Build Your Own
AI-Powered GPTs

Reflection from last time (2 min): Quick pulse check. "What stuck with you from Session 1? What did you actually use?" Go around the room or take 3-4 volunteers. This reconnects them to the material and sets a tone of application, not theory.

Show and tell (2 min): "Has anyone built anything, tried anything, or found a use case they're excited about?" If someone shares, celebrate it. If nobody volunteers, have a backup: "I spoke to [name] before the session and they mentioned they'd been using ChatGPT for [X]." Pre-seed this with 1-2 people before the day.

After the keynote chatter (1 min): If there's been internal buzz, hallway conversation, or Slack threads about AI since Session 1, acknowledge it. "I know there's been some conversation about [topic]. We'll touch on that today." This shows you're listening and creates psychological safety.

Keep it to 3-4 things that matter for business leaders, not a full industry roundup. Think: new model releases, major enterprise features shipped, Australian regulatory updates, or anything that changes what they can do today vs. last session.

Make it punchy. "Here are 3 things that changed since we were last in a room together." No slides needed. Just talk through it with energy.

Frame anything relevant to what they're about to build. If a new capability landed that affects GPTs, connectors, or agents, connect it to the session ahead: "This is exactly why we're building what we're building today."

This is the "wow" moment. Hannah shows a pre-built GPT she actually used to prepare for today's session. This makes it real, personal, and immediately credible.

Walk them through what the GPT does: "I built a GPT connected to Outlook and Calendar that helps me prepare for meetings. Here's what it did for me this morning."

Live demo hits:

1. Calendar pull: "Show me my meetings for today and give me a prep brief for each one." Show it pulling real calendar data and generating concise briefings.

2. Email context: "Summarise my recent emails with [name] and flag anything outstanding." Show it searching Outlook and pulling relevant threads.

3. SharePoint retrieval: "Find [relevant document] in SharePoint and summarise the key points." Show it reaching into their document library.

Then peel back the curtain: "Here's how I built this. It took me about 15 minutes. By the end of today, you'll each have one of your own."

Briefly show the GPT's instructions and knowledge files. Don't explain everything yet. Just enough to make every executive think: "I want one of those." The detailed walkthrough comes next.

Facilitation note: The power of this is that it's Hannah's real GPT, not a demo toy. If she can show it pulling her actual calendar and emails (even if pre-tested), the credibility is 10x higher than a hypothetical example.

Screenshare chatgpt.com/gpts/editor. Walk through the Configure tab, not the Create conversational tab (too slow for workshops).

Cover each field briskly:

Name & description — keep it clear, not clever

Instructions — this is where the magic happens. Show a strong example with role definition, output format, constraints, and tone. Emphasise: "The quality of your instructions = the quality of your GPT."

Knowledge files — upload templates, policies, brand guides (up to 20 files per GPT)

Capabilities — toggle web search, Canvas, code interpreter on/off

Apps — tick the checkbox, select Outlook/Calendar/SharePoint. Explain this gives the GPT access to M365 data. Note: Apps and Actions are mutually exclusive.

Conversation starters — pre-set prompts that model ideal usage

Have everyone open the builder on their own laptop as you go. If WiFi is slow, have them pair up on one device now.

Pre-assign pairs if possible (one more tech-comfortable with one less). If not, let people self-select but nudge mixed-ability pairings.

Present the GPT Build Menu (see section below). Each pair picks one. If a pair is stuck, recommend the Meeting Prep GPT — it's the fastest to build and most universally useful.

Assign explicit roles:

Navigator: Thinks strategically, drafts the instructions on paper/notes, thinks about what knowledge files to upload, quality-checks the output.

Driver: Operates the GPT Builder, types instructions, uploads files, tests prompts.

They'll switch roles at the halfway mark.

Hand out the one-page build brief for each GPT option (or point them to the relevant card in this guide).

This is where the real learning happens. Your job as facilitator:

First 5 minutes: Make sure every pair has the builder open and knows which GPT they're building. Unstick anyone who's frozen.

Minutes 5-20: Roam. Check instructions quality. The most common issue is vague instructions. Push pairs to be specific: "Don't just say 'write in a professional tone.' Say 'write in a warm, confident voice. Use Australian English. Keep paragraphs under 3 sentences.'"

Minute 20: Call out "Switch roles!" Navigators become Drivers and vice versa.

