Most AI workshops are enjoyed and then forgotten. Here's how to plan one around real work, a half-day agenda you can use, and the follow-up that makes it stick.
TL;DR
Plenty of AI workshops get enjoyed and then forgotten. The room is lively, the demos land, everyone leaves with a cheat sheet, and three weeks later the work looks exactly the same.
The difference between that and a workshop that changes things is mostly in the design, before anyone walks into the room. This is how I'd plan and run one for a team, whether you facilitate it yourself or bring someone in.
A good AI workshop does three things. It gives people enough understanding of the tools to use them sensibly. It gives them practice on their own work, so the first real use happens in the room rather than alone at their desk. And it ends with a specific commitment about what changes next.
What it can't do is change incentives, redesign a process, or make a manager model new behaviour. Those sit outside the room. If you expect a workshop to do that work, it'll disappoint you, which is why I treat it as one part of a wider plan for building AI literacy and adoption.
Decide the outcome. Write one sentence about what people should be able to do by the end. Something like: "Everyone uses our approved AI tool on two tasks from their own role, and knows how to check the output." If you can't write that sentence, you're not ready to book the room.
Find a sponsor and champions. Atlassian's AI training workshop play recommends a senior sponsor plus champions in each department, and makes a point I agree with: choose enthusiastic users rather than tool administrators, because admins tend to have the curse of knowledge and struggle to remember what beginners find hard.
Collect real tasks. Ask each participant, a week before, for two tasks they do repeatedly that involve reading, writing, summarising, or analysing. A monthly report, a client email they dread, a policy they have to read and apply. These become your exercise material.
Settle the rules. Confirm which tools are approved and what data can go into them. If you don't have a policy yet, write a one-page interim version. People won't experiment if they're unsure what they're allowed to do.
Offer short pre-work. Free introductions like Microsoft's Work smarter with AI learning path or Anthropic's AI Fluency course mean fewer people arrive cold. Keep it optional, and design the session so nobody falls behind without it.
Atlassian's play runs in an hour, which works for a team already using a tool. For a group starting close to zero, I prefer half a day. Here's the structure I'd use.
| Time | Block | What happens |
|---|---|---|
| 0:00 | Opening | The sponsor explains why this matters and shares how they use AI themselves |
| 0:10 | Baseline | A two-minute survey on confidence and current use |
| 0:15 | How it works and fails | A plain explanation of the tools, plus live examples of confident errors |
| 0:45 | Use case one | "Show, then do" on a common task, like summarising a long document |
| 1:30 | Break | Fifteen minutes |
| 1:45 | Use case two | Pairs work on the tasks they submitted, with facilitators circulating |
| 2:30 | Judgement drill | Groups critique AI outputs and decide what to keep, check, or bin |
| 3:00 | Commitments | Each person names one task they'll do differently next week |
| 3:20 | Close | Closing survey, questions, and the sponsor's close |
The opening matters more than it looks. Atlassian suggests the sponsor shares a personal example of using AI, which gives people permission to try it. When a leader says "this is how I drafted last week's board update", the room relaxes.
The "show, then do" rhythm is the core. The facilitator demonstrates a task, then everyone does it immediately on their own material. Watching a demo creates the illusion of understanding, while doing it yourself reveals the gaps.
Most workshops spend almost all their time on getting output. I'd spend a fair share on judging it. Harvard Business Review argues that as AI makes polished work easier to produce, judgement becomes the scarce skill, and yet organisations mostly train people to use the tools rather than to think critically with them.
The judgement drill in the agenda is my answer to that. Give small groups three AI outputs on a task from their function. One is good, one is plausible but contains a factual error, and one gets the tone badly wrong. Ask them to decide what they'd keep, what they'd check, and what they'd throw away, then explain why. It's the most useful 30 minutes in the session.
Group people by function. An HR team and a finance team need different examples. If you have to mix them, use breakout pairs from the same function for the hands-on blocks.
Keep groups small enough to help. I'd plan for one facilitator or champion per ten people in the hands-on blocks, so nobody sits stuck for long.
Plan for the tech to fail. Record your demos in advance so a slow connection doesn't derail the session. Atlassian makes the same suggestion, and it saves a lot of awkward minutes.
Welcome the sceptics. The person who says "I tried it and it was wrong" is useful, because they've already learned the most important lesson. Ask them to share the example.
This is where workshops tend to lose their value. Book two check-ins before the session ends: one at two weeks and one at four. In each, ask every person which task they changed, what worked, and what got in the way.
The research supports more than a single session. BCG's AI at Work survey found that regular use rises among employees who get at least five hours of training and access to in-person coaching. A half-day workshop gets you close to that, and the follow-up sessions carry you past it.
Share what people discover. A short internal channel where people post one useful prompt or workflow a week spreads practice faster than any second workshop. Without follow-through, a workshop tends to join the pattern I describe in why AI pilots fail.
Run the same short survey at the start and end, as Atlassian suggests, to capture the change in confidence. It's a useful signal, but it measures how people feel on the day.
The measure I care about comes later. At four weeks, how many people are still using AI on the task they committed to? Has anyone found a second use? Can their manager point to work that's now done differently? Those answers tell you whether the workshop changed anything.
A capable internal champion can run this format, especially for a team that already has an approved tool. Bring in outside help when nobody internally has the time, when the team is starting from zero, or when leadership needs its own session on the decisions only they can make. Our guide to corporate AI workshops in South Africa covers what to look for in a provider and what it typically costs.
An hour works for a team already using an approved tool. For beginners, half a day gives enough time for hands-on practice and a judgement drill, with follow-up sessions afterwards.
How the tools work and where they fail, hands-on practice on participants' real tasks, how to check output, the organisation's rules for AI use, and a specific commitment from each person about what they'll do differently.
Any number can attend, but plan for roughly one facilitator or champion per ten people during hands-on practice, so nobody gets stuck.
Both can work. In person makes hands-on support easier, and BCG's research links regular use to in-person training and coaching, so I'd choose in person for a team's first session where possible.
Someone who uses AI regularly and remembers what it was like to start. Atlassian recommends enthusiastic users over tool administrators for exactly that reason.
If you'd like a workshop built around your team's real work, with a plan for the weeks afterwards, that's how we run ours. Our free AI Readiness Diagnostic is a good place to start if you want to know where your team stands first.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic