A Half-Day AI Workshop Agenda for Executive Teams (Remote-Friendly)
A good half-day AI workshop for executives is four hours of hands-on work on the tools the company already pays for, inside a clear set of data rules, with a named follow-up owner. Lectures on "the future of AI" are the part to cut. Below is the agenda I use as a starting point, with pre-work, the data-safety slide, and what to measure after 30 days.
What should a half-day executive AI workshop achieve?
It should leave each leader able to do three things alone: complete one real task in ChatGPT Business or Claude Team, state which data may and may not go into it, and name one workflow they will change in the next 30 days. If a session ends with ideas but no changed workflow, it was a talk, not a workshop.
I design for these outcomes:
- Every participant has a working, configured account before the session.
- Every participant has produced at least one real work output during the session (a board paper summary, a customer reply, a meeting prep brief).
- The team has agreed written data rules, short enough to fit on one page.
- One person owns follow-up, with dates.
Keep the group to the CEO and direct reports, ideally under ten people.
What pre-work should participants do?
Send pre-work five working days ahead and keep it under 45 minutes per person. The goal is that no workshop time is spent on logins, and that the examples come from the executives' real week, not from slides.
The pre-work I ask for:
- Confirm access. Each person logs into the company's ChatGPT Business or Claude Team workspace and sends a screenshot to the organiser. Admins fix seat or SSO problems now.
- Bring one real task. A document, email thread or spreadsheet they would normally spend 30-60 minutes on, with confidential details removed or approved for use.
- Answer three questions. What do you already use AI for? What worries you most? What would you most like to stop doing manually?
- Read the one-page data rules draft. The organiser sends a starting draft (see below) for them to mark up.
- Check the room setup. For remote sessions: camera on, two screens if possible, one for the video call and one for the tool.
Collect the answers before the session. They shape the examples and show who is a beginner and who is already experienced.
What is the hour-by-hour agenda?
Here is a four-hour version with two short breaks. For a three-hour version, shorten block 4 and the wrap-up. Times are relative so you can place them in any time zone.
| Time | Block | What happens |
|---|---|---|
| 0:00-0:20 | 1. Frame and baseline | Goals, ground rules, results of the pre-work survey, what the tool can and cannot do |
| 0:20-0:50 | 2. Data safety and tool setup | Walk through the plan's data handling, agree the rules, set workspace options |
| 0:50-1:30 | 3. Core skills, live | Prompting with context, asking for structure, iterating, checking output |
| 1:30-1:40 | Break | |
| 1:40-2:30 | 4. Your own task | Each leader works on their pre-work task, with coaching and a short share-back |
| 2:30-3:10 | 5. Reusable assets | Turn good prompts into saved projects, custom GPTs or Claude Projects the team can share |
| 3:10-3:20 | Break | |
| 3:20-3:50 | 6. Where it fails | Hallucination, confident errors, bias, what must always get human review |
| 3:50-4:00 | 7. Commitments | One workflow each, one owner, 30-day check-in date |
A few notes on running it:
- Block 3 works best as a live build. I take a real kind of task, such as a messy meeting transcript, and show a weak prompt, a better prompt, and how to check the result. Then they repeat it.
- Block 5 is where value compounds. A leader who saves a "weekly board update" project once saves it for the whole team.
- Block 6 matters more than people expect. Executives who have seen the tool fail in front of them trust it more appropriately.
What data-safety rules should the team agree on?
Agree the rules in the room, but come with facts about your plan. Vendor claims differ between consumer and business plans, so say which plan you are on. The two vendors document their business-plan commitments publicly, and I link them in the sources below so you can quote them in your policy.
What the vendors say, as of my reading in October 2026:
| Topic | OpenAI (ChatGPT Business) | Anthropic (Claude for Work / commercial) |
|---|---|---|
| Training on your data | "We do not train our models on your data by default"; opt-in feedback may be used | "By default, we will not use your inputs or outputs from our commercial products" to train models; reported feedback may be used |
| Ownership | "You own your inputs and outputs (where allowed by law)" | Customer retains rights to inputs and owns outputs under the Commercial Terms; Anthropic "may not train models on Customer Content from Services" |
| Admin controls | SAML SSO and access controls per OpenAI's page | Owners can turn off the thumbs up/down feedback in organisation settings |
| Security statements | SOC 2 audit, AES-256 at rest, TLS 1.2+ in transit | See Anthropic's Trust Center, linked from its privacy article |
Two cautions. First, these are vendor statements, not guarantees about your own handling. Second, the protection applies to the business plan, so staff using a personal free account with company data is outside it. Make that a rule.
