Unartificial Intelligence

No one signed off on it. Yet somewhere in the last year, the hour and a half I spend reading AI’s output line by line before I put my name on it has become the most expensive thing in the room – more than the compute that built it in a third of the time. I have been unearthing what that hour and a half underwrites.

Unseated

Wage labor, licensed professions, and schooling rose to prominence across the 1800s, displacing a world in which pay (what one earns), status (where one stands) and meaning (what one is for) were settled by station: the land your family held, the rank of your name, the guild that would admit you. In this new world, you could outlearn your birth: what you know and can execute is eligible to re-settle all three, and two centuries allows for an arrangement to become our natural order.

Competence has gone cheap, and the three prizes are coming loose from it as they had from station.

I tried to sketch out what holds value next: eight simulation rounds alongside an AI model, each of us forecasting the answer, both sealed before comparing. Round after round the model reached for what institutions can verify (capital, license, governance), and I for what only time can (trust, attention, presence), each of us betting on the half native to us.

What sits on neither of our lists is competence alone, and where we agreed was the rule that exposes it: a field’s repricing comes down to how long a capable stranger would take to learn its work and what contracts are in place. What you could only have accrued by wanting to be somewhere a long time – the places you’ve walked with your five senses, the client who calls at 11pm because of what you spotted in 2022 – will be around the longest.

Whatever the three prizes settle on, it will not be competence as we were raised around. Nobody had to ask what competence was evidence of, because it never arrived alone: someone had wanted a thing, kept wanting it through the stretch where it paid little, and stayed long enough to be trusted. Wanting takes years to underwrite in a person, competence an afternoon. Every knowledge profession set aside the years for this, when someone senior signed for your work until they didn’t have to, and then until someone else’s work went out with your name beneath. That signature made wanting legible. What that hour and a half underwrites is a name that has wanted long enough for influence to grow roots in it.

Unmanufactured

Up until this point, we’re circling within familiar boundaries, yet the boundaries are what’s moving under our feet. (No one in 1950 could have ranked the “software engineer,” the “index fund,” or anyone in 2000 the “influencer.”) Cheap cognitive power is setting loose things with no present-day silhouette: a lab running thousands of variations overnight and waking its scientist with the two worth their bet; a university turning lecture series into an interactive atlas of a subject; every inbound teaser diligenced the morning it lands, so the first call is already a second meeting. And at the foundation of them all, something that anyone a year into working with a model has already started building: a bilateral relationship. The model and you learn each side’s blind spots and reflexes until the pair reaches a depth no one could have read their way to, and a transparency into each other no one else is offered.

How the relationship deepens, from asking questions to letting it act with agency. Read from the bottom, it doubles as an operational plan.

the loopwhat it holds of youwhat you’ve learned of itwhat breaks here

Beyond you

7Self-directeda deputy running a team you no longer checkYou set outcome and limits; it splits, runs and reviews the workA rubric: your judgment, written as a testWhere its own checks can’t be trustedToo fast and deep to check; only models can review models

to climbVerification at scale: checking outruns doing, so the name on the work becomes an institution’s

6Entrusted agencya chief of staff answering in your nameActs alone within your limits; you see results, not stepsWhich tasks it may run in your nameWhich decisions you’d let it make, and which neverNo one has agreed who answers when it’s wrong

to climbLiability: nobody will say which tasks it may run alone, so none run alone

Across sessions

5Track recorda founding pair with one track recordWhat the pair made that neither could alone, and how it held upYour calls, its objections, and how each turned outWhere its work is yours now, where you still hold the material yourselfCredit blurs: no one can tell its calls from yours, so the record can’t be priced

to climbHonesty: writing down which parts were its, in a place others can read

4Standing correctionsthe assistant who has learned your styleCorrections become rules it keeps; it corrects you back by your standardsYour blind spots, preferences, and how you decideIts tells and failure points; that it stops when told, and calls you out when you’re lazyIt forgets a rule, or keeps one that no longer fits

to climbMemory: somewhere to keep corrections, the habit of writing them down, and in many firms, permission to keep anything

Inside a session

3Delegationthe intern’s task for the dayYou hand over a task and review the result; it ends with the sessionThe brief, the deadline, what done looks likeIts reflexes: drifting to the average, padding, folding when pushedCorrections don’t carry: it repeats the same mistakes every session

to climbTrust: checking its work costs more than doing yours, until you learn where to look

2Contexta temp handed the briefing binderOne way: you feed it material, it works from thatThe material for one project: files, terms, the threadWhat it does with context, and what it can’t inferMaterial doesn’t carry: each new project, you gather and feed it again

to climbThe material: it lives in heads, inboxes and habit, and someone has to gather it

1Questionsasking the reference librarianOne question, one answer; nothing carries overOnly this turn’s questionWhat it knows, and what it invents just as confidentlyIt answers the question, not your situation; a search does as well

Under/writtenunder-written.comSave imagePortraitLandscape

© 2026 Underwritten. Free to share with credit and a link.

Beneath these changes, two jobs once bundled as intelligence are coming apart. Our working mind has done both at once: hold the material, and form an angle on it, including edge cases. The first job is now mass production, and goes the way of arithmetic by hand or fire from flint. The second cannot. An angle steers the width of that production toward a call that can be right or wrong. Wanting does the aiming.

Split between a person and a model, those halves add up to more than one worker today, because each lacks what the other brings: the model, width with no aim of its own; the person, an angle held back by how much one mind can carry. When securities trading went from floor to screen, capital market rebuilt its choreography. Much of a working day’s shape has been constructed around that first job. Soon you’ll wake up to days shaped around the angle.

Unpriced

That pair still has no term nor KPI, no arrow nor box on an org chart. Newest forms wear old names first. The car was a horseless carriage; “assistant” and “copilot” are the same move, defining what’s new as an old kind of help, better equipped within the same frame.

AI brings output at a tempo no human can match, and in time it can map the next steps about as well as anybody. The human has seen enough steps through to stand by their view under turbulence, compounding into a name others learn to trust. Next to be priced are track records that neither could have achieved by force alone, and being the person who coins the standard.

---

What outlasts even the want, then? I’d expect most people to find what they’re for in stewardship: a place, a tradition, a group of people. I expect to also find it in craft: a piece well made, with one’s name, for its own sake.