AI, craft, and what's left when implementation is cheap
There's a pattern that repeats every time a craft gets a powerful new tool. Photography didn't kill painting, but it killed portrait likeness as a paid skill — and painting responded by becoming about everything a camera couldn't do. Desktop publishing didn't kill graphic design, but it killed "can operate a phototypesetter" as a career. The tool commoditizes the mechanical layer of the craft, and the craft retreats — or advances — to the layer the tool can't reach.
AI is doing this to building software, and I think it's worth being precise about which layer is actually being commoditized. It's not "development." It's something narrower: average execution of well-understood things.
What's getting cheap
A competent CRUD app. A landing page that looks like landing pages look. A React component that does what React components do. Code in the shape of all the code that came before it — which is most code, most days.
This was already cheap, honestly — templates, themes, and Stack Overflow commoditized it years ago. AI just collapsed the remaining cost toward zero and removed the assembly labor. Anyone can now produce the median artifact of any digital discipline: median copy, median layout, median code. The floor rose overnight.
The uncomfortable part for people in my line of work: a lot of what clients used to pay for was the median artifact. If your value was "I can build the normal thing," the market for that is going where the market for portrait likeness went.
What's getting expensive
Everything that made the median the median.
Specificity. AI outputs converge on the average of their training — the most statistically likely design, phrase, structure. That's what "AI-generated" looks like: not wrong, just unplaced, belonging to nobody. Meanwhile the entire job of brand, of design, of voice, is to be unmistakably somebody's. When average is free, the premium on particular goes up. The warm near-black and the italic eyebrows on this site are worth more, not less, in a sea of competent defaults.
Judgment. Someone still decides what to build, what to cut, what "good" means for this client, this audience, this constraint. AI widens the option space enormously — it will happily generate ten plausible directions — and that makes the ability to choose the bottleneck. Curation was always part of taste; now it's most of it.
Knowing when it's wrong. AI produces confident output at every quality level, and the output looks the same whether it's right or subtly broken. The person who can tell the difference — in code, in design, in copy — becomes the load-bearing element of the whole workflow. Craft used to be the hands; now it's mostly the eye. The eye was always the harder part to train.
The feel layer. The last few percent of any interface — timing, weight, friction, the difference between "works" and "feels right" — lives below language. You can't fully specify it in a prompt because it isn't fully specifiable in words; it's found by hands on the thing. That layer was always where craft lived, and it's the layer the tool reaches last.
The honest risk
I don't want to write the comfortable version of this essay, where craftspeople are safely irreplaceable. The tools will keep climbing, and some of what I called "expensive" above will get cheaper too. Anyone who tells you they know where the line settles is guessing.
But the direction of the pressure seems clear: AI punishes the generic and rewards the specific, because the generic is exactly what it automates. The developers and designers who were coasting on "can produce the normal thing" are in trouble. The ones whose work you could identify without a signature are, if anything, more visible than before — because the background noise they stand out against just got infinitely louder.
What I'm doing about it
Practically: I use AI for everything mechanical and nothing directional — the full inventory is in how I actually use AI. Strategically: I'm investing in the layers that resist automation — taste, judgment, the willingness to make things that look like me and not like the average.
The craft isn't dying. The definition of it is moving, the way it always has — away from what tools can do, toward what someone has to decide. That part of the job has never been more of the job.