James Barton argued in a recent article that AI may not be making us worse thinkers at all. Instead, it may be teaching us to think differently by shifting more of our mental effort away from producing answers and towards framing questions, challenging assumptions and judging what to trust. That got us thinking about a related issue: if AI gives more people access to answers, what becomes the real marker of expertise inside an organisation?
For a long time, knowledge work rewarded people for having answers.
Expertise was often tied to what someone knew, what they had seen before, or what information they could access faster than everyone else. In that world, value sat close to recall. The person with the insight, the precedent, the framework or the explanation had a clear advantage.
AI is starting to unsettle that model.
As James notes, generative AI can now draft, summarise, analyse, present options and respond to questions in seconds. That does not make human judgement irrelevant. But it does change where human value is created. When answers become easier to generate, the differentiator shifts towards choosing the right problem, adding the right context, spotting the weak assumption, and deciding whether an answer deserves trust.
There is a popular concern that the more we use AI, the less we think for ourselves.
It is not a baseless concern. James acknowledges that evidence around cognitive offloading suggests that when people rely on external systems without questioning them, some mental skills can weaken over time. He also notes that users who accept AI outputs without challenge tend to engage less deeply with the problem in front of them.
But that is only part of the picture.
The more interesting point in the article is that many effective AI interactions do not look like passive dependency at all. They look like conversation. A user asks a question, challenges the answer, adds more context, requests another angle, tests the assumptions, and refines the prompt until something more useful appears. James argues that this process closely resembles traditional critical thinking: questioning assumptions, analysing information, exploring alternatives and making reasoned judgements.
That changes the debate.
The real risk may not be AI use itself. It may be unthinking AI use.
This is where the article becomes especially useful.
James draws a clear distinction between people who use AI as a shortcut to avoid thinking and people who use it as a tool for exploration. In his framing, the problem is not volume of use but quality of use. Someone who accepts the first answer and moves on is probably outsourcing judgement. Someone who uses AI to test ideas, pressure assumptions and refine thinking may actually be exercising judgement more actively than before.
That is a much more practical lens for leaders.
It suggests that AI does not have a single effect on thinking. It amplifies the habits that already surround it. Strong users tend to bring structure, curiosity and scepticism. Weaker users tend to bring vagueness, passivity or overconfidence.
In that sense, AI may be less of a replacement for thinking and more of a mirror for it.
One of James’s sharpest observations is that prompt engineering sounds technical, but is really about structured thought.
To get meaningful output, people have to define the goal clearly, provide relevant context, recognise constraints, explore alternatives and refine their instructions. Vague thinking produces vague results. AI simply exposes that more quickly than many other systems do.
That matters because it reframes what organisations should be developing.
If AI performance depends heavily on the quality of human framing, then the real capability gap may not be technical literacy alone. It may be the ability to define problems clearly, ask sharper questions and apply better judgement under uncertainty.
That is not a small shift.
It means the organisations that get the most from AI may not be the ones with access to the most advanced tools. They may be the ones whose people are best at thinking before they ask.
If AI responses are being used without challenge, refinement or verification, the organisation is not speeding up thinking. It is skipping it.
When teams focus only on prompt formats and hacks, they often miss the more important issue: whether people know how to define the problem well in the first place.
If the conversation is dominated by speed, automation and output volume, with little attention paid to evaluation and decision quality, the organisation may be scaling weak thinking faster.
If status continues to depend on recall rather than framing, diagnosis and judgement, the business may be measuring the wrong capabilities for an AI-shaped environment.
If James is right, then organisations need to become much more intentional about the human skills that sit around AI.
That does not mean resisting the technology. It means recognising that the value of AI depends heavily on the quality of the thinking that guides it.
Employees need more than access to AI. They need to know how to define the task, supply context, clarify the objective and recognise what a good answer would need to include.
If people are only praised for producing work quickly, they will naturally be tempted to accept plausible outputs too early. Teams need to value verification, judgement and thoughtful refinement as well as productivity.
The best AI users do not just ask for output. They probe reasoning, request alternatives and test whether the conclusion stands up. Those habits can be taught and reinforced.
As answers become easier to generate, expertise increasingly lies in judgement: knowing what matters, what is missing, what does not ring true, and what question should be asked next.
This matters because many organisations are still framing the AI challenge in the wrong way.
They worry that easy access to answers will lower capability across the business. That can happen. But James’s article suggests a more useful concern: whether organisations are helping people adapt to a new form of cognitive work, where the hardest and most valuable thinking happens before and after the answer appears.
That is a leadership issue, not just a tooling issue.
Leaders shape whether AI is adopted as an answer machine or a judgement tool. They influence whether teams are encouraged to question outputs, add context and think critically, or simply move faster with less reflection.
The long-term effect of AI on expertise will depend in large part on those choices.
James ends with a powerful question: if AI gives everyone access to answers, what becomes the true measure of expertise? His suggestion is that the advantage may no longer belong to the person who simply knows the answer, but to the person who can identify the right problem, spot the flawed assumption, or ask the question that changes the discussion altogether.
That feels like the right place to focus.
The future of expertise may not be about defending human value against AI by holding on to answer ownership.
It may be about strengthening the distinctly human capabilities that matter more when answers are plentiful: framing, discernment, scepticism, contextual judgement and the ability to navigate uncertainty well.
In other words, AI may not be reducing the need for critical thinking.
It may be making it far more visible who is really doing it.