Operation: Productivity · Day 1
16 JUNE 2026 · 14:00–17:00

Operation: Productivity

Artificial intelligence on your side

Day 1 — What AI is, how to write prompts, and what Microsoft offers.

🧠 AI foundations ✍️ Prompting 🟦 Microsoft Copilot

Employee Experience Domain · Productivity Services Team

Today's agenda

3 hours, 3 modules, 1 goal

TimeModule
14:00–14:15Opening · rules · the prize teaser
14:15–14:55Module 1 — What is artificial intelligence
15:05–15:50Module 2 — The art of the prompt
16:00–16:40Module 3 — Microsoft AI ecosystem + demo
16:40–17:00Day 2 preview · access check · Q&A

Tomorrow we build. Today we understand.

Why we're here

The Grand Prize 🏆

Tomorrow the team solves 6 missions of rising difficulty.

Complete them all and you unlock a personal photo certificate to print as a keepsake.

The real prize

Skills you'll use every working day — making AI do the boring work for you.

Module 1 · 40 min

What is artificial intelligence?

Where it comes from, how it "thinks", and what it can actually do.

A short history — how we got here

EraIdea
1950–2000Rules hand-written by people ("if-then"). Brittle.
2000–2017Machine learning — the machine learns from data, not rules.
2017Transformers ("Attention Is All You Need") — the breakthrough architecture.
2022 →ChatGPT and large language models (LLMs) reach everyone.

The key shift: we stopped writing rules and started training models on huge amounts of data.

How an LLM "thinks" — the intuition

One plain truth

An LLM is a machine that predicts the next word.

It has read a vast amount of human writing and learned which words tend to follow which.

Why it looks smart

At enough scale, "predict the next word" turns into the ability to summarize, translate, write code and reason.

📌 It is not a database and it does not "look things up". It generates plausible text.

Two concepts worth knowing

Tokens

Models read and write in tokens — chunks of words (≈ ¾ of a word each). You pay and are limited in tokens.

Context window

The model's short-term memory — how much text it can "see" at once. Beyond it, earlier content drops off.

Practical takeaway: give the relevant context, but don't dump everything — keep it focused.

How a model is built (simplified)

📌 This is why the same base model can be made friendlier, safer, or more specialized.

How a model is trained — the pipeline

1 · PRE-TRAINING Massive data web · books · code Predict the next token self-supervised, trillions of times Base model raw text completer 2 · POST-TRAINING SFT follow instructions RLHF / Constitutional AI helpful, honest, safe RL: code / math + tool use Assistant Claude · GPT · Fable
SFT — Supervised Fine-Tuning: learning from curated instruction → answer examples
RLHF — Reinforcement Learning from Human Feedback: humans rank answers, the model learns what's preferred
RL — Reinforcement Learning: learning by reward/feedback instead of fixed examples
Constitutional AI — the model critiques & revises its own answers against written principles (Anthropic's method for Claude)

What they do really well ✅

  • Summarizing long text
  • Drafting emails and documents
  • Translation and rephrasing
  • Extracting structure from chaos
  • Explaining things in plain words
  • Generating and explaining code
  • Ideas and brainstorming
  • Classification and sorting

Where they fail ⚠️

Hallucinations

They sometimes invent facts — confidently.

The 2026 landscape — who's who

ProviderRegionModelsNotable for
Microsoft + OpenAI🇺🇸 USGPT‑5In M365 Copilot & Copilot Studio
Anthropic🇺🇸 USClaude (Opus, Sonnet 4.5/4.6)Also selectable in Copilot Studio
Google🇺🇸 USGeminiGoogle Workspace
Mistral🇪🇺 FranceMistral Large 3, Ministral 3European, open‑weight (Apache 2.0)
Alibaba — Qwen🇨🇳 ChinaQwen 3.7 Max / PlusLeading Chinese flagship
DeepSeek🇨🇳 ChinaDeepSeek V4 / V4 ProOpen‑weight, low‑cost, self‑hostable

In Copilot Studio you pick the model — GPT‑5 and Claude are both available. Open‑weight models (Mistral, DeepSeek, open Qwen) can even run on your own infrastructure — relevant for data privacy.

Sources: Microsoft Copilot Studio 2026 release wave 1 (learn.microsoft.com); Mistral AI — Mistral 3 (Dec 2025); Alibaba Cloud — Qwen 3.7 (May 2026); DeepSeek V4 (2026).

Open vs closed weight

Weights = the billions of learned numbers that are the model.

