By Diana Garcia Quevedo, Anna Glaser & Caroline Verzat
AI tools are transforming organisational workflows. They summarise, predict, generate, and assist – all in seconds. Their speed is impressive, and often helpful. But speed is not accuracy. And writing is not knowledge. Recent studies show factual error rates of up to 79% in AI-generated reasoning tasks. Yet these answers are often persuasive, presented with such linguistic ease that users accept them without hesitation.
This is the real risk of “fast answers”: they create the illusion of understanding. We believe we’ve grasped something – when in fact, we’ve merely received a fluent output. Understanding is superficial. Judgment is deferred.
The consequences are already visible. Junior software engineers report higher productivity with AI-assisted coding – until bugs accumulate and code becomes harder to maintain, then productivity falls. In publishing, AI-generated books about ADHD and autism have appeared on Amazon, filled with misleading or dangerous claims, yet sold without expert oversight. In academia, a philosophical article credited to a fictional author – created entirely by a language model – made it through peer review. The author didn’t exist, but the text was convincing enough to pass.
What these examples show is not just AI models’ limitations and ethical problems. They reveal something deeper: a lack of critical evaluation of AI answers. The more we rely on AI’s probabilistic outputs, the more we risk outsourcing not just tasks, but thinking itself.
This matters – especially in leadership. Decisions are rarely about information alone. They’re about context, trade-offs, and values.
So how should we respond?
The answer is not to reject AI, but to embed it in structures that support reflection. That begins with a concept often overlooked in digital transformation: reflexivity.
Reflexivity means questioning our beliefs, values, and actions that shape our thinking. It requires us to see knowledge co-created in relation to others. It means asking not only what an AI system produces, but how it produces it – and how we, as users, interpret and act on it. It’s about clarifying roles: what we delegate to machines, and what remains a human responsibility.
In organisations, reflexivity isn’t built alone. It depends on team dynamics – on bringing together diverse forms of expertise: technical, operational, contextual. In practice, that means including not only engineers and managers, but also those who understand human systems, social dynamics, and anticipate unintended consequences. These are not peripheral perspectives; they are essential to the responsible and effective design and implementation of AI.
This kind of reflexive diversity isn’t a luxury. It’s a safeguard. It protects against blind spots, premature certainty, and the narrowing of inquiry to what AI can handle easily.
In our research on AI use in qualitative analysis, we recommend combining techniques and testing models, and critically evaluating the outputs and language models themselves.Of course, all this takes time. But speed, by itself, is not a virtue. The goal is not to slow everything down – but to protect space for thinking where it matters most. Sometimes, a moment of hesitation is the most intelligent move a system – or a leader – can make.
Reflexivity is not inefficiency. It’s what makes judgment possible. And in a world of fast answers, it may be the only thing that protects us from shallow thinking.