Minutes 20-35: Pairs should be testing by now. Encourage them to try edge cases and weird prompts. "What happens if you give it a really vague brief? How does it handle a complaint?"

Last 5 minutes: Ask pairs to refine their conversation starters and do one final test they're proud of.

Energy tips: Celebrate wins out loud ("Table 4 just generated a full event run sheet in 10 seconds!"). Play background music. Keep the vibe workshop, not classroom.

Safety net: Have a pre-built "clone-ready" GPT for any pair that gets stuck. They can clone it and customise rather than building from scratch.

Select 3-4 pairs that built different GPTs for variety. Each pair gets 3-4 minutes:

30 sec: What does your GPT do and why did you pick this?

1.5 min: Live demo with a real prompt. Screenshare for the room.

1 min: Audience throws test prompts at it. Encourage people to try to break it or push it in unexpected directions.

After the demos, briefly highlight patterns: "Three pairs built operations tools, two focused on communications. That tells us where the biggest pain points are, and where we should invest in building more GPTs as a team."

Canvas demo (5 min): Open Canvas, draft a short board update or event proposal. Show inline editing (highlight a section, ask for revision), suggest-edits mode (like track changes), and reading level adjustment. Export to Word. Key message: "Canvas turns ChatGPT from a Q&A tool into a document co-author." Type "use canvas" in any chat to activate it.

Vibe Coding (5 min): This is the one that blows minds. Show how ChatGPT can build a working app, tool, or prototype just from a description in plain English. Live demo: "Build me a simple event capacity calculator where I enter the venue name, room, and event type and it tells me max capacity, recommended layout, and staffing levels." Watch it generate a working interactive tool in real time. Key message: "You don't need to be a developer. You describe what you want, AI builds it. This is how non-technical leaders prototype and test ideas."

Codex (5 min): Briefly introduce OpenAI's Codex agent. This is a dedicated coding agent that can work on entire codebases, not just one-off scripts. It reads your code, understands the full context, writes changes across multiple files, runs tests, and commits. For this audience, frame it as: "You won't use this directly, but your tech teams will. What matters for you is understanding that AI isn't just helping people write code faster. It's writing code autonomously. That changes the economics of every build-vs-buy decision you'll make."

Facilitation note: The vibe coding demo is the "wow" moment of this block. Pick something relevant to their business. If the live build stalls, have a pre-built example ready to show. The Codex piece is a 60-second framing, not a deep dive. Plant the seed that this exists and move on.

This is the "so what" moment. You've built GPTs. You've seen what's possible. Now the question every leader in this room should be asking: how does this go beyond me?

The pattern that works: Leaders use it first. Then they identify 2-3 champions in their team who are curious and willing to test. Champions find what works in the real workflow, not in a training room. Then you scale the wins, not the tools.

Three things to decide as a leadership team:

1. Who gets access next? You don't need to roll out to everyone at once. Pick the teams where the GPTs you built today would have the most impact. Event ops, marketing, finance reporting, guest experience are natural starting points for a venue business.

2. What's the governance layer? Who approves new GPTs before they're shared workspace-wide? What data is OK to upload as knowledge files and what isn't? You need a light framework, not a 40-page policy. One page is enough.

3. How do you measure it? Time saved is the obvious one, but the real signal is adoption. Track how many people are actually using GPTs weekly, not just how many were created. A GPT nobody uses is just a good intention.

Common traps: Don't wait for perfect governance before letting people experiment. Don't build 20 GPTs when 3 good ones would do more. Don't assume the IT team owns this alone. The best rollouts have a business sponsor (someone in this room) and a tech enabler working together.

Facilitation note: This should feel like a brief, energising framing, not a deep strategy session. Plant the seed. The Reimagine Workshop in a few weeks is where they'll do the full strategic planning on this.

Everyone writes down two things:

This week: One GPT to refine and actually use in your workflow.

This month: One new GPT to build for your function or team.

Assign GPT Buddies: pair people up (different from their build partner). Buddies check in at week 1 and week 4. This simple accountability structure is what makes the difference between "fun workshop" and "actual behaviour change."

Announce a 30-day virtual check-in (30 min, calendar invite goes out today) where each person shares their favourite GPT iteration.

Optional: nominate a GPT Champion who curates the team's shared GPT library and sends a "GPT of the Week" highlight.