ORGANISATIONAL HEALTH
AI Certainty, Paradox, and the Limits of “Holding Both Sides”
23.10.2026
Fifty-one of the sixty-one senior leaders KDVI interviewed for The Human Multiplier admitted to conflicting feelings about AI. The study's first finding, published on 22 October 2026 as "What Is Your AI Certainty Costing You?", is that this ambivalence is an asset and the real risk is becoming convinced too early. I agree with the diagnosis. I am less convinced by the cure.
Holding a paradox is a posture. Working it is a capability, and it needs an organisation built to carry what it decides. That is the argument of this piece, and the work we do.
Two Exits from the Same Room
The article's sharpest observation is that champions and sceptics make the same move: both relieve the discomfort of holding two valid concerns at once, then leave in opposite directions. KDVI, true to its psychodynamic roots, reads this as a defence against anxiety. Might be true, but only in part: it is rather that many leaders lack the cognitive capability to do anything with a contradiction except resolve it.
Holding Is Not Working
KDVI's prescription is to "work with the paradox" rather than settle it. Here is where Otto Laske's dialectical thought forms begin. Formal-logical thinking (Hegel's Understanding) sees a contradiction and treats it as a defect: true or false, adopt or resist. Dialectical thinking (Reason) treats it as information about how the system is about to change.
A leader who merely holds "efficiency versus values" produces the French case from the study: the data is "not clean enough", so nothing moves. A dialectical thinker asks what the two poles share, which is the work itself, and where the system is about to change in kind rather than in size. A QA function cut from fifteen people to three after AI came in is not just a smaller QA function. It is a different one, and nobody had decided what it should be.
The same lens explains the study's best phrase, "organisational hallucination": management believed adoption was widespread while the frontline saw almost none. "Adoption" had been frozen into a state; nobody was watching it as a process.
Seen this way, KDVI's three paradoxes (belief versus reality, values versus efficiency, short versus long term) collapse into one: treating AI as a thing acquired rather than a transformation of the work system.
A Role Is Not a Capability
The authors propose a "judgement steward" who decides where AI adds value and where human judgement is essential. As Jaques taught, naming a role does not create the capability it requires. One leader in the study put the real problem in a sentence: AI makes it harder to see where he adds value. That is a stratum question. When machines absorb the first layers of work, value migrates upward to setting context, which demands a capacity few executive teams have ever had the chance to look at in themselves.
Where We Come In
The numbers behind KDVI's paradox are not kind. Gartner expects over 40 per cent of agentic AI projects to be cancelled by 2027 and opens its 2026 Hype Cycle for Agentic AI by noting that hype and confusion are outpacing progress. A 2026 Strategy Partners study of the Russian market found that 7 to 10 per cent of AI-agent pilots reach production, and that estimates of adoption differ thirtyfold, from 97 per cent in executive self-reports to under 3 per cent in the strictest counts. Organisational hallucination, confirmed. The study's conclusion is that success is decided not by the technology but by the owner of the process. That is an organisational finding, and organisations are our natural and favourite ground.
Basically, we help leadership teams with two things. The first is the paradox itself: not resolving it for them, but turning it into decisions with a price, a sequence and a point of review, so that neither the champion nor the sceptic has to win. The second is putting AI where it is needed, inside an organisation healthy enough to carry it: where the intention behind AI matches the organisation's design, and where that design is coherent in itself, across people, working practices and the structure of roles and relationships, and fitted to the stage of life the company is in.
Take an entertainment business of a few hundred employees we worked with this year as an illustrative case. The leadership came with two sensible positions: it wanted to understand whether AI was worth starting at all, and there would be no major AI budget next year. That is the ambivalence KDVI describes, and a good place to start talking. A quick review added something else: nineteen AI services were already in use across twenty-one units, mostly through personal accounts, because people had found them useful before leaders had decided anything. So the question changed shape, from whether to start to how to look after something that had started on its own, at a pace the organisation could afford. We came back with a staircase rather than an investment case: first, the same tools people already like, on company accounts, with clear rules and a named owner, two months and cheap enough for almost any budget; then order in the data before any agent touches it; then AI taking the routine, with one principle agreed with the chief executives up front: decisions on content stay with people. Numeric gates between steps make moving on a matter of evidence, not of faith in the consultant. Nobody had to pick a side, and the organisation gained a way to find out, step by step, what AI is for in its own work.
The capacity to work a contradiction is measurable and developable. Three self-reflection questions will not do it. Read the piece. Then listen to how your team talks about AI, not what it concludes: what do they treat as fixed, what do they see already moving, what do they connect, and where do they expect the system to outgrow itself?
References
Kets de Vries, O., Kets de Vries, M., Chupina, Y. and Florent, E. (2026). What Is Your AI Certainty Costing You? KDVI, 22 October 2026. Based on The Human Multiplier, the KDVI AI Leadership Study (61 senior leaders, 20+ sectors, 14 countries).
Laske, O. (2008; 2nd ed. 2017). Measuring Hidden Dimensions of Human Systems: Foundations of Requisite Organization. Medford, MA: IDM Press. The Dialectical Thought Form Framework and the table of 28 thought forms, adapted from Basseches.
Gartner (2025). Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. Press release, 25 June 2025.
Gartner (2026). Hype Cycle for Agentic AI, 2026. 2 April 2026.
