“There are basically two ways to know you have a future. One, you have some vocational training. Or two, you’re neurodivergent.”
Alex Karp · Palantir · on TBPN, March 2026
He credits his company’s success to what he calls his lifelong struggle with dyslexia. Palantir also recruits through a “Neurodivergent Fellowship” that the company is careful to describe as something other than a diversity programme, and whose final interviews Karp runs himself. It has drawn more than two thousand applications.
I can’t stand the man. I can’t stand what his company builds, who it sells to, or the world it makes possible. Nobody is getting me to applaud a hiring programme presented as good news by a surveillance company.
And the line stays with me anyway, because half of it is true. The false half is exactly the one that costs the most.
The half that’s true
ADHD and AI go together extremely well. That isn’t a figure of speech, it’s what I have been living for three years — and what has accelerated exponentially since the start of spring 2026.
I have ADHD. Formally diagnosed, paperwork and all, and well before it became a topic on social media.
A brain that runs in twelve directions at once was always a handicap in a world where each direction cost three months of work. It becomes an advantage the day each direction costs an afternoon. What looked like scattering now looks like exploration. Nothing changed inside my head; what changed is the price of following an idea all the way through.
Corporate life, on the other hand, has always rewarded the opposite. It favours process, the people who write it and the people who apply it to the letter. Not taking risks is a reliable long-term career strategy there. If there are two things I cannot do, those are precisely them: playing internal politics, and following a process without trying to improve it.
Now, if AI replaces first whatever follows a written, documented, letter-of-the-law logic, then it is aimed straight at a significant share of corporate roles — and at first-line management, whose real work consists of checking that processes are being applied. The profile companies spent thirty years selecting for is the first one they automate. The profile they barely tolerated is the one taking off.
For scale: in four months I rebuilt my three professional sites and wired an invoicing ERP into them, shipped two sites whose features are advanced enough for an agency to bill for them, wrote a ten-chapter guide, and opened an internal portal of a dozen tools for sourcing, recruiting and copywriting that could each find a real audience.
This has to be said without reservation, because it’s true: it is absolutely extraordinary. I’m not going to open a piece about exhaustion by pretending I don’t know how lucky I am. The field of possibilities opened all at once, and behind every door I open I see twelve more. I did not get smarter in three years. I simply stopped hitting walls.
What the line leaves out
Before, the difficulty of execution acted as an external brake. You didn’t decide to abandon a project — you simply couldn’t get it done. Lack of time, lack of technical skill, lack of budget imposed a queue, a pace, and a set of automatic renunciations. That wasn’t discipline. It was physics.
AI removed the brake and put nothing in its place. The number of projects running in parallel is no longer limited by the capacity to produce. It is limited by the willingness to say no, which for almost everyone is the weakest resource available — and for an ADHD brain, the weakest of all.
Then add the nature of the work. Writing code by hand is roughly 80% mechanics and 20% judgement, and the mechanics are restful: they let the brain come down between two decisions. Directing an AI in a domain you don’t fully master is one hundred per cent judgement. No dead time. And on top of that sits a cost nobody counts: validating work you couldn’t have produced yourself. I am not a software architect. I still make architecture calls, accept technical trade-offs, and live with the consequences. On every iteration.
Output multiplied by ten, cognitive load multiplied by considerably more.
That is what the Palantir line leaves out. It promises a future to people who think fast and explore without a plan, in a world where producing is no longer the limiting factor. On that point it is right. The gain simply goes to the same people as the cost.
Because the limiting factor now is mental load. Not time, not skill, not budget: the number of projects a brain can hold open in the background. They don’t shut down when I close the laptop. They keep running — at night, during a meeting, in the middle of a conversation about something else entirely. And I can’t stop them.
The worst part is that it doesn’t converge. Every step forward on a project makes me develop a skill I didn’t have; that skill opens possibilities I couldn’t see; those possibilities produce ideas I would never have had. Moving forward doesn’t shorten the list. Moving forward lengthens it.
Shipping is not selling
And despite all of it, I ship. That’s where the problem gets interesting, because it doesn’t stop where I thought it did.
Shipping a project and selling a project are two different jobs.
I have tools that work. A personalised outreach system for LinkedIn, for instance, that goes well beyond the assistant built into the platform and leaves the user far more freedom. It runs locally, on the recruiter’s own machine: no candidate data leaves for the cloud. This isn’t a prototype. It works, and it works well.
Behind it sits a thesis I believe in: local AI, sovereignty over HR data, a way of using these systems that doesn’t amount to dumping candidate profiles with an American vendor to save eight minutes. And one very concrete consequence: because nothing leaves the workstation, this tool can clear a large company’s security review, where most of what’s on the market stops. It answers the question legal departments are already asking, and that almost nobody answers.
I need to correct myself before going further, or what follows turns into portfolio fantasy. Until someone has paid, it isn’t an asset — it’s a well-built hypothesis. I know the illusion, I’m living in it right now: when you build fast and well, every project looks obvious, unstoppable, ready. Nearly everything I see behind those doors would be invalidated by three real sales conversations.
So I don’t have ten companies in hand and a simple shortage of time. I have ten hypotheses, of which perhaps two stand up. And even those two have to be carried all the way to a client.
The channel closed at the same time
What costs isn’t the building. It’s the last mile. Going to see someone, understanding their situation, explaining why this changes something for them, absorbing their objections, coming back. That trip doesn’t compress. It can’t be automated, it can’t be prompted, it is made of human time and presence, and I have a fixed amount of both.
