180 People, Six Sessions, One Pattern
In 2025, I delivered six sessions for a Hong Kong statutory productivity body. Different formats each time -- a creative-AI workshop, an executive class, a travel-industry seminar, a Copilot productivity session, a mainland sharing session, and a bank co-hosted training. Different audiences: thirty participants in one room, ten in another, fifty in the largest. A hundred and eighty people across the series.
The mix was varied enough that finding a common pattern should have been unlikely. But one showed up in every session, regardless of format or topic, and it had nothing to do with the technology.
The executives were more anxious about AI than their staff.
Not openly. Nobody in an executive session announces that. It showed up in questions -- careful, hedged, circling the same concern from different angles. Questions about reliability. About risk exposure if someone uses the tools incorrectly. About whether comparable organisations had started already. The framing was strategic: what's our liability? The substance was personal: if the people I manage understand this technology and I don't, what does that mean for my position?
The staff sessions had a different quality. Frontline participants asked practical questions -- whether the tool could draft in their tone, whether it worked with their existing systems, how to stop it generating inaccurate information. The anxiety was there, but it sat at the task level: will this make my day harder before it makes it easier? The executives carried something heavier. Their concern was about identity -- does this technology make me less relevant?
In the ten-person executive class, this was sharpest. Smaller room, more senior people, and the gap between "I should understand this by now" and "I don't, actually" was visible in how carefully everyone approached the tools. The thirty-person staff workshops had more energy -- more willingness to try things, fail visibly, laugh at bad output. The executives were engaged and notably more cautious about making mistakes in front of peers.
What this means for who you train first
Most organisations start AI training with frontline staff. The logic is straightforward: train the people doing the work the tools are supposed to improve, measure the result, brief leadership afterward. It's a clean sequence that misses the bottleneck.
If the executives are more anxious than the staff -- and across six sessions with this client, they consistently were -- training them first does two things. It gives them firsthand experience with the tools before they're asked to make budget and policy decisions about them. And it removes the situation where leaders evaluate an AI initiative they haven't personally used, which tends to produce either excessive caution or misplaced enthusiasm, both more expensive than the training itself.
The training still has to reach the frontline eventually. But a leader who has sat in a room, tried the tools, watched them fail, and come away with a working sense of what they're actually useful for asks better questions when the rollout proposal lands on their desk. The approvals come from experience instead of a vendor's slide deck.
I notice the pattern more clearly now than I did at the time. During the sessions, the executive rooms just felt quieter and the staff rooms felt easier to run. It took six of them -- different topics, different audiences, same dynamic -- to see that quiet and easy were describing the same gap from opposite ends. If I planned the sequence again, I'd start with the executives.
