AI is everywhere. Tools, dashboards, pilots, consultants arriving with a promise. The temptation is to begin with the question that feels easiest: what can this do for us?<\/p>\n
That is not the question that matters.<\/strong><\/p>\n The question that matters is harder and less comfortable: what does this do to us, to how we lead, steer and work together? That is not a technical question. It is a question for those who govern, and it is precisely the question most organisations skip, not out of unwillingness, but because the technical question gives a faster answer and exposes less.<\/p>\n I see that pattern recur everywhere. An organisation wants to introduce AI, and the first conversation is about which tool, which platform, which vendor. That is a rational conversation. It feels forward-looking. But beneath that rationality often lies an avoidance. Because the moment you start with the tool, you do not have to answer three questions that are genuinely harder: what do we actually stand for as an organisation? What must never be decided by an algorithm? And what does it mean for my own role as a leader, when part of the thinking work is taken over by something I do not fully understand?<\/p>\n Those questions do not disappear because you skip them. They simply wait, and they return harder later, at the point when it is already more difficult to correct anything.<\/p>\n AI is not a neutral addition to an organisation. It sharpens existing tensions, whether you want it to or not.<\/strong><\/p>\n I see three recurring tensions in this respect. The first is the tension between efficiency and human scale. Do you mainly want to do more with fewer people, or do you want to free up time for attention, judgement and relationship? Both answers are legitimate. The problem is that many organisations never answer that question explicitly, and then discover afterwards that they chose the first implicitly, simply because that was the easiest route through an implementation process.<\/p>\n The second tension is data-driven steering versus professional autonomy. Does the system become the leading voice, or does the judgement of the professional remain central? And what do you say to someone who disagrees with the outcome of the model? That is not an abstract question. It is a question that arises on a Tuesday afternoon, when an experienced employee says: this is not right, even though the system says it is. If you have not prepared an answer to that, you improvise under pressure, and improvising under pressure rarely produces your best answer.<\/p>\n The third tension is control versus trust. Do you use AI to help, or to control? That distinction sounds subtle, but everyone in the organisation senses it immediately. People notice the difference between a system that relieves them and a system that monitors them, even when the technology is identical.<\/p>\n It is worth pausing on this, because it is precisely this tension that is least explicitly discussed in boardrooms, and most explicitly felt on the work floor. A dashboard measuring productivity might in theory be intended to flag bottlenecks so people can be helped. In practice it is almost always experienced as an instrument of control, simply because the context in which it is introduced, and the tone with which it is announced, make that interpretation unavoidable. Technology, in that sense, is never fully neutral. It inherits the intention of whoever introduces it, whether that intention has been made explicit or not.<\/p>\n What people hear does not depend on what you say, but on what your first pilot does.<\/strong><\/p>\n This may be the most important point of this entire essay. An organisation can have a beautiful vision statement about human-centred AI, full of words like trust, development and support. But if the first practical application of AI is a monitoring system tracking individual performance, nobody consciously ignores the vision statement. They simply believe the practice more than the document. And the practice says: we trust the system more than you.<\/p>\n That is not a matter of bad intentions. It is a matter of sequence. The first application of a new technology sets the tone for everything that follows, because people infer from it what the organisation genuinely values, regardless of what is written on paper.<\/p>\n That is not a matter of bad intentions. It is a matter of sequence. The first application of a new technology sets the tone for everything that follows, because people infer from it what the organisation genuinely values, regardless of what is written on paper.<\/p>\n There is something else that reinforces this dynamic, and it is rarely named out loud. People are remarkably good at detecting the gap between declared policy and actual behaviour, often far better than organisations themselves realise. An employee who sees that the first AI application is built around tracking individual speed draws an immediate conclusion, even though nobody has stated that conclusion explicitly. That conclusion then spreads informally, through corridors and coffee machines, faster and more credibly than any internal communication. By the time management realises which story is actually circulating, it has already taken root.<\/p>\n Research into human oversight of AI systems confirms why this carries such weight. When people are repeatedly exposed to systems that prepare or make decisions, the risk of automation bias arises: the tendency to accept the system’s outcome uncritically, even when one’s own experience suggests otherwise. That risk grows, not shrinks, the less explicit an organisation has been about when dissent is welcome and when it is not. The European AI Act recognises this problem too: for high-risk AI systems, human oversight is a mandatory design criterion, precisely because oversight without explicit room to intervene tends, in practice, to dissolve into a formality.<\/p>\n That is precisely why this first instalment of the series begins with purpose, not technology. Not because technology is unimportant, but because technology introduced without purpose ends up filling in the purpose itself. An organisation that does not explicitly choose what AI may and may not be used for ends up doing what is technically possible, rather than what it genuinely wants to be.<\/p>\n Nobody chooses that consciously. It simply happens, unless someone stops it.<\/strong><\/p>\n What does this require now, concretely, today? It starts with something small. Name one place in your organisation where AI already plays a part in a decision, even if nobody calls it that. Ask the question there: what is this actually doing to how we work and speak to one another? Not as a technical audit, but as an honest conversation.<\/p>\n Over a slightly longer horizon, it helps to design a fixed rhythm, say every quarter, in which the management team and a number of key figures from across the organisation reflect on what AI is actually doing, not what it is supposed to do according to the plan. That rhythm is more valuable than a one-off kick-off, because the impact of AI often only becomes visible after months, in patterns you only notice if you are deliberately looking for them.<\/p>\n And what is best avoided is starting with a pilot primarily aimed at control, without first having had the conversation about trust. That is the sequence that does the most damage, because it sets a tone that is difficult to reverse later.<\/p>\n This essay is the first of five. The following instalments address making purpose and boundaries concrete, how AI redistributes power and roles, the governance and learning capacity needed to carry that, and finally a compact compass in which everything comes together. But it begins here, with the question most organisations skip.<\/p>\n Are you mainly occupied with tools and use cases? Or is there already a conversation underway about who you want to be in a future where AI is present everywhere?<\/p>\n Notes for those who wish to read further:<\/em><\/p>\n Most organisations start with the question of what an AI tool can do for them. That feels sensible. It is often a flight forward, away from three conversations that are more uncomfortable than an implementation plan. The first instalment in a series on implementing AI without losing yourself.<\/p>","protected":false},"author":1,"featured_media":931,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[26,81],"tags":[89,100],"class_list":["post-5012","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","category-english","tag-english","tag-leadership"],"_links":{"self":[{"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/posts\/5012","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/comments?post=5012"}],"version-history":[{"count":1,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/posts\/5012\/revisions"}],"predecessor-version":[{"id":5127,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/posts\/5012\/revisions\/5127"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/media\/931"}],"wp:attachment":[{"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/media?parent=5012"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/categories?post=5012"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dbvp.nl\/en\/wp-json\/wp\/v2\/tags?post=5012"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}\n