Nine modules. Less watching, more doing.

You have seen the keynote. Today we roll up our sleeves: we discuss, you practice, we debrief, you apply.

• 01
  Beyond the headlines
• 02
  Your thought partner
• 03
  Make it yours
• 04
  Sharper thinking
• 05
  Connect AI to your world
• 06
  Build with Cowork
• 07
  What are AI agents?
• 08
  AI adoption in Canada
• 09
  The path for your business

Is AI going to replace all the jobs?

The 2026 headlines say AI is taking the jobs. Zoom out: Amazon ran the largest corporate cuts in its history and still employs roughly double its 2019 staff.

But sometimes it really is AI.

The work is going to explode, not disappear.

Jevons Paradox, 1865: when James Watt made the steam engine more efficient, Britain burned more coal, not less. Cheaper to use meant used everywhere. Efficiency expands demand. The same thing happens with capability.

You bracket the work. AI does the middle.

The skill is not handing everything to the AI. It is knowing which part is yours. You frame the work and you verify the result. The AI carries the load in between.

The shape of a strong ask: Role, Context, Task.

A prompt structure is a repeatable shape you fill in. Same skeleton every time. It forces you to name what the AI needs up front, so it stops guessing and answers what you actually meant. Role, Context, Task is the one to learn first.

Context is what makes it sharp.

Same question, three layers. Watch the answer climb as you give it a role, then the real specifics, then the exact output you want.

Rewrite a real prompt as Role, Context, Task.

Think of something you would type into ChatGPT or Claude today: an email, a plan, a tough message. Pull it apart and rebuild it with these three layers.

• R
  Role

  Who should the AI be? What experience or perspective should it bring to this?

• C
  Context

  What does it need to know to be useful? The audience, the constraints, the specifics only you have.

• T
  Task

  What exactly do you want produced? Length, format, tone, what to include and leave out.

Debrief

Turn to the person next to you. Read each other your before and after.

Your pocket template.

Copy this. Keep it where you work. Fill in the blanks any time you want a sharper answer on the first try.

Role, Context, Task - Three layers. Always in this order.Role: You are [the role with the experience or perspective that fits this].

Context: [What this is for. Who will see it. The constraints. The specifics that make this not generic.]

Task: [What you want produced. Length. Format. Tone. What to include, what to leave out.]

Before you give me the final version, ask me anything you'd need to make this right.

Four things worth telling it once.

Draft your custom instructions.

Fill in the brackets. Five minutes now saves you an edit on every chat from here on. Keep it tight; you can always add more later.

Starter: tell the AI who you are - Edit everything in brackets, then paste into Settings > Custom instructions.Who I am: [your name, your role, your company and what it does].

What I care about right now: [your top 2-3 priorities or challenges].

How I like answers: [direct or detailed? bullet points or prose? any formats you use a lot?].

Things to remember: [anything it keeps getting wrong, names, products, your market].

Debrief

Compare with the person beside you.

The lazy way: let it interview you.

Do not stare at a blank box. Paste this, answer its questions, and let it write a clean set of instructions for you to drop into settings.

Have the AI build your instructions - Paste into a fresh chat. Answer the questions. Copy the result into settings.I want to set up strong custom instructions so you understand me in every future chat.

Interview me: ask me one question at a time about my role, my company, my current priorities, and how I like answers. Keep going until you have enough.

Then write a tight set of custom instructions in the second person ("You are working with...") that I can paste straight into my settings.

Reverse prompting: make it ask before it answers.

When you only half-know what you want, do not write a long prompt. Hand it the messy intent and say: "Interview me until you have what you need, then answer." It pulls the context out of your head, one question at a time.

Devil's advocate: have it argue with itself.

The first draft is rarely the best one. Before you accept anything, make it attack its own work: "What is weak here? What would a tough critic say? Now fix it."

Run both on something real.

Pick a live one, not a hypothetical. The point is to feel it change your thinking.

Debrief

What did it surface that you had not considered?

Two prompts worth keeping.

Save these two lines. They turn the AI from a faster typist into a sharper colleague.

Reverse prompt + devil's advocate - Use the first when you are unsure; the second before you commit.REVERSE PROMPT:
"Before you answer, interview me one question at a time until you have everything you need to do this well. Then answer."

DEVIL'S ADVOCATE:
"Now take the other side. What are the three weakest points in this? What would a sharp critic say? Then rewrite it to fix them."

Custom instructions gave it memory. Connections give it reach.

Once connected, the AI can read and act where you already work, with your permission.

What it unlocks, and where to stop.

Turn one connection on this week.

You just saw one live. Now make it yours, starting safe.

• 01
  Pick the safe one

  Your own email or calendar. Nothing client-confidential to start.

• 02
  Connect it

  Two minutes in settings. Grant access to that single account.

• 03
  Ask it something real

  "What needs me today?" or "Prep me for my next meeting."

Describe it. It builds it.

Cowork is a mode where the AI does not just answer, it builds. You explain the tool you want in plain language; it writes the software, shows you the result, and changes it as you react. The same skill you have been practising all day.

What would you build?