🔓 Open weight

  • You can download the weights and run them yourself
  • Host on-prem, fine-tune, inspect — no vendor lock-in
  • e.g. Mistral, DeepSeek, Qwen (open), Llama
  • ✅ Data privacy   ⚠️ needs GPU infra, you maintain it

🔒 Closed / proprietary

  • Weights stay with the vendor — API / cloud only
  • Vendor handles infra, updates and safety
  • e.g. GPT‑5, Claude, Gemini
  • ✅ Easiest, often most capable   ⚠️ data leaves your tenant, ongoing cost

Note: open weight ≠ open source — you get the trained weights, but usually not the training data or recipe.

Module 2 · 45 min

The art of the prompt

Same model, garbage or gold — it depends on the prompt.

Weak prompt → weak result

❌ "Write an email"

The model doesn't know about what, to whom, in what tone. It returns something generic.

✅ With context

"You are a consultant. Write a short, professional email to a client who is postponing a meeting to next week. Polite, 4 sentences, propose a new date."

The RCTFE framework

LetterElementExample
RRole"You are an experienced legal assistant…"
CContext"…we're working on a vendor contract…"
TTask"…extract all the deadlines."
FFormat"Return as a table: clause | deadline."
EExamples"e.g. Payment | 30 days."

≥3 of these 5 = already a strong prompt.

Patterns that work

A note on grounding

A prompt only uses what's in front of the model.

🎬 Mini-exercise (5 min)

Take a weak prompt and improve it live with RCTFE.

Start: "summarize this"
Target: a 3-bullet summary + 1 recommended action, professional tone.

This is a direct warm-up for Mission 2 tomorrow.

Module 3 · 40 min

Microsoft AI ecosystem

From chat to your own agent — and the bridge to your world.

The three surfaces

M365 Copilot Chat

Free AI chat with web access. Your starting point.

Agent builder

Build your own agent with instructions — no code, inside Copilot Chat.

Copilot Studio (Full)

Full low-code portal — knowledge, actions, connectors, publish to Teams.

Source: learn.microsoft.com — "Agents for Microsoft 365 Copilot Chat" & "What is Copilot Studio" (2026).

What is an "agent"?

An agent coordinates a language model with:

📋 Instructions 📚 Knowledge sources 🗂️ Topics 🔧 Tools / actions ⚡ Triggers

It can answer questions, follow a conversation flow, and take actions on your behalf — across web, Teams and more.

Source: learn.microsoft.com/microsoft-copilot-studio (Overview, 2026).

Lite vs Full — when to use which

AspectLite (in Copilot Chat)Full (Studio)
For whombusiness userspower users / IT
Dataweb, simple filesSharePoint, connectors, APIs
Actionslimitedcustom actions / HTTP
PublishingsharingTeams, channels
Licensemostly freeCopilot license / PAYG

Free vs metered — the honest map

Free (no extra cost)

  • Copilot Chat prompts
  • Agents grounded in instructions
  • Agents grounded in public websites

Metered (Copilot Credits)

  • Grounding in SharePoint / Graph data
  • Custom actions / connectors
  • Included free with an M365 Copilot license

That's exactly why tomorrow has a FREE mode and a FULL mode.

Source: learn.microsoft.com/copilot/agents (Feb 2026). Since 1 Sep 2025 the meter is "Copilot Credits".

The bridge to your work 🧭

You already knowIn the AI world it's…
M-Files — a document repositoryKnowledge sources (grounding in documents)
A UiPath workflow — steps and logicTopics / conversation flow
A UiPath bot that "does" thingsAgent actions — the agent acts
Routine manual tasksPrompts for productivity

You're not learning something foreign — you're learning the AI version of what you already do.

Where this is going (2026) 🚀

Agents working together

Agent-to-agent communication — agents delegate tasks to each other.

Computer use

Agents that operate browsers and desktop apps on your behalf.

Agent evaluations

Test agents against sets of scenarios to measure quality.

Real-time voice

Talk to your agent naturally, out loud.

Source: Microsoft Copilot Studio 2026 release wave 1 & Copilot Studio blog (Apr–May 2026) — all generally available.

🎬 Live demo

We build a simple agent in 5 minutes.

  • Give it a name and a role
  • Write its instructions
  • Ask 2 questions → it stays in role

This is exactly Mission 3 tomorrow. For now, just watch.

Before we leave

Access check ✅

Now, together, everyone confirms they can:

Find a problem now → we have time to fix it. Tomorrow this would eat a whole mission.

Tomorrow: Operation: Productivity 🚀

M1 First Contact M2 Perfect Prompt M3 Build Your Agent M4 Knowledge Vault M5 Automation Link M6 Real-World Rescue ⭐ Deploy & Demo

Questions?

Employee Experience Domain · Productivity Services Team

References (verified)

Sources

1 / 1
← → to navigate · F for fullscreen