Real work
Use the actual document, email, spreadsheet, meeting note, proposal, or admin task someone already has to handle. If the work is fake, the learning usually is too.
AI sandbox concept
Stop training people on AI. Let them play with it instead.
This is a working idea, not a current team-training offer. I am interested in how people learn AI through real work, clear guardrails, and repeated practice. For now, my public offer is strictly one-on-one coaching.
The gap
Canada has the awareness. The practice is missing.
The concept
A practice rhythm for learning with real work.
This is not a certification program or a giant transformation project. It is a way of thinking about how AI skill actually develops: through safe, repeated use on tasks people recognize.
A clear safety line
Write down what is in scope and what is out before anyone starts. No personal information, client data, private financial details, or regulated material.
Reps, not courses
A single workshop teaches people that AI is something they tried once. Twenty or thirty minutes a week teaches them that it can become part of how they work.
Failure is useful
The weird answer, the hallucinated detail, and the broken prompt are not embarrassing. They are where people learn how verification and judgement actually work.
Use cases from the floor
Good use cases usually come from the people closest to the work. Leaders set the safety and rhythm; staff discover the practical spots where AI helps.
What it looks like
Small enough to start. Concrete enough to matter.
Week 0
Set the safety line
Choose 6-10 curious people, pick the weekly time, and agree on a one-page guide for what can and cannot be tested.
Week 1
Try one thing together
Start with a simple shared task like turning a long thread into action items. Compare what worked, what failed, and what felt risky.
Weeks 2-4
Build weekly reps
Practice on common work: email drafts, summaries, messy spreadsheets, proposal sections, internal templates, or meeting notes.
Week 5
Follow the sideways use cases
People start bringing tasks nobody put on the agenda. That is the point. The useful workflows often show up from the floor.
Weeks 6-8
Debug the habits
Shift from demos to judgement: verification, privacy checks, prompt repair, when to use AI, and when to leave it alone.
End
Keep what stuck
Review which workflows became useful, which did not, and what the next small round should test.
Safety line
The boring part is what makes the playful part possible.
The first deliverable is a plain one-page safety guide. It gives the team permission to experiment without turning every experiment into a privacy or compliance risk.
In scope
- Drafting plain-language emails with names and details removed.
- Summarizing internal notes that do not contain sensitive information.
- Brainstorming options for public-facing or low-risk material.
- Reformatting messy non-sensitive spreadsheets or templates.
- Comparing versions of a document for clarity, tone, or completeness.
Out of scope
- Personal information.
- Client or patient data.
- Private financial details.
- Regulated, privileged, or confidential material.
- Anything that would create harm if copied into the wrong place.
Canada is good at AI research. The weaker part is using AI inside everyday work. The numbers are uncomfortable: adoption is high, measurable return is low, and employees still say they need more practical skill.
The disconnect tells us something important. Leaders often think they have provided AI training. Workers often experience that training as too general, too abstract, or too far away from the actual work on their desk.
A webinar can explain that AI exists. A vendor demo can show what a tool does under perfect conditions. A policy document can name the risks. None of that, by itself, teaches someone what to do when the tool confidently invents three facts, misses the obvious point, or makes a decent first draft that still needs human judgement.
A learning sandbox is different. It is a small group, a regular rhythm, a clear safety line, and permission to try real work in front of other people. The point is not to make everyone an AI expert. The point is to help people build the judgement to know when AI is useful, when it is risky, and how to check the work.
Imagine a 25-person accounting firm in Winnipeg. Mostly bookkeepers and accountants, a few admin staff, two partners. Nobody is the official tech person. People have heard about ChatGPT, a few have tried it, and almost nobody is using it at work because nobody has made the safety line clear.
Week zero is simple. One partner picks eight curious people and writes the one-page rule: we are trying AI on real work for eight weeks. Internal templates and anonymized email drafts are in scope. Personal client data is not. The group meets every Tuesday for 30 minutes.
In week one, they summarize a long email thread into action items. A few outputs are useful. A few are garbage. A few are weird in the middle. Everyone sees, right away, that AI is neither magic nor useless. That lesson lands because people watched it happen on familiar work.
By week five, the interesting stuff starts showing up sideways. A senior bookkeeper finds a way to turn last quarter's notes into a cleaner report structure. Someone else uses AI to rewrite an email she had been avoiding. Another person discovers the tool is bad at a task everyone assumed it would handle easily. Those discoveries are better than a curriculum because they come from the work itself.
By the end of eight weeks, the firm has not transformed into a science-fiction company. That was never the point. Three or four workflows have stuck. People are more comfortable asking better questions. They are also more alert to mistakes. The team has built a little shared language around what AI can and cannot do.
That is what I think many organizations are missing. Not a bigger AI strategy. Not more hype. Not another training portal. Just a protected space where people can try, fail, compare notes, and get better at the actual judgement calls AI requires.
Try it next week
Bring one real task. Learn the judgement by doing it.
If you want help applying this kind of practical learning to your own work, start with a focused 1:1 coaching session.
Ask about 1:1 coaching