It is early. The screen lights up before the meeting room fills. The first question of the day is not a greeting but a prompt. An answer rolls out that hangs together better than yesterday’s notes. Someone sighs with relief, someone else frowns, a third feels an inexplicable irritation.
In those three microreactions lies the whole field exposed: AI as promise, as threat, as mirror.
What happens here is not purely technical or procedural. It is also psychological, relational, governance-related and legal. It touches how you provide direction, how teams make meaning, and how the organisation carries accountability. The core message is simple and sharp: AI does not change people into a different species, but amplifies what is already there, and makes the consequences explicit in governance and law.
In classical relationships, omnipotence, certainty and omniscience were projected onto the leader. Now the algorithm slides into that role. It appears as the new object: smart, fast, seemingly neutral. That appearance of neutrality makes it attractive as a carrier of idealisation and denial. The model says so has grown into a modern defence mechanism, a way of avoiding shame, doubt or moral burden.
Psychodynamically, this is not a deviation but predictable: transference and countertransference play out between people and systems just as much as between people themselves. We talk to the system, but really to our need for certainty. We listen to the score, but really to our fear of falling short.
AI is not a standalone tool. It is a system node creating new feedback loops, dependencies and power structures.
From a systems-thinking perspective, the core question shifts: not whether AI works, but which patterns it reinforces or suppresses. Organisations that want to handle complex reality need variation in perspectives. Many models do the opposite: they score, sort, normalise. That is efficient, but also a source of cultural drift towards conformity and risk aversion. Less deviation, less dissent, less creativity, with a governance side effect: blind spots that only become visible in incidents and complaints, at the point when correcting is already more expensive than prevention would have been.
In systemic language, AI codes implicit norms. Power shifts towards datasets, features and thresholds. Whoever does not pay attention entrenches exclusion as objective logic. That is why governance is not primarily a document, but a relational design: who defines the data, who chooses the models, who may make exceptions, who may override? For executives, this leads to a simple rule with hard consequences: AI initiatives are change initiatives. They are interventions in roles, processes, identity and accountability. Whoever runs them purely as an IT project gets technical delivery and social collateral damage.
AI can increase productivity, shorten turnaround times and make quality more predictable. But without a goal architecture, it collapses into scattered pilots and demo wins without lasting value. Strategically, the order is: problem, value, risk, design. First define clearly which problem and which value, only then choose tools. Otherwise technology becomes the solution and people become the problem.
Learning is the weak spot of AI. It excels at getting better at what we already do, but the question of whether we are measuring the right things remains a human process.
Whoever strips out that layer amplifies blindness with high precision: beautiful dashboards, wrong decisions. Learning organisations therefore organise three practices. Reflection on outcomes: not only is it correct, but what does this mean? Dialogic decision-making: data as a conversation starter, not as arbiter. And co-evolution of skills: digital literacy plus judgement, moral sensitivity and legal alertness, which must keep pace with each other rather than drift apart.
AI operates within a normative landscape that, in Europe, is layered and in motion. The basic principles of data protection have stood for years: lawfulness, purpose limitation, data minimisation, accuracy, storage limitation, integrity and accountability. Around that lie rules for platforms, data portability, security and duties of care in essential sectors. Specific AI frameworks demand controlled risk analysis, documentation, monitoring and means of correction.
The governance verb here is demonstrable. Not only being careful, but being able to show that you were careful: model cards, logs, version control, impact assessments, override decisions with reasoning, incident notes with remedy. Establish who may stop the system, and when. Legitimacy is more than legality. It rests on explainability, contestability and remedy. Whoever automates must also make visible what is not automated.
AI touches professional identity. As systems analyse, advise and create, what does it mean to be an expert or a leader?
The task shifts from knowing more to being able to carry more: tension, uncertainty, slowness, conflict and moral ambiguity. The core question becomes: what do we not do, even though we could? Which applications do we refuse because they hollow out relationships, dignity or the future, even when the technology works perfectly?
Embodied work is a precondition here, not an afterthought. AI accelerates rhythm and increases information pressure. The body pays first. Teams that organise rhythm, silence, recovery and reflection hold their course. Teams that skip that layer burn out, and that is not only a wellbeing issue, but a governance risk: absenteeism, mistakes, eroding trust.
Governance reality is a field of tensions that AI makes visible and unavoidable. Efficiency versus resilience: faster and cheaper can be more fragile, so choose resilience where disruption is costly and efficiency where recovery is easy. Centralisation versus autonomy: one platform gives scale but reduces local wisdom, so central standards with local variation in application. Automation versus human dignity: delegation relieves but can dehumanise, so clear limits with contestability and remedy. These tensions do not disappear through policy. They call for rhythm and conversation. Governing is holding course amid contradictory truths.
Great ambitions succeed by doing a small number of things consistently.
A working minimum consists of clear roles: a system owner who holds the mandate, a model owner who guards performance and bias, a data owner who manages quality and access, risk and compliance who translate frameworks into workable requirements, and a business owner who anchors value and effect in daily operations. It also consists of a fixed rhythm: monthly monitoring and incident review, a quarterly challenge session with independent counterforce to prevent groupthink, and a biannual audit on documentation, drift, fairness and rights.
What does this ask of you, today? Not optimising a tool, but answering an old question in a new form: what do we do because we must, and what do we deliberately leave undone, even though we could? AI makes that question more urgent than ever, simply because, once an answer is given, it applies it at a scale no previous technology has ever reached.
Notes for those who wish to read further:
- W. Ross Ashby, An Introduction to Cybernetics (1956, Chapman & Hall). The original source for the concept of requisite variety, relevant to why AI systems that reduce too much make organisations blind.
- Wanda Orlikowski, Sociomaterial Practices: Exploring Technology at Work (2007, Organization Studies). On how technology and organisational relationships continuously shape each other.
- Chris Argyris & Donald Schön, Organizational Learning: A Theory of Action Perspective (1978, Addison-Wesley). On the distinction between single-loop and double-loop learning, crucial to understanding what AI can and cannot do.
- Ralph Stacey, Complex Responsive Processes in Organizations (2001, Routledge). On how power and patterns form within complex, responsive systems.
- ISO/IEC 42001:2023, Information Technology — Artificial Intelligence — Management System. The international standard for structural governance of AI systems within organisations.