastest quality gain there is.
Move 2
Sharpen it
"Replace every vague phrase with a specific one. Point to the exact line."
Move 3
Change the reader
"Rewrite this for the finance director, who has ninety seconds."
Run them in that order, one at a time.Asking for all three at once gets you a worse version of all three.
Section 3

The same three moves,
in Arabic.

The draft
نود أن نحيطكم علماً بأنه وبعد الدراسة المستفيضة للموضوع قد تقرر تأجيل الموعد.
After "cut it, keep every fact"
تم تأجيل الموعد بعد المراجعة.
The moves are the same in both languages. Ask in the language you want the answer in.
Task 4 In Claude

Rewrite it.
Tone and length.

Task 410 minutesIn Claude
One clunky paragraph, six commands, same chat
  1. Copy the paragraph from the lab page into Claude.
  2. Run six rewrite commands in turn, in the same conversation. Do not start a new chat.
  3. The last one asks for it in Arabic, keeping the same tone.
  4. Notice what survives every version. That is the actual message.
Open task 4 →
Task 5 In Claude

Draft a real
document.

Task 515 minutesIn Claude
A brief in, a document you would send out
  1. Pick one of three briefs: a staff memo, a customer notice, or a status update.
  2. Write a full six part prompt for it. Do not skip Format or Constraints.
  3. Generate the draft, then edit it by hand. Fix the tone, check every fact, cut repeats.
  4. Read it out loud once. If you stumble, so will the reader.
Open task 5 →
This is your Day 1 deliverable. Bring it tomorrow.
Break

Section 3 done.
Lunch and prayer.

Two documents built. After lunch: research, then the rules for your own work.
Back at the time on screen. Bring your laptop.
Welcome back

Turn to the person
next to you.

Three minutes, seated. One question: which prompt from this morning will you actually use next week?
If your answer is "none yet", say that out loud. We will fix it in the next hour.
Question

Can I use it to
find things out?

Callback Section 1

Look at the
weak column again.

"Today's facts." "Knowing when it is wrong." Those two are exactly what a research question asks for.
So the answer is no for facts, yes for structure. The fix is to change where the facts come from.
Section 4

Give it the source.
Do not ask it
for the source.

Paste the report. Attach the PDF. Link the page. Then say: "Answer only from what I gave you. If it is not in there, say so."
This turns a guessing machine into a reading machine. The single most useful habit on this slide deck.
Section 4

The summarizing
brief.

Who reads it
"A department head with no background in this."
How long
"One page. Around 300 words."
Keep
"Every figure, and anything with a deadline attached."
Drop
"Background, methodology, and anything before 2024."
A summary without a named reader is just a shorter document.
Task 6 In Claude

Summarize,
then verify.

Task 614 minutesIn Claude
One operations report with five planted faults
  1. Download the report from the lab page and attach it to Claude. It is sample data.
  2. Ask for a one page summary: the position, the risks, the open questions.
  3. Then ask it to cite the exact line for every number it used.
  4. Pick three numbers and check them yourself. The report has five deliberate faults.
Open task 6 →
Finding one is a pass. Finding three is very good. Finding five means you are ready for Day 2.
Break

Section 4 done.
Short break.

Last stretch. Trust, privacy, and the rules you take back to your desk.
Pick a length. The countdown goes full screen and shows the room when to be back.
Question

How much of your week
is your inbox?

Say a number. Most people say too low.
✉️

Email, end to end

Three things Claude handles well on email. All three are just draft, rewrite and summarize pointed at your inbox.
📥Summarize a thread such as "what was decided, what is still open, and what do I owe anyone?"
✍️Draft a reply such as "answer the customer below, apologize for the delay, give Thursday as the new date, under 100 words"
🔁Match your own voice such as "reply in the same tone I used earlier in this thread"
Nothing here is new. It is Sections 2, 3 and 4 pointed at one place you already spend your day.
Section 5 5.1

Right now, all of this
is copy and paste.

You open the thread. You select it. You copy it. You switch tabs. You paste it. You read the answer. You copy that back.
It still saves you time. But it is the part people quietly stop doing after two weeks. Hold that thought.
Task 7 In Claude

Tame
the thread.

Task 712 minutesIn Claude
A long thread you were copied into, and the reply you owe
  1. Copy the thread from the lab page. Five messages, four people, one deadline.
  2. Ask for the summary in the shape you need: decisions made, questions still open, who owns what.
  3. Check it against the thread. Did it miss anything? Did it invent an owner?
  4. Draft your reply, then rewrite it once for tone.
Open task 7 →
There is a trap in the thread. Nobody actually agreed a date, but it reads like they did.
Question

What if it could read
the inbox itself?

Flexible block. Moves to Day 2 if we are short on time.
🔌

Connectors, and MCP

A connector gives Claude permission to read one of your tools directly. MCP is the standard that makes those connections work the same way everywhere.
📁No more pasting because Claude opens the thread, the file or the calendar itself
🔐You grant it, and you can remove it because a connector is a permission, not a copy of your data
🧰It works across tools because MCP is one standard rather than one integration per app
This is the difference between a clever tab and something that sits inside your actual work.
Watch this one

Live, on screen.

01
Turn a connector on
What the permission screen actually asks for, and what it does not.
02
Ask the same question as Task 7
Except this time nobody pastes a thread. It reads the thread itself.
03
Turn it back off
Because knowing how to revoke it is the part your IT team will ask about.
You are watching, not doing. Day 3 is when you build your own.
Callback This morning

The task you said
you dreaded.

Go back to the answer you gave at 09:15. Would you do it differently on Sunday?
Be honest about the ones that did not move. Some of them are Day 2 problems, and some are Day 3 problems.
Today

What you built.

01
An improved document
Lab 5. Cut, sharpened, aimed at a new reader.
02
A research summary you checked
Lab 6. Three facts traced back to the source.
03
Five reusable prompts
Labs 1 to 6, plus your paste rule at the top.
Homework, five minutes

Name your prompts
before you leave.

A prompt you cannot find again is not a saved prompt. Give each one a name that says when you would reach for it.
Good name
"Shorten a long email for a director"
Bad name
"Prompt 3"
Good name
"Summarize a supplier report, source only"
Bad name
"Untitled chat"
Bring the library tomorrow. Day 2 adds to it.
Tomorrow

Day 2: you turn this
into slides and numbers.

Claude Design and Gamma in the morning. Real data, real charts, and checking the maths in the afternoon.
Bring your prompt library. Bring a laptop that opens Excel.
End of Day 1

Thank you.

Day 2 starts at 09:00. Same room.