AI Foundations: fluency and confidence with AI, for everyone
The format
- 6 weekshalf a day a week: two hours live, the rest on their own work
- Claude or Copilotmatched to the tool each function already uses
- Before and aftera confidence survey and a skills assessment, at the start and in Week 6
In person at your office, online or hybrid.
4.8 / 5 from 78 Google reviews
1,000+ learners trained
Trusted by teams at
Who it is for
Everyone, in every function
AI Foundations is not just for the technical team. No coding is required, and nobody needs to be technical to start. It suits people in any role.
The format
- Length
- Six weeks.
- Time
- Half a day a week: a two-hour live session, about an hour of preparation, and about an hour applying it to their own work.
- Where
- In person at your office, online or hybrid, as you prefer.
The tools
People learn on the AI tool their team already uses: Claude Enterprise or Microsoft 365 Copilot, matched by function.
The skills are not tied to one tool, so people on Claude and on Copilot learn side by side in the same cohort.
The curriculum
Six weeks, week by week
Each week pairs a hands-on activity with the skill it teaches, and adds one saved file to each person's AI workspace: a Project in Claude, an agent or notebook in Copilot.
Week 1
Find your use cases, work safely
How the model works, and where it fails, before any rules.
- How AI writes by predicting the next word, and why that lets it sound sure when it is wrong.
- Why long documents cost more and hit limits, and why an overloaded chat gives worse answers.
- Why AI forgets everything between chats, and how saved instructions and files in a Claude Project or a Copilot agent fix that.
- Writing background information as clear, structured plain text that AI reads reliably.
- Scoring possible uses of AI, and ruling out first anything your organisation's rules do not allow.
- Keeping personal data safe: removing names and other identifying details before anything is uploaded.
They leave with: A scored shortlist, one chosen task and its brief.
Week 2
Prompt it
People run an interview prompt: the AI asks about their task one question at a time, then writes the prompt for them.
- The AI hands back a brief, a description of what good output looks like, the prompt itself, and a numbered list of the assumptions it made.
- Correcting those assumptions before trusting the prompt.
- The six parts of a good prompt: who you are, what you are giving it, what you want back, the rules, an example, and when it should stop and ask.
- Running the same prompt several times to see how much the answers vary, then writing down the version that works.
- Noticing when a long chat starts to go stale, and starting a fresh one inside the same workspace.
- Trying the prompt on a fresh example, checking the result against what good looks like, and fixing the prompt rather than editing the answer.
- The common ways prompts fail, and the fix for each.
- Six real decisions about what is safe, and allowed, to upload.
They leave with: A brief, a specification and a prompt that runs.
Week 3
Break it
People run their prompt on inputs it has never seen, watch where it gives way, and write the line that fixes it.
- Three set ways to try to make it fail, each with a clear pass or fail.
- Asking it to “list every figure and the row it came from” or to “quote the line you used”, so every answer can be checked against its source.
- Asking “what is in this document I have not asked about?”, so nothing important is missed.
- Keeping a short record of each change at the bottom of the prompt.
- A personal library of prompt pieces to reuse.
They leave with: A tested, versioned prompt and a snippet library.
Week 4
Your prompt becomes a button
The prompt becomes something a colleague can run without its owner in the room.
- Saving the prompt as the standing instructions of a Claude Project or a Copilot agent, so it runs from one short phrase.
- Which instructions win when a person's own settings and the shared workspace disagree.
- The stranger test: a colleague runs it cold, the owner says nothing, and the questions are counted.
- A handover prompt, so whoever picks the work up next can run it.
- Choosing which model to use, and how hard to ask it to think, where the tool offers a choice.
- Sharing a project or agent, plus scheduled tasks and connections to live data where your organisation has turned them on.
They leave with: Shared instructions a colleague can run, and a stranger test passed.
Week 5
A workspace that knows things
The same workspace becomes a team knowledge base that answers questions from its own files.
- Turning a department's shared files into a knowledge hub that answers questions from those files alone.