A one-page rule set I would start from:
- Use only company-provided accounts for work content. No personal accounts.
- Never enter passwords, API keys, payment card data, health records or anything covered by a client NDA unless legal has approved it.
- Personal data about customers or staff needs a lawful reason and minimum necessary detail. Ask your data protection lead first.
- Turn off feedback sharing where admins can, and remember that if someone clicks thumbs up or down, the conversation can be used by the vendor.
- Human review before anything leaves the company: contracts, numbers, legal or medical statements, anything with your name on it.
- Say when AI was used where a client, regulator or contract expects it.
- Report problems. Wrong output that nearly shipped is useful information, not a confession.
If you sell into the EU, keep the attendance list and materials: an executive workshop can be a first entry in your AI literacy records under Article 4 of the EU AI Act.
How do you run it remotely?
Remote works if you design for it. Use one video call for the room and have everyone share their screen in the hands-on blocks. Keep the group to a size where the facilitator can watch a few screens quickly, and use breakout pairs for the "your own task" block.
What changes versus in person:
- Shorter blocks. Cap any talking segment at 10 minutes. Then give a task.
- A second facilitator or chat monitor if the group exceeds six. Someone has to answer "it doesn't work for me" without stopping the room.
- Shared written notes. One shared document where each leader pastes their best prompt and the output. It becomes the first team prompt library.
- Time zones. From Cairo, I overlap a full UK working day. US East Coast mornings work as Cairo afternoon or evening, and Sydney or Melbourne afternoons are Cairo early morning. US West Coast means an early start for the attendees.
What follow-up should happen in the 30 days after?
Follow-up decides whether the workshop pays off. Plan three touchpoints before you run it.
- Day 2: Send the one-page data rules in final form, the shared prompt library and a short recap of commitments.
- Day 14: A 30-minute check-in call. Each leader shows one workflow they changed or explains what blocked them.
- Day 30: A 60-minute review. Look at usage numbers, share what worked, and decide what to roll out to the next layer of managers.
Name one internal owner, ideally someone close to operations rather than IT only. Without an owner, follow-up slips by week three.
What should you measure after 30 days?
Measure a few things you can collect without a big project: who is using the tools, what they use them for, and whether leaders think it saved them time. I do not promise percentage gains, because they depend heavily on the team and are easy to inflate. Use your own baseline.
| Measure | How to collect it | Why it matters |
|---|---|---|
| Active leaders | Admin console usage or seat reports | Shows whether use persisted past week one |
| Workflows changed | One line per leader at the day-30 review | The real outcome of the workshop |
| Self-reported time saved | Short survey, same question before and after | Rough, but your own comparison |
| Reusable assets created | Count of shared projects, GPTs or prompts | Predicts spread beyond the leadership team |
| Rule breaches or near misses | Log from the report channel | Tells you if the data rules are working |
| Next-layer requests | Managers asking for training | A sign of demand |
Treat self-reported figures as indicators, not proof.
Where Karem fits
I'm Karem, an independent AI consultant and corporate AI trainer in Cairo. I run half-day ChatGPT Business and Claude Team workshops for CEOs and leadership teams, built around the tools the company already pays for. Participants work on their own tasks during the session, so they leave with something real built. Details for UK, US, Canadian and Australian teams, including time zones and pricing, are on the remote AI training page.
FAQ
How long should an executive AI workshop be? Three to four hours covers setup, data rules, hands-on practice and commitments. Shorter sessions tend to become demos. If the group is new to AI, a second short session two weeks later works better than a longer first day.
Should we pick ChatGPT or Claude before the workshop? Use whichever the company already has. The skills transfer between them. If you have both, pick one for depth and use the other for comparison.
Can executives paste confidential documents into ChatGPT Business or Claude Team? Both vendors state that business-plan data is not used for training by default, but that does not replace your own policy, client contracts or legal review. Define what is allowed in your data rules and have legal approve edge cases.
Do we need technical staff in the room? No. It helps to have one IT or security person for the data-safety block, to confirm settings such as SSO and feedback options.
What if some leaders are sceptical? Good. Block 6 on failures is for them. Ask sceptics to bring the task they think AI will get wrong, then test it together.
Sources
- OpenAI, Enterprise privacy at OpenAI: https://openai.com/enterprise-privacy/
- Anthropic Privacy Center, Is my data used for model training?: https://privacy.anthropic.com/en/articles/7996868-is-my-data-used-for-model-training
- Anthropic, Commercial Terms of Service: https://www.anthropic.com/legal/commercial-terms