Strategy Partners (2026). ИИ-агенты: успех внедрения решает не технология, а владелец процесса. Moscow: Strategy Partners.
Authors:
Pavel Charny, ORTALEX, Managing Partner, Dubai
Holding a paradox is a posture. Working it is a capability, and it needs an organisation built to carry what it decides. That is the argument of this piece, and the work we do.
Two Exits from the Same Room
The article's sharpest observation is that champions and sceptics make the same move: both relieve the discomfort of holding two valid concerns at once, then leave in opposite directions. KDVI, true to its psychodynamic roots, reads this as a defence against anxiety. Might be true, but only in part: it is rather that many leaders lack the cognitive capability to do anything with a contradiction except resolve it.
Holding Is Not Working
KDVI's prescription is to "work with the paradox" rather than settle it. Here is where Otto Laske's dialectical thought forms begin. Formal-logical thinking (Hegel's Understanding) sees a contradiction and treats it as a defect: true or false, adopt or resist. Dialectical thinking (Reason) treats it as information about how the system is about to change.
A leader who merely holds "efficiency versus values" produces the French case from the study: the data is "not clean enough", so nothing moves. A dialectical thinker asks what the two poles share, which is the work itself, and where the system is about to change in kind rather than in size. A QA function cut from fifteen people to three after AI came in is not just a smaller QA function. It is a different one, and nobody had decided what it should be.
The same lens explains the study's best phrase, "organisational hallucination": management believed adoption was widespread while the frontline saw almost none. "Adoption" had been frozen into a state; nobody was watching it as a process.
Seen this way, KDVI's three paradoxes (belief versus reality, values versus efficiency, short versus long term) collapse into one: treating AI as a thing acquired rather than a transformation of the work system.
A Role Is Not a Capability
The authors propose a "judgement steward" who decides where AI adds value and where human judgement is essential. As Jaques taught, naming a role does not create the capability it requires. One leader in the study put the real problem in a sentence: AI makes it harder to see where he adds value. That is a stratum question. When machines absorb the first layers of work, value migrates upward to setting context, which demands a capacity few executive teams have ever had the chance to look at in themselves.
Where We Come In
The numbers behind KDVI's paradox are not kind. Gartner expects over 40 per cent of agentic AI projects to be cancelled by 2027 and opens its 2026 Hype Cycle for Agentic AI by noting that hype and confusion are outpacing progress. A 2026 Strategy Partners study of the Russian market found that 7 to 10 per cent of AI-agent pilots reach production, and that estimates of adoption differ thirtyfold, from 97 per cent in executive self-reports to under 3 per cent in the strictest counts. Organisational hallucination, confirmed. The study's conclusion is that success is decided not by the technology but by the owner of the process. That is an organisational finding, and organisations are our natural and favourite ground.
Basically, we help leadership teams with two things. The first is the paradox itself: not resolving it for them, but turning it into decisions with a price, a sequence and a point of review, so that neither the champion nor the sceptic has to win. The second is putting AI where it is needed, inside an organisation healthy enough to carry it: where the intention behind AI matches the organisation's design, and where that design is coherent in itself, across people, working practices and the structure of roles and relationships, and fitted to the stage of life the company is in.
Take an entertainment business of a few hundred employees we worked with this year as an illustrative case. The leadership came with two sensible positions: it wanted to understand whether AI was worth starting at all, and there would be no major AI budget next year. That is the ambivalence KDVI describes, and a good place to start talking. A quick review added something else: nineteen AI services were already in use across twenty-one units, mostly through personal accounts, because people had found them useful before leaders had decided anything. So the question changed shape, from whether to start to how to look after something that had started on its own, at a pace the organisation could afford. We came back with a staircase rather than an investment case: first, the same tools people already like, on company accounts, with clear rules and a named owner, two months and cheap enough for almost any budget; then order in the data before any agent touches it; then AI taking the routine, with one principle agreed with the chief executives up front: decisions on content stay with people. Numeric gates between steps make moving on a matter of evidence, not of faith in the consultant. Nobody had to pick a side, and the organisation gained a way to find out, step by step, what AI is for in its own work.
The capacity to work a contradiction is measurable and developable. Three self-reflection questions will not do it. Read the piece. Then listen to how your team talks about AI, not what it concludes: what do they treat as fixed, what do they see already moving, what do they connect, and where do they expect the system to outgrow itself?
References
Kets de Vries, O., Kets de Vries, M., Chupina, Y. and Florent, E. (2026). What Is Your AI Certainty Costing You? KDVI, 22 October 2026. Based on The Human Multiplier, the KDVI AI Leadership Study (61 senior leaders, 20+ sectors, 14 countries).
Laske, O. (2008; 2nd ed. 2017). Measuring Hidden Dimensions of Human Systems: Foundations of Requisite Organization. Medford, MA: IDM Press. The Dialectical Thought Form Framework and the table of 28 thought forms, adapted from Basseches.
Gartner (2025). Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. Press release, 25 June 2025.
Gartner (2026). Hype Cycle for Agentic AI, 2026. 2 April 2026.
Strategy Partners (2026). ИИ-агенты: успех внедрения решает не технология, а владелец процесса. Moscow: Strategy Partners.
Authors:
Pavel Charny, ORTALEX, Managing Partner, Dubai