Mark Emdin
March 2026
The opportunities and risks of artificial intelligence are already well documented. Some of the writing is careful and thoughtful, some is evangelical, some alarmist, and much of it focuses on what AI can do faster, cheaper, or at greater scale than we humans. That conversation matters, but by now it is also familiar. And increasingly, it feels incomplete.
Towards the end of last year, I joined a webinar titled AI for Change Makers, hosted by organisation development specialists Mayvin. During the session, participants were asked to reflect on a deceptively simple question: where might you deliberately add more friction into how you use AI? Not whether we should adopt AI at work, nor how quickly we can do so, but why we might choose to slow certain things down.
From an organisation development perspective, that question goes to the heart of what our work is all about. OD has always lived in the tension between the technical and the social. We design structures, processes, roles, and operating models, while simultaneously working with meaning, identity, power, emotion, and relationships. This is not a new insight. The sociotechnical tradition, articulated by Emery and Trist, made clear that organisational effectiveness depends on the joint optimisation of technical and social systems, not the dominance of one over the other[1]. Galbraith’s work on organisation design reinforced the same idea: structure alone never produces performance without alignment to people and behaviour.[2]
What AI does is dramatically amplify one side of this equation. It is extraordinarily effective at technical work: synthesising data, generating options, spotting patterns, and producing plausible, fluent outputs at speed. Where OD work is technical, AI can be a powerful ally. The risk emerges when that technical fluency drifts into domains where judgement, meaning, sense-making, and responsibility cannot be automated, but are quietly displaced nonetheless.
This is where friction becomes important.
Philosopher of technology Shannon Vallor offers a helpful lens. In her 2024 Turing Lecture, she argues that much contemporary AI, particularly large language models, is best understood not as a mind but as a mirror, reflecting patterns from human-made data back at us. That mirror can be useful, and in some domains life-saving, but it can also distort, magnify injustice, and produce what she calls confident nonsense[3].
Vallor’s deeper concern is not simply about error or bias. It is about what the widespread, largely unreflective use of AI takes away from us as humans. She suggests that current AI systems tend to erode three human necessities: space, time, and stories.
By space, Vallor means the “space of reasons”: the social and relational space in which humans test ideas, argue about justifications, sit with uncertainty, and change their minds together. Many AI tools are designed as space-fillers, providing instant answers that reduce the felt need for collective reasoning. The danger here is not only that the answers might be wrong, but that our capacity to reason together slowly atrophies through disuse.
Time is also quietly compressed. Generative AI is conservative in a specific sense: it is bound to the past data on which it is trained. Even its apparent creativity is a recombination of what has already been. Humans, by contrast, experience the future as open and are capable of breaking patterns rather than simply extending them. When AI outputs are treated as forecasts or inevitabilities, the future subtly closes down. As OD practitioners, much of our work involves noticing patterns and then deliberately experimenting with what happens when we interrupt them.
Stories may be the most fragile of all. Humans make meaning through narrative: about who we are, what matters, and where we are going. Vallor worries that we are increasingly outsourcing reading, writing, teaching, and sense-making to systems for which nothing is at stake, because they do not live with the consequences of the stories they tell. From an existential perspective, this is not a small loss. From an OD perspective, it strikes at the core of change work itself.
Space, time, and stories are not abstract philosophical ideas. They are the raw materials of organisational change, and for leaders they are also the conditions for sound judgement: without space to test reasons, time to hold uncertainty, and stories people can own, change becomes compliance theatre rather than commitment.
A Story From The Inside
This becomes particularly tangible when you listen to people doing OD work inside organisations. In a recent conversation with a senior internal OD leader, she described a growing unease about the direction her function was being pulled. Her concern was not that AI or new technologies were being introduced, but that the broader environment of constant pace, efficiency pressure, and delivery demand was nudging OD toward becoming more transactional: less deep craft, more “deliver the thing”.
Without quite naming it, she was pointing to the erosion of reflective space. OD, she suggested, was increasingly positioned as execution support rather than as a place where the organisation could slow down enough to think about what kind of system it was becoming. The implication was sobering but familiar: human-centred work is harder to sustain in contexts optimised for speed, productivity, and relentless forward motion, and AI intensifies those conditions.
This is not only happening in OD. It is happening across much of organisational life.
This is why friction matters. In social systems, friction is not simply inefficiency. It is often the signal that something carries weight. Friction creates pauses, invites challenge, and forces responsibility back into human hands. It protects precisely those capacities that AI, if left unchecked, tends to compress.
Karl Weick’s[4] work on sense-making is instructive here. Organisations rarely fail because they lack information; they fail because they stop noticing, questioning, and revising their interpretations. AI makes it easier to generate interpretations, but it also makes it easier to stop thinking once a plausible one appears. Friction interrupts that slide into premature certainty. Friction asks “what and how are we missing”.
Seen this way, friction is not a nostalgic preference for slowness. It is a design choice.
There are several places where friction can be deliberately introduced into the use of AI, particularly in organisational change contexts.
- At the point of sense-making, not production
AI excels at producing outputs. That is precisely where friction is most needed. Rather than treating AI-generated analyses or recommendations as conclusions, OD work benefits from slowing down what happens next. Asking what feels too neat or too confident, and what lived experience, political reality, or emotional data is missing, keeps interpretation human and collective. The friction comes after the answer, not before it. - In how problems are framed
Some of the most consequential AI failures are not wrong answers but well-formed answers to the wrong questions. Friction here means resisting the rush to refine prompts and instead disputing the framing itself. Who benefits if this framing becomes dominant? What assumptions or biases are quietly embedded? What would an inconvenient stakeholder call this problem? Only then is AI invited in. - Around authority and authorship
AI outputs often arrive with an aura of objectivity that can quietly displace responsibility. Friction can be introduced by making authorship visible and accountability explicit. Labelling AI-assisted material, asking whether a group would stand behind a conclusion without the tool, or deliberately appointing someone to challenge AI-derived insights all help return authority to human judgement. - In emotional and relational domains
Change work is never purely cognitive. In some domains, friction is best created by restraint rather than inclusion. Refusing AI-generated empathy maps, stakeholder narratives, or cultural interpretations unless they are grounded in direct human engagement protects the messy, embodied intelligence at the heart of adaptive change. - Through time itself
Speed is one of AI’s strongest seductions. Delay can be a powerful design move. Revisiting AI-informed decisions after a pause, re-running the same analysis weeks later, or delaying commitment to AI-supported forecasts all help reveal how contingent and context-dependent many insights really are. Time reopens the future rather than collapsing it into the past. - Where power is at stake
AI systems tend to reflect and reinforce existing power structures by default. Friction here involves asking questions that are uncomfortable but necessary: whose voices are absent from the data, who gains legitimacy from this output, and who loses it? Framed as design risks rather than moral accusations, these questions prevent neutrality from becoming a cover for consolidation.
The point, is not that AI threatens OD, nor that OD should resist AI. It is that AI strengthens the technical side of our work in ways that make the social side more, not less, important. Friction is how we protect that social work.
Further, friction can result in the sparks of creativity that can only come with space, time and stories
