AI Will Reward People Who Think Clearly
The bottleneck moved from production to cognition. That changes who wins.
Sahir Maharaj11 min read
For most of the history of knowledge work, a significant fraction of professional value was captured by the ability to produce: to generate text, to compile information, to format and present analysis, to move from raw input to polished output efficiently. The person who could write quickly, research thoroughly, and produce clean deliverables reliably was valuable partly because these productive capabilities were scarce and time-consuming. AI is changing this in a specific and important way: production is becoming cheap in ways that devalue the productive capabilities that have underpinned much of professional value, while the underlying thinking that determines the quality of what gets produced remains as scarce and as valuable as ever, possibly more so. The shift is from rewarding people who can produce to rewarding people who can think, and the distinction, which has always existed but was partly obscured by the scarcity of productive capability, is now becoming starker.
I want to be specific about what I mean by thinking clearly, because the phrase risks being too vague to be useful. The cognitive skills that I think will be most rewarded in the AI era are not mysterious or exotic. They are the skills that good epistemologists, good scientists, and good decision-makers have always valued: the ability to frame a problem correctly before attempting to solve it, the ability to evaluate the quality of evidence rather than just its volume, the ability to identify the assumptions embedded in an argument and assess whether they are warranted, the ability to distinguish between what is known, what is inferred, and what is assumed, and the ability to hold complexity and uncertainty without collapsing it prematurely into false simplicity. None of these skills is new. What is new is the premium that the AI era places on them relative to the productive capabilities they were previously bundled with.
The reason AI specifically rewards clear thinking rather than fast production is that the quality of AI outputs is highly sensitive to the quality of the thinking that directs them. A person who thinks clearly about what they need produces better prompts, evaluates outputs more accurately, identifies the gaps and errors that need correction, and integrates AI outputs into work that reflects genuine understanding rather than assembled plausibility. A person who does not think clearly but types fast now has access to tools that can produce the same volume of output with much less effort, but the quality of that output remains bounded by the quality of the thinking directing it. The productivity bottleneck has moved from production to cognition, and the people who invested in cognitive quality rather than productive speed are finding that the tools work better for them.
The cognitive skill that I observe being most consistently undervalued in professional contexts, and that I expect to become most clearly valuable in the AI era, is problem framing: the ability to define what the actual problem is before attempting to solve it. This sounds obvious, but the frequency with which people bring solutions to the wrong problem, or bring the right kind of solution to a poorly specified version of the right problem, suggests that it is harder than it sounds and less consistently practiced than the sophistication of the solutions it precedes would suggest.
AI makes the problem framing skill more important in a specific way: AI systems are very good at solving the problem they are given and have no capacity to notice when the problem they have been given is not the problem that needs solving. A coding AI will write excellent code for the function you specified, even if specifying that function was the wrong approach to the underlying system architecture problem. A writing AI will produce a polished version of the argument you outlined, even if the argument is the wrong response to the strategic situation. A research AI will compile comprehensive information on the question you asked, even if the question you asked is not the question that would actually inform the decision you need to make. The human in the loop who can identify when the AI is doing excellent work on the wrong thing is providing irreplaceable value. The human who cannot is simply accelerating the production of misaligned outputs.
Developing the problem framing skill requires practice in a specific kind of intellectual discipline: the discipline of staying with a problem in its undefined state long enough to understand it rather than immediately moving to solution mode. This is cognitively uncomfortable in ways that are specific and well-documented. The human mind prefers the reduction of uncertainty that solution generation provides, and the pressure of professional environments to produce and to move rewards the person who arrives with solutions rather than questions. The AI era does not eliminate this pressure, but it does change the relative value of arriving with the right question versus arriving with the wrong solution.
The second cognitive skill that the AI era places a premium on is evidence evaluation: the ability to assess the quality and relevance of information rather than just its volume and apparent confidence. This is a skill that has always been valuable but that was partly masked by the difficulty of accessing sufficient information to make it the primary bottleneck. In information-scarce environments, the person who could find the information was more valuable than the person who could evaluate it, because without the information there was nothing to evaluate. AI has made information access easy enough that evaluation is now the bottleneck.
The specific evaluation skills that matter are the ones that allow a person to assess whether information is relevant to the specific question at hand, whether the source and method of the information are reliable for this type of claim, whether the uncertainty in the information has been appropriately characterised, and whether the information is being interpreted in ways that are consistent with what it actually shows rather than what would be convenient for it to show. These are the skills of good scientific reasoning applied to the everyday information environment, and they are skills that most professional education does not develop with any deliberateness.
One of the specific evaluation challenges that the AI era creates is the evaluation of AI outputs themselves, which requires a meta-level application of evidence evaluation skills: the ability to assess not just whether the content of an AI output is accurate but whether the methodology by which it was produced is appropriate for the question. A person who asks an AI to analyse a business situation and then evaluates the AI output as though it were the product of a skilled analyst is making a category error: the AI has produced a plausible analysis, not necessarily a sound one, and the difference requires the kind of domain expertise and methodological awareness to detect that the prompt-and-evaluate workflow does not automatically provide.
The practical question that follows from the argument that AI rewards clear thinking is how clear thinking is developed, and whether the practices that develop it are compatible with AI-assisted work. I think the answer is yes, but it requires some deliberateness about how AI tools are used rather than using them as default shortcuts for the cognitive work that produces the clear thinking in the first place.
The most important practice for developing clear thinking in an AI world is maintaining the habit of doing the foundational cognitive work before going to the AI. Thinking through what the actual problem is before asking AI to help solve it. Developing a preliminary view on a question before asking AI to research it. Forming an initial argument before asking AI to strengthen it. The preliminary work is not wasted even when the AI output supersedes it: the discipline of having thought through the problem develops the cognitive muscle that makes the AI-assisted work better, and the preliminary view provides a standard against which to evaluate the AI output rather than simply receiving it as the answer.
The educational implication is uncomfortable but important: the development of clear thinking requires engaging with difficult cognitive work without shortcuts, and the availability of AI shortcuts creates pressure toward avoiding the specific kinds of difficulty that produce cognitive development. The student who uses AI to solve problems rather than to check their own solutions is not developing the problem-solving capability that the solution process develops. The professional who uses AI to draft the structure of their argument before thinking through the argument themselves is not developing the structuring capability that the drafting develops. This does not mean AI should not be used for these purposes. It means the people who also maintain a practice of doing the hard work without AI are building something that the AI-only workers are not, and that something turns out to be the primary source of competitive advantage in the AI era.
Topics in this article
You might also like
View all
AI Has Created a New Kind of Expert
Fluent, confident, and thin. What happens when the expertise signal stops working.
Why the Quality of Your Curiosity Matters
When answers are cheap, the scarce resource is a question worth asking.
Your Company Will Not Replace You With AI
It will replace five of you with one person who knows how to run the tools.