AI at work

The room

EuroCham Vietnam — the European Chamber of Commerce — invited me and my colleague Manh Nguyen to design and deliver a two-session AI training series for member companies, hosted at the Caravelle in Ho Chi Minh City. Registration was capped at one attendee per company, which made the attendance list a cross-section of European business in Vietnam: pharmaceutical and consumer-goods multinationals, global logistics operators, real-estate and construction firms, law firms, an international NGO, and manufacturers, consultancies, and operators across more than fifteen sectors.

Twenty-seven of the thirty-five registered for the full series before the first session had run, and thirty were in the room on the day — an 86% show rate for a complimentary event, in a market where free registrations routinely convert at barely half that.

The finding

The opening poll asked participants to place their organisations honestly on an adoption scale. Eight said Experimenting. Eight said Piloting. Twelve said Scaling — this was, by self-assessment, the most AI-mature room I had worked with across any sector.

And not one placed their organisation at Embedded.

That single empty column is the corporate AI problem in miniature. These companies are adopting — many are scaling — but none would claim AI has become how the organisation actually works. Individuals are learning fast. The institutions around them aren't. The gap between those two speeds is precisely what the series was built to address.

The design problem

The same poll revealed the harder challenge. The tools data ran from one extreme to the other: participants running n8n automations and local models sat alongside participants using AI to check the grammar in their emails. Copilot led the count — enterprise IT had made its decision — but "agents" was already among the most-cited answers, which means the frontier had arrived in these companies whether or not anyone had signed it off.

A three-hour session for that room cannot teach to the middle, because there is no middle. Session one was built hands-on around participants' own real tasks — each person working on something from their actual job — so that the n8n user and the grammar-checker were both stretched by the same structure rather than bored or lost by the same content.

Responsive by design

The series was published with an unusual commitment: session two's content would be adjusted based on what session one revealed about the room. That is not a courtesy line — it is the same principle that runs through everything on this site. You sense what is actually happening, and then you respond to it, rather than delivering the deck you wrote before you met the people.

Session one produced the data above. Session two — on leading AI adoption in an environment where agents already act on the organisation's behalf — runs on 15 July 2026, redesigned around what the room revealed about itself. After the first session, registration demand led EuroCham to expand the second session's capacity to forty seats. Outcomes will be reported here.

What this demonstrates

The frameworks on this site were built in higher education and have been tested on diplomats, accountants, governing bodies, and police commanders. This series is the corporate test: thirty-five companies, fifteen-plus sectors, maturity ranging from grammar-checking to agent-building, in a room convened by the European business community's own chamber. Same problem underneath. Same gap between individual capability and institutional absorption.