Do not think about the technology. Name the small tool or dashboard you keep wishing existed, the one nobody will ever build for you. Then start it with a line like this.

Cowork starter - Open Cowork, paste this, fill in the brackets, and let it build a first version.I want to build a simple [tool or dashboard] that helps me [the job to be done].

It should show [the few things that matter most] and let me [the one action I take with it].

Start simple. Build a first version I can look at, then we will refine it together.

One word, three meanings. Ask which one.

Underneath, every agent is the same five parts.

Strip away the branding and an agent is a junior employee you have hired. You give it a job, some knowledge, some skills, and keys to certain systems, and you decide how much rope it gets. Manage it like one.

Agents you may already own: Microsoft Copilot 365.

With Copilot Studio you build agents grounded in your own company data, living in Teams and SharePoint, and publish them to a shared catalog for the org.

And they are showing up everywhere.

Take an agent that lives in Slack, right where your team already works. It is the same five parts you just saw.

Real momentum, real limits.

Five questions before you trust an agent.

Whether a vendor is selling it or your team is building it, the same five questions tell you whether it is ready for real work.

• 1
  What can it access?

  Which systems and which data, exactly.

• 2
  What can it do without asking?

  Where is the line it cannot cross on its own.

• 3
  Who owns it?

  A named person responsible for its care and feeding.

• 4
  What happens when it is wrong?

  Because it will be. What is the catch and the fix.

• 5
  Can we audit it?

  Is every action logged so you can see what it did.

Canada set a number. We are nowhere near it.

In June 2026, the federal "AI for All" strategy put adoption on the record. Today, just over one in ten Canadian businesses use AI to do real work. The target is six in ten by 2034.

The gap is people, not technology.

The prize is on the table right now.

The same strategy that named the gap also sized the upside. And adoption is already moving faster than most people realize.

It is a change program, not a software rollout.

Most leaders start with "which tool?" That is the wrong first question. Look at what actually happens when companies adopt AI:

Five stages from pilot to payoff.

You do not have to leap to the end. Each stage earns the next, and each one compounds the value of the last.

First, name an owner.

Adopting AI is a change program, not a software purchase. Like any reorg, it needs a single accountable owner, not a committee and not "everyone."

Set the guardrails: one AI acceptable-use policy.

Know your governance, privacy, and regulatory requirements, then write them down. A one-page policy beats a six-month committee.

Decide what business outcome you are actually buying.

Not "use AI." A number you want to move. The common three:

Then choose the tool, in that order.

ChatGPT, Claude, Copilot, Gemini are all capable. Pick on fit, not on the demo you saw last. Four questions decide it:

Build fluency: train on real work, leaders first.

Generic training fades. Train people on their own workflows, put the exec team through first, and name a champion on each team.

Automate the painful, frequent, easy-to-check work.

Three filters: it hurts, it happens often, and you can quickly tell if the output is right. Good first candidates show up in almost every business:

Every automation needs an owner.

A workflow, an SOP, even a saved prompt is now a small system. It needs care and feeding, or it quietly breaks.

From a laptop to the org: four ways to share a build.

Someone builds "a little app that runs on my computer." Getting it to colleagues safely is the real work. The watch-out: by default, "publish" means public on the internet.

Make it on-brand, and make it safe.

"It works" and "it's safe" are different claims, and the gap between them is exactly where AI-built apps fail.

IT's job: a paved road, not a roadblock.

Shadow IT exists because the official path is too slow. 69% of security leaders already suspect staff use banned AI tools (Lansweeper). Don't ban it, pave it. Match scrutiny to blast radius:

Make wins findable. Pull beats push.

Scaling is not more pilots. It is making the wins repeatable and easy to find, so other teams ask for them instead of being told.

Cross the trust threshold: let agents earn it.

You met agents in Module 6. The final stage is letting them do real, multi-step work, and the rule is the same as with any new hire: trust is earned, in stages.

Two threads run through all five stages.

Get these two right and the stages take care of themselves. Skip them and even good tools fade within a month.

One registry. An owner per row, or it doesn't exist.

The same rule covers a built app, a workflow automation, a saved prompt, or an AI SOP. One row each in a shared registry:

Your first 90 days.

Three moves. Write yours down before you leave the room.

• 01
  Days 1-30: set the foundation

  Decide what data can and cannot go in. Pick one business outcome to move, and baseline it.

• 02
  Days 31-60: build fluency, top first

  Get your leadership team genuinely using it on real work. You set the tone the org follows.

• 03
  Days 61-90: automate one thing

  Find a painful, frequent, checkable workflow. Wire AI in. Measure before and after.

Five commitments. Pick one to start.

• 01
  Set your custom instructions today

  Five minutes. Pays back on every chat from now on.

• 02
  Use Role, Context, Task on your next real prompt

  The one you'd have typed half-baked. Slow down. Use the shape.

• 03
  Devil's-advocate one decision this week

  Before you commit, make the AI argue the other side.

• 04
  Connect one tool to your AI

  Start safe: your own email or calendar. Ask it what needs you today.

• 05
  Build one small thing in Cowork

  The little tool nobody else will ever build for you. Start rough.

Work with us.

Three ways to keep going after today.