The tool is not the transformation.

Adoption is broad. Impact remains limited.

In 2026, a research team working with the National Bureau of Economic Research surveyed almost 6,000 senior executives in the United States, the United Kingdom, Germany and Australia.¹ More than two-thirds said they regularly used AI. Regular use averaged about one and a half hours per week. And 89 % reported that AI had had no impact on labour productivity in their firm over the previous three years — even as the same executives expected stronger effects over the following three.

Read carefully, the study does not describe failure. It describes shallowness: adoption that is widespread and thin at the same time. The tools are present; the work is largely unchanged. The interesting question for a leadership team is therefore not whether to adopt AI — that has mostly happened — but why adoption so rarely deepens into changed work.

A tool can be used without changing the work.

There is a difference between using AI to complete an isolated task and redesigning a recurring workflow. Summarising a document is a convenience; redesigning how reports are produced, reviewed and distributed is an operational change. Generating a piece of sales copy is a time-saving; an approved sales-content workflow, with agreed inputs and checks, is a capability. Asking a chatbot questions is exploration; a governed internal assistant, with defined data and escalation paths, is infrastructure. Automating an output is easy to demonstrate; deciding who remains accountable for it is the actual work.

The first half of each pair saves minutes and proves nothing. The second half changes cost, quality or risk — and it requires exactly what a licence does not include: a chosen workflow, a baseline, an owner, and a decision about where human judgement remains essential.

The manager as translator.

Between strategy and daily work sits the manager, and this is where translation succeeds or quietly stops. In Gallup's 2025 research on employees in organisations already implementing AI, only 28 % strongly agreed that their manager actively supported the team's use of it.²

Support here is not encouragement. It means helping the team select useful applications, protecting time for experimentation, setting boundaries, deciding what good output looks like, creating review processes, making learning discussable, and integrating the use case into actual work. Gallup found strong associations between this kind of support, more frequent use, and employees' assessment that AI helped them perform at their best. The relationship has not been proven exclusively causal — but it is difficult to imagine deep adoption without it.

Before you buy another AI tool.

Five questions separate experimentation from organisational change:

1. Which recurring workflow are we actually trying to change?
2. What baseline will show whether it became faster, better or safer?
3. Who owns the new way of working after the pilot?
4. Where must a human review, challenge or approve the output?
5. What does the team need to understand about data, limitations and responsible use?

Each question is practical, not rhetorical. Workflow: name a recurring process, not a vague ambition. Baseline: know what is currently slow, expensive, inconsistent or risky, or you will never know whether anything improved. Ownership: identify who carries the implementation after the workshop or pilot ends. Human review: decide in advance where output must be challenged, approved or corrected. Literacy: make sure the people using the system understand its capabilities, its limitations, the data implications, and when to escalate.

If these questions cannot yet be answered, the next step is probably not another licence. It is a leadership decision.

AI literacy is role-specific.

In Europe, one of these questions now has a regulatory dimension. Article 4 of the EU AI Act requires providers and deployers of AI systems to take measures to ensure a sufficient level of AI literacy among staff and others using AI systems on their behalf.⁴ The European Commission's guidance emphasises that the appropriate level depends on technical knowledge, experience, education and training, the context of use, and the people affected by the system — a context-specific approach, rather than one universal course for everyone.³

That framing is useful well beyond compliance. AI literacy is not the ability to recite definitions. It is the ability to use a particular system with enough understanding to recognise when confidence, caution, review or escalation is required. This article and the Bon Sentiment programme do not constitute legal advice.

Why many AI workshops disappear after the workshop.

Most AI workshops fail politely. Everyone learns something; nothing changes. The reasons are rarely mysterious: no workflow was selected, so the energy had nowhere to land. No decision owner was named. No baseline existed, so improvement could not be shown. No implementation time was protected, so the daily business reabsorbed everyone within a week. No review standard was agreed, so quality disputes stalled the pilot. No feedback process existed. The executive ambition never connected to everyday work — or too many tools were introduced at once. In short: the workshop ended before the organisational work began.

A programme must extend beyond the room.

This is the premise of Leadership in the Age of AI, the programme we have developed with Regine Hüttebräuker and digitalbeirat. Seven modules form the complete system. Each engagement is configured from those modules and sequenced as one implementation cycle, according to the organisation's priorities. The cycle has three phases, and each exists for a reason.

Preparation, within the organisation: identify real workflows, select participants, establish a baseline, and clarify objectives and constraints — so the intensive works on the organisation's actual material rather than on hypotheticals.

The Mallorca intensive: build shared understanding, make decisions away from ordinary interruptions, work on real use cases, build or define an assistant, establish human-review requirements, and assign ownership.

Implementation, back in the team: sequence next steps, test with users, refine, document the working method, follow progress, and transfer responsibility into the organisation — the phase most workshops never reach.

Why the setting is part of the method.

Place does not replace method. It protects the conditions in which difficult decisions can be made. In practice that means distance from routine, fewer competing meetings, uninterrupted working blocks, shared meals, time for informal integration, hospitality that removes logistical friction, and a rhythm designed around concentration rather than around a calendar. We would not claim that a beautiful room produces innovation. We would claim that a protected one makes honest decisions more likely — and Bon Sentiment's craft is the protecting: setting, rhythm, hospitality, logistics, matched to each team.

What participants should leave with.

• A clearly defined AI use case from their own organisation
• A functional AI-assistant prototype for one defined use case, together with the requirements for safe implementation
• Named ownership and a sequence of next steps
• Agreed human-review and escalation points
• A role-specific AI-literacy framework
• A shared language for discussing AI use in the organisation

The programme is designed for a selected group of 6–12 leaders who can carry one shared method into larger teams and functions.

The closing question.

The most important AI decision may not be which tool to adopt next. It may be which part of the work deserves to be redesigned — and who is prepared to take responsibility for what follows.

If that question is live in your organisation, the programme exists to work on it. Leadership in the Age of AI is developed with Regine Hüttebräuker and digitalbeirat and configured for one organisation at a time: 6–12 leaders, a cycle of roughly ninety days with six working days on Mallorca, held in German or English, in a Mallorca location matched to the team. Fee on request.

The programme is facilitated by Regine Hüttebräuker, a change-management consultant, coach and trainer; Robert Fahle, who builds and implements AI assistants in organisations; and Dr Olivier Blanchard, who has spent two decades building digital products and services. Bon Sentiment composes what carries the work: the setting, the rhythm, the hospitality and the logistics.

Bring us one workflow, one leadership challenge or one AI pilot that has not yet translated into real work.