- Instructions that make the hub quote the file, say “not in the hub” rather than guess, and never give a figure no file contains.
- When two files disagree, trusting the one with the later “last checked” date.
- A list of what the hub cannot yet answer, and a named person who checks it each week.
- A results log: the task timed three times, checking included.
They leave with: Hub instructions and a results log.
Week 6
Show and tell
Each person shows the cohort and sponsors, in five minutes, how they now use AI in their own work, one thing that went wrong and how they fixed it. The skills assessment and confidence survey are repeated.
Afterwards
What people can do by Week 6
Pick the right work for AI.
Score possible uses, rule out first anything your rules do not allow, and remove identifying details before anything is uploaded.
Write prompts that run reliably.
Use the six parts of a good prompt, try it on new examples, and fix the prompt rather than the answer.
Check what comes back.
Ask the AI to show where every figure came from, test it on inputs it has never seen, and spot what it left out.
Give it the right context.
Keep instructions and files in a workspace, not in one-off chats, and start fresh when a chat goes stale.
Share what works.
Hand a colleague instructions they can run without you, and keep a team workspace that answers from its own files.
Know what to keep human.
Tell a confident answer from a correct one, and decide where AI belongs in their own work.
How we teach it
How every session works
Mechanism before rules
People learn how the model works, and where it fails, before any rules.
Plain language first, technical term alongside
Every skill is taught in plain words first. The technical name comes second, for anyone who wants it.
Files, not chats
Each week adds one saved file to each person’s AI workspace: a Project in Claude, an agent or notebook in Copilot.
Testing is part of the work
Every prompt is run on new inputs and checked against its specification.
Nothing real on screen
Demos use fictional material, so no real data appears in a session.
Working rooms by stage
The second hour of each session splits into small groups by where each person is up to, so faster and slower learners both progress.
How it is measured
Confidence and skills, before and after
Before Week 1
An onboarding survey asks each person about their role, where AI could help, and how confident they feel. A skills assessment sets the starting point. We check everyone's AI access before the start.
Weeks 1 to 6
Attendance reporting every week, and weekly feedback on how each session landed and what we changed in response.
Week 6
The skills assessment again, the repeat confidence survey, and an end-of-programme survey.
The report
An end-of-programme report for each cohort: skills, confidence, attendance and feedback in one place.
Recognition and next steps
A certificate, an optional qualification, and a route further
For everyone who completes
Certificate of completion
Each person who completes the programme receives an iO-Sphere certificate of completion.
Optional
A regulated Level 3 qualification
People can also take The AI Board's Level 3 Award in AI: Concepts, Ethics and Applications, an Ofqual-regulated qualification (610/6527/4) at RQF Level 3. It is assessed by a portfolio that draws on the work people do in the programme.
Going further
AI Champions
For the few people who want to take AI further on a real team problem: six more weeks, at about eight hours a week, with their manager's backing.
AI Champions →Training several groups at different depths? The whole-organisation programme puts them in one plan.
Questions
AI Foundations FAQs
Do people need to be technical?
+
No. No coding is required, and people do not need to be technical to start. The programme is for every function, not just the technical team.
Which AI tool do you teach?
+
The one each person's team already uses: Claude Enterprise or Microsoft 365 Copilot. The skills are not tied to one tool, so people on Claude and on Copilot learn side by side in the same cohort.
How much time does it take each week?
+
Half a day a week for six weeks: a two-hour live session, about an hour of preparation, and about an hour applying it to their own work.
Is it in person or online?
+
Any of the three. We run it in person at your office, online, or as a hybrid, whichever your team prefers.
How will we know it worked?
+
Each person takes a confidence survey and a skills assessment before Week 1, and the same again in Week 6. You get an end-of-programme report for each cohort, with skills, confidence, attendance and feedback in one place.
Talk to us about AI Foundations
Tell us about your teams and the tools they use. We will come back within one working day.
Start with one team, or with everyone
Tell us how many people and which AI tools they use, and we will set out how a cohort would run.