Alex Karp, CEO of Palantir, recently offered a provocation worth sitting with. AI will displace most routine white-collar work, he argues, but two groups retain a genuine edge: people whose work requires real-world physical presence and judgement, and people who think differently — those with ADHD, autism, or dyslexia, who approach problems like artists rather than following a playbook. That mindset, Karp suggests, is harder for AI to replicate than most assume.
Bradley Wilson, a researcher who has spent nine months analysing data on the experiences of neurodivergent employees at work, finds the provocation substantially validated, and substantially complicated. The edge is real. Neurodivergent employees in his research panels show greater eagerness to take on responsibility, stronger confidence in their leaders’ ability to handle change, and more likelihood of recommending their employers. These are markers of genuine investment, not deficits.
But the structural conditions in most organisations systematically prevent this edge from expressing itself. The same employees report significantly greater stress, more role ambiguity, greater fragmentation of their working days, and higher departure intent. The edge exists. The conditions that would allow it to operate largely do not.
This is a recognisable pattern. overoptimisation describes how systems narrow to the measurable and rewarded, shedding unmeasured load-bearing capacity in the process. The same logic applies to cognition. Standardised processes, uniform meeting cultures, rigid attendance requirements, tightly defined role expectations — these are optimisations. They reduce variance that feels unmanageable and increase consistency that feels like control. What gets shed quietly in the process is the cognitive diversity that cannot be easily specified, measured, or rewarded.
The irony is precise. The employees most likely to think differently from AI — to pattern-cross, tolerate ambiguity, depart from the playbook — are operating in environments optimised for the cognitive style AI is best at replacing: consistent execution of defined tasks. Organisations are removing the capacity they most need by tightening the conditions that suppress it.
the-homogeneity-trap describes a related problem in AI development itself: when the people building AI think in similar ways, the frame they bring becomes invisible as a frame. The same trap runs in organisations as employers. When the conditions of work are designed around a narrow cognitive norm — the person who processes information linearly, operates well in fragmented attention environments, is comfortable with ambiguous verbal expectations — the norm disappears as a norm. It simply looks like what work is.
Wilson’s data adds a further layer. The highest-leverage variable across his research panels is the manager relationship. The engagement gap between a neurodivergent employee who has a manager they feel comfortable talking to and one who does not is larger than any other variable in the dataset. The edge is not suppressed by grand policy failures. It is suppressed at the most proximate level, in the daily question of whether someone with authority is actually attending to the specific person in front of them.
This is also a version of separated-knowledge. The people who can see most clearly when a frame no longer fits — whose cognitive style makes them better at noticing what the room has stopped questioning — are systematically separated from the conditions that would allow that noticing to do any work. They leave, or they adapt, which generally means learning to think within the norm rather than around it.
The organisations likely to hold this edge are not necessarily running neurodiversity programmes, though some are. They are doing something simpler and harder: improving the baseline conditions that allow any employee to do their actual best work. Clear expectations. Adequate resources. Managers who follow through. Protected time for sustained attention on genuinely difficult problems. These improvements benefit everyone. They benefit people who think differently disproportionately, because those people are currently absorbing the greatest cost of the gap between what organisations say they value and what they structurally reward.
Karp says people who think differently approach problems like artists. The data suggests this is accurate. The question is whether the organisation has built the conditions to use them.
Further reading:
↳ Productivity Is the Wrong Word — on the distinction between AI-for-speed and AI-for-capacity, and which human capacities most need protecting
Garden notes
- Overoptimisation — the mechanism by which organisations shed unmeasured capacity in pursuit of consistency, applied here to cognition itself
- The Homogeneity Trap — the parallel argument for AI labs: when everyone thinks alike, the frame becomes invisible as a frame
- Separated knowledge — the people who can see differently are separated from the conditions that would let that seeing do any work
- Productivity Is the Wrong Word — on which human capacities AI cannot replace, and why organisations are reaching for the wrong distinction