A widely circulated quote categorises the oasis of safety for workers in the face of Artificial Intelligence (AI) advances as having a binary existence: vocational, hands-on trades, whose highly contextual, real-world practical work operates beyond the means of natural language systems and the current capabilities of robotics; and non-linear, neurodivergent thinkers, ranging from those with ADHD and autism to dyslexia, whose minds operate outside a conventional playbook. Perhaps ironically, these are the very minds a rigid schooling system would throw into a special educational needs class.
These are the core observations about the working economy regularly made by Palantir CEO Alex Karp, and unlike most widely circulated quotes, this one has the merit of truth. On a daily commute into London, Karp’s words resonate meaningfully against the backdrop of an advisory industry urgently needing to adapt to and integrate the multitude of AI models and tools now available.
When the last great disruption of social media swept the industry, it centred on the basic dynamic of communications no longer being a one-way street (broadcast out, publish an article, distribute a statement), but a two-way revolution where stakeholders of all guises engaged in conversation - whether you liked it or not. In turn, organisations had to prepare themselves for radical transparency, where information could be shared as easily as hitting publish on this Substack article.
AI takes this a generational leap forward. The mass ingestion of the world’s knowledge online, combined with siloed training on highly specific datasets within companies, provides the perfect context for automation, pattern detection and predictive analysis - but above all, relentless speed. All of this is backed by a nonsensical business model where it’s possible to purchase £4,000 worth of compute for a mere £100. Horror stories of CFOs receiving budget-crushing invoices of thousands are still discount tokens - literally - as AI companies race to maintain frontier status and achieve superintelligence.
It’s the immediate cognitive impact for workers that matters
As a reader of this Substack, you likely know all this. It’s the psychological impact of the in-between state that has weighed on my mind lately. How should information workers across the world’s cities manage with an ever-increasingly smart copilot sitting by their side?
As a 14-year-old, I found solace against a Christian evangelical upbringing in the late polemicist, journalist and columnist Christopher Hitchens, starting with his book Letters to a Young Contrarian. In it, he wrote that the essence of the independent mind lies “not in what it thinks, but in how it thinks”.
Hitchens cared a great deal about his conclusions and defended them fiercely, but he thought a view was only worth holding if you had reasoned your way to it. He kept coming back to this idea, closing Why Orwell Matters with it and repeating it in God Is Not Great and his memoir, Hitch-22. “Contrarian” suggests someone who looks for an intellectual fight, hence Hitchens’s publicly known distaste for the publisher’s choice of title. But that isn’t the goal. The book, through a series of letters to a student, encourages independent thought and freedom of enquiry.
In professional advisory, as in life, this has never mattered more, with the ever-tempting ability to offload our thinking to AI. Our experience and our advice, backed by data, are what businesses rely on. That only works if the thinking behind the insight is our own, shaped by the situation it’s applied to.
The research on offloading is still early, but it is worth reading. AI use has been linked to weaker critical thinking. In a survey of 666 people, published in January 2025, Michael Gerlich of SBS Swiss Business School found that frequent use of AI tools was associated with lower critical-thinking scores, with cognitive offloading explaining much of the relationship. Younger participants used AI most and scored lowest. These are correlations that warrant careful examination in everyday work. They also raise the question of whether people who are less inclined towards critical thinking are more drawn to AI, rather than AI use itself being the cause.
How dated those days feel: wandering university libraries - often early in the morning, in my case - among a physical forest of books. Framing an essay, structuring an argument or pursuing a line of enquiry to find your own authorial voice took time. I reminisce about some of my essays, where a lecturer would remark upon how I had reasoned my way to a particular conclusion - on some occasions better than others. With high AI usage, dare I say undisclosed in academic settings, can this even happen anymore?
A 2025 MIT Media Lab study led by Nataliya Kosmyna recorded the brain activity of 54 students writing essays with ChatGPT, a search engine or no tools. The ChatGPT group showed the weakest brain connectivity, reported the least ownership of their work and had the weakest recall of it. The authors describe the potential accumulation of these effects as “cognitive debt”. The findings are preliminary, and the study has not been peer-reviewed, but they raise questions worth asking about how we learn to think. A friend of mine who is a university professor described the error-ridden performance of a student challenged over undisclosed AI use, who failed to accurately explain the meaning of an archaic word brought back into the English language.
This challenge is older than the technology
The philosopher John Stuart Mill made the case for reading beyond your own side in On Liberty in 1859: “He who knows only his own side of the case, knows little of that.” Hitchens, and many others, gave me the other side of a great many cases, teaching me scepticism: the habit of asking how to approach a subject and whether something is actually right.
In Plato’s Phaedrus, written around 370 BC, the Egyptian king Thamus warns that the newly invented art of writing “will produce forgetfulness in the minds of those who learn to use it” and give pupils “the appearance of wisdom, not true wisdom”. Writing turned out rather well for humanity, so the comparison is imperfect, but Thamus’s distinction between looking wise and being wise is exactly what an AI-drafted report puts to the test.
In his 1784 essay What Is Enlightenment?, Immanuel Kant described enlightenment as a way out of self-imposed immaturity: the inability to use your own understanding without someone else’s guidance. His motto was Sapere aude: “Have courage to use your own understanding”. Among his examples of convenient guides were a book that does your understanding for you and a pastor who does your conscience for you. AI is a far more convenient guide than either.
Applied to AI, these thinkers point to a few working habits. Ask what each claim in an AI answer rests on, and check the ones that matter. Ask the tool for the strongest case against its own answer, as Mill would want. Read beyond what it gives you, and draft your own view before you ask for its.
I value the Enlightenment and what it means for how we think, and I want to make sure I don’t lose my own identity, my own way of thinking, to the AI tools I now use every day. The pressure this time is subtler, because it comes from a tool that is always available and very easy to lean on.
The truth of the matter is that undirected and decontextualised AI usage is akin to airline journalism, the kind of writing you find in an in-flight magazine. It has to cater for all and be inoffensive, so it tends to hand back an average of what people think, in an unapologetic way that an untrained readership would likely miss. In strategic advisory, the consensus is rarely where the value lies. The value is in creative ideas that sit outside the obvious, that understand the psychology of human behaviour in ways the machine struggles to, and that are led by experience.
Standing behind our advice matters more than ever
I’m not arguing that people should avoid AI. In business, it is important to use the technology in front of us, and telling a team to avoid AI would be like telling them to use a library instead of the internet. The risk is in how we adopt it. If AI is rolled out before people have adapted culturally and can work with it confidently, I think it’s a recipe for disaster. Writing is thinking, and many of us find our conclusions by playing with an idea until it holds up. AI can help with that, but only if the person is still doing the thinking.
We all have to have values, and the most important one here is that we stand behind every idea we put in front of a business - the ones we lay on the table and put our names to. Communications teams will increasingly be asked to stand behind conclusions that AI helped to produce. The test will be whether we can show what we checked, what we couldn’t check, and where our own judgement and experience took over.
If you’re working through any of this in your own organisation, the comments are open, and so is my inbox.