If AI is the answer, what was the question?
XML and violence
There this old joke: “XML is like violence: if it doesn't solve your problems, you didn’t use enough of it.”
When I came across this recently, it struck me as both funny and telling. Funny because XML feels like a relic of a simpler, more enlightened Internet civilisation. Telling because I’ve been rubbing uncomfortably against something of roughly this shape recently: the idea that when a technology causes problems, the solution is always more of it. I think we can do better.
All paths lead to middle management
In that post, Scott Werner uses the analogy of foxes hunting mice, then enlisting the help of hawks, leading to a surplus of mice, and so more foxes appearing—catapulting them straight into a middle-management economy.
The total number of problems doesn’t change, but now they are of the meetings-and-paperwork type instead of the mice-hunting type.
The argument is neat: resisting hawks (or AI) is pointless, so embrace them more deeply. Maybe the hawks should learn to take meeting minutes. But then we’ll need hawk-archivists too. And then vet bills start piling up.
It might be time we (and the foxes) start thinking hard about the question: to what end?
On pace
Unlike foxes, we don’t have to be passive participants in an evolutionary arms race. We can choose. We can slow down. We can slow down in some areas and speed up in others. We can go crazy and do neither.
Crucially, we can decide what to optimise for. We can decide what work we find worth doing, instead of functioning as human putty, filling in the gaps between tools.
Complexity is not neutral. It shapes the work we do, the jobs we have, and the kinds of lives those jobs enable. We must be clear about what kind of world our tool landscape is creating.
Fun-employment
I often come back to the idea that employment is the first and most important service any company provides. Coming up in the startup era of the 10s and early 20s, this was easy to believe.
Companies wrapped the basic fact of a salary in free lunches, on-site childcare, kombucha taps, and colourful gazillion-dollar offices designed like adult playgrounds. It was easy to believe that work itself—the act of employing people—was central, that enjoying work was central. The market had spoken, and the puritan fantasy of suffering gloriously through hard work was officially over.
But now, with layoffs treated as a mystical cleansing ritual and AI framed as a way to “do more with less,” that belief seem quaint, naive even.
Hidden costs
Markets shift. And when they do, companies will find that the talent they most need might not be interested. Still, none of this threatens their bottom line. Shareholders will still get their returns, executives will still collect their bonuses.
The cost is borne instead by everyone else, in the form of increasingly precarious, hollowed-out jobs that offer little security or dignity. It doesn’t destroy companies—it corrodes working life. And when enough of those jobs accumulate, the world itself becomes a little harder, a little meaner, a little worse to live in.
In a landscape where AI multiplies output endlessly, what happens to the meaning of work, to its shape, to its role in providing a structure in our lives?
On responsibility
This is why the project of AI at work can’t be left abstract. The people thinking, writing, and reading about AI are a subset—but an important, powerful subset nonetheless, with a large overlap with those working in the industries most affected by this shift.
These sectors are full of people with a measure of voice, leverage, and power. For our benefit, companies once built adult-playground offices and kombucha taps; perhaps in the next round we will harness this power to rethink the kind of working world we are choosing to sustain.
AI can speed us up, scale us out, and create efficiencies we could previously only dream of. But unless we seriously ask ourselves: “To what end?” we risk becoming spectators to our own systems, watching as complexity metastasises. Mistaking acceleration for progress.