From one programme I ran inside a Fortune 100 company. Client names stay private unless they tell me otherwise.
Who runs this
I'm Arvindh Sundar. I'm a facilitator, and I run my own business on these tools every single day. I'm not an engineer, which I think helps more than it hurts in a room full of non-technical people.
I've run six session-days of one programme inside a Fortune 100 company, and it came back at 8.82 out of 10 across 113 respondents. I've also run AI training for marketing teams at the same company.
I design the games I run, so when a table pushes back at 2pm I can usually rebuild the exercise on the spot. And when somebody says "yes, but that would not work for what I do", I can often show them, live, that it would. That's my favourite part of the day.
What a quiet AI rollout tends to look like
None of these is anybody's fault, really. They just don't show up on a licence dashboard as anything except a number going sideways.
- Licences bought, a policy circulated, a launch email sent, and usage flat after month one.
- Two or three enthusiasts using it brilliantly, and everyone else quietly not.
- People pasting things into tools you didn't approve, because nobody showed them where the line is in a way that stuck.
- A report goes out that AI wrote and nobody checked, and it's wrong in a way that's now a bit embarrassing.
- A training session that everyone attended, and that nobody used on their own work afterwards.
- Senior people who'd rather not say out loud that they're unsure how to use it, so nobody helps them.
Three sizes. Pick by what feels most stuck.
Same spine every time. Hands on, their own real work, no code, no jargon. What changes is how far we go.
And in every size, the final call stays with the person. AI drafts, sorts and checks. Your people decide what goes out with their name on it.
They start using it
For a team that has access and isn't using it. Short enough that you can actually release people, and it runs in batches across a large group.
- The difference between a question and a brief, practised on their own task
- A repeatable prompting method they can still remember in three weeks
- A serious game in the middle, with no AI at the table: The Potter from Belagavi, where you can spend your coins on more pots by hand now, or on a wheel that pays nothing the round you buy it and makes pots for you every round after
- The tells of a confident wrong answer, and where your policy line sits in their real terms
- Each person rebuilds one repeat task before they leave
They start trusting it
For a team that uses it but can't always tell good output from fluent nonsense. Everything in the half day, with room for the part three hours can't reach.
- Everything in the half day, unhurried
- Thinking in systems rather than single prompts - why most people's setup starts from scratch every Monday
- What these models actually are, in plain English, and why that explains the ways they go wrong
- A second, harder task rebuilt with feedback in the room
- Where a decision must not be handed over, worked through on their own live examples
They change how they decide
For when the tool was never really the problem. Your people can usually spot what's wrong. What's hard is getting anyone to act on it. This is that, with AI doing a lot of the heavy lifting on the real work.
- Separating the thing that's annoying them from the thing causing it, with AI working through the evidence
- Reading a messy, multi-source pile the way the real world hands it over
- Structured divergence before convergence, so a room of eight produces eight angles
- Mapping who actually has to say yes, and building the case in their language
- The impact case in the language finance uses - the step between "good idea" and a budget line
- 30/60/90 commitments, written and said out loud, with names against them
This one is built on Insight to Impact, my leadership decision-making programme. Its most recent scored day, 6 August 2026, came back at 8.37 out of 10 across 19. I've rebuilt it so AI runs through both days rather than showing up occasionally. The rebuilt version is new - the 8.37 belongs to that one day of Insight to Impact, and I'd rather tell you that than let a number do work it hasn't earned.
What your people can actually do afterwards
Every one of these gets practised on their own work during the session, rather than shown on a slide.
- Turn a vague ask into a brief that gets a usable answer - the one change that fixes most of what people complain about.
- The "make it ask you questions first" move - a handful of words, and what comes back gets noticeably better.
- Hand over a two-hour reading pile and get the decision-ready version back before the coffee's cold.
- Spot a confident wrong answer - the specific tells that an answer is fluent and false, so nobody signs one off.
- Know where the line is - what should never go into these tools, in your policy's terms rather than a generic warning.
- See the whole system, not just their one step - the game does this, and it's the part people quote back to me months later.
Before anyone on your team asks an AI tool anything, have them type: "Ask me 3 questions before you start." That one habit fixes most of the junk answers people complain about, because now it's working from their context instead of guessing. Send it round this afternoon and see what happens. No charge :)
What you get as the sponsor
- Sizes that fit real calendars. Three hours is the length people actually get released for. Two days is what you book when the problem is worth it.
- Batch-by-batch tuning. What batch one found hard changes batch two. That's a lot easier when the same person runs every batch, which is me.
- Things that outlive the room - a prompt library in your own vocabulary, an agreed policy line, written commitments with names on them.
- Feedback data per batch, shared honestly, including the criticisms.
- Built around your approved tools and your real policy limits, including what's explicitly off the table.
- A debrief with me on what the room revealed - where the work is actually getting stuck.
Honestly, who this is and isn't for
Probably a good fit if
- You've given people AI access and usage has stalled.
- Your group is non-technical and mixed in seniority.
- You'd like behaviour change more than attendance numbers.
- You'll let real work into the room.
Probably better served elsewhere if
- Your audience is engineers wanting technical depth.
- You'd prefer a recorded module people can click through on their own time.
- Certification matters more to you right now than daily use.
- Real work can't come into the room for policy reasons.
And if it's really your managers you have in mind, using AI on their own calls and their own week rather than a whole team learning the tool, that's a separate programme called AI for Managers, and it's probably the better fit. You'll find it with the rest on the leadership training programs page.
Tell me what feels stuck
Three things is all I need. Roughly how many people, which AI tools they're allowed to use, and what you've already tried. I'll come back with the size I'd recommend. And if none of the three fit, I'll tell you that instead - it happens, and I'd rather say it early.
Worst case, we have a conversation and you come out of it with a clearer brief.
I only take two or three programmes a month, mostly because I run all of them myself. So it does fill up, and I'd rather tell you that now than when you have a date in mind.
Straight answers
Which size should we pick?
Which AI tool do you teach?
Can you run this for a few hundred people?
Is three hours really enough?
Is the AI inside the game?
In person or online?
We've already had our AI vendor's training. Is this different?
Do people need accounts before the session?
None of this will turn your team into AI experts in a day, to be fair. It just gets them using it on the work that's already on their desk, which is usually where it was meant to go in the first place.
With love from Bengaluru, this is Arvindh saying over and out.