For years, LinkedIn did part of that trip for me. Publishing regularly was enough to bring clients in. Content was an acquisition channel.
That is over. LinkedIn has become an influence algorithm rather than a commercial link. It manufactures visibility rather than encounters, and it has decoupled itself from the moment you publish. Content still builds a reputation; it no longer fills a calendar.
The result: prospecting has become a full-time job again, at precisely the moment when I have the least time.
And between knowing how to build tools and making it known that you have them, there is an entire profession. Presenting them, positioning them, selling them, supporting the clients who use them, keeping versions current, answering when something breaks. A product isn’t a delivery, it’s a commitment that starts over every month. When there’s nobody to hand it to, that stops being a workload: it becomes a brake on releasing anything new.
And I have to be honest here, because this is the knot and it isn’t flattering: it isn’t only that I don’t have the time. It’s that in parallel, I want to build. Running a campaign, following up, doing sales — I know how, I did it for twenty years, and today I’m choosing something else without admitting it to myself.
What it costs, and the word I keep avoiding
An unlaunched project doesn’t cost you its construction. It costs you its upkeep.
Every object finished but never brought to market stays open somewhere. You have to remember it, keep it current, think about it again, explain it again, justify not having released it. Twelve tools built and unlaunched are not twelve successes waiting: they are twelve open loops consuming attention permanently, for free, without producing a cent. Building has become cheap. Maintaining the unfinished has stayed exactly as expensive as it was.
There used to be rest inside the work. Commutes, dead time, slow tasks during which the brain came back down. Those windows are gone, and nothing replaced them.
So I’ll write the word rather than circle it: I’m being careful not to walk into burnout. I’m not there. I know how to recognise the slope, and I’d rather write it here, cold, than have to explain it later.
There’s one more thing, less comfortable than the rest. I spend part of my time building tools that make finding people faster, broader, more automatic. This is the profession I’ve practised for twenty years, and the parts of it I love are the most artisanal ones. I expected these tools to kill the pleasure. In fact it was already dead before they arrived: I have a machine for finding people, and no hour left to go and talk to them.
It isn’t automation that damages the craft. It’s saturation.
Surrounded, yes. Managing people, no.
The logical conclusion would be to hire. It doesn’t suit me, and that isn’t a whim.
I don’t want employees. I have no desire to manage people, to run annual reviews, to carry somebody’s salary in a month when a client pays late. Employment isn’t the right vehicle for what I’m describing, and a holding company stacking stakes in projects nobody pushes isn’t either.
What I want is an ecosystem. A collective. A guild, to say it the way I actually think it: people whose personal interests converge and who decide to work together, in good faith, with clear rules, and where everyone gets their share.
People who take a piece and genuinely carry it, with a direct stake in the outcome. Someone whose actual job is selling the training, not enduring it between two assignments. Someone who takes a tool and turns it into a product, with a price, support and customers. Each of them owning their piece, not executing mine.
And me as one link in that chain, not at its head: seeing, building, passing on. That’s where I’m good, that’s where I’m hard to replace, and that’s probably where I should stay.
I know this is likely still utopian. The dominant model remains the employer and the employee who does as they’re told, and it dominates because it provides a frame, a hierarchy, someone to ask. A guild has none of that: it holds together on rules set in advance and on everyone having more to gain by staying than by leaving. It’s more fragile. It’s the only shape I want to work in.
Few people have done the work
This is where I’m genuinely stuck.
It would take a double competence: understanding recruiting from the inside, and understanding what AI actually changes — not having heard about it, having practised it long enough to know what it gets wrong.
I already know this pattern; I’ve watched it for ten years in my own field. There is no doubt that mastering LinkedIn Recruiter, taking the time to understand the tool, making the effort to go further than the interface suggests, produces considerable results. I built a training programme out of it, and the results are there. And yet the people willing to put in that effort remain few. You only have to look at the rise of magic tools, which promise the result without the work — and which sell extremely well.
AI follows exactly the same curve. Doing what I do isn’t within everyone’s reach, but it is within many people’s reach. What’s missing isn’t technical capability: it’s perspective on the profession, which comes only from experience, and an understanding of the systems and what’s at stake in them — where the data goes, who owns it, how long a model stays available and at what price. That takes genuine interest.
In my circle, few people have done that work. Plenty have vibe-coded things here and there. The ones who have really gone deep, I know very few.
And knowing how to build a house doesn’t mean knowing how to make it desirable. I can make the tool; I can’t give it a name that sticks, a page that makes you want it, a price that holds up, a story an HR director repeats to their committee. That isn’t false modesty, it’s an assessment of my own skills: I build something sound and it stays invisible.
Creating is not succeeding
Creating is what I love. Creating is also what exhausts me. And creating is not succeeding.
It takes a set of people, a grouping. My skills may be turning into expertise — I don’t know, and it isn’t for me to say — but expertise needs an environment to deploy in and to flourish. I don’t have that environment yet.
And the uncomfortable part has to be said immediately: this advantage may be temporary. Nothing says a model released six months from now won’t do everything I do — including the decisions I make, which are the real work — better, from a single simple request. I can’t predict it, and I’m not going to pretend otherwise.
So if the CEO of Palantir claims you have to work with your hands or be neurodivergent to have a future, here are the observations of a neurodivergent. Corporate was never built for us: it rewards process, caution and internal politics, which is precisely what we do worst. The internal opportunities he promises don’t exist yet. What works isn’t being neurodivergent. It’s being surrounded. And protecting yourself.