Where Taste Came From

Nothing is wrong. That’s the part I can’t get used to.

Three implementations of the same function, open in three tabs, and the suite is green under all of them. Not green-ish. Green. The agent produced them inside a minute, politely, the way it produces everything, and now they sit there being correct at me.

One is compact to the point of showing off. One is what a careful person writes at four in the afternoon. One does something with a lookup table that I don’t love and can’t argue with.

There is no bug to find. There is no failing case to chase. There is only the question of which one goes in — and the slow discovery that the whole job has quietly relocated into that question, which no one ever trained me for. They trained me to make.

Pick one, I think, in the voice I use on myself when I’m stalling. They all work. Just pick one. And I don’t, because some part of me that has been doing this a long time knows the three are not the same, even though every instrument on the desk says they are.

When correctness was the expensive part

For most of the years I’ve been doing this, correctness was the scarce good and taste was a garnish. You fought the thing until it worked. If it worked, it shipped, and whether it was the right correct thing was a conversation for a quieter quarter that never arrived.

There was a stretch where I was the one building the thing, and a longer stretch after it where I was the one who had to live with whatever got chosen. One taught me what a decision costs to build. The other taught me what it costs to keep.

But the making was the bottleneck for so long that the choosing barely registered as work. It looked like the thing you did on the way to the work — a doorway, not a room. You stood in it for a moment and then got on with the part that was hard.

Now the making costs a minute and the choosing costs the afternoon.

The scarcity that used to decide for me

When one option is expensive to produce, judgment barely matters. You take the one you’ve got and you’re grateful for it. Scarcity does the deciding on your behalf, and it feels like rigor because it hurt.

When ten good options arrive in a second, the scarcity is gone, and the deciding comes back to you — all of it, at once, with no friction left to hide inside.

Which of these will the engineer who inherits it understand at eleven at night, six months from now, with the pager going? Which one bends when the requirement changes in March, and which one snaps? Which one is clever in a way I’ll be quietly embarrassed by? Which one is boring in the way good infrastructure is boring?

The tests answer none of it. The tests only know that all three work.

The tuition

Taste is trainable, but not the way a skill is trainable. There’s no curriculum for it, and there’s no prompt for it, which is the first thing everyone tries.

It comes from exposure to consequence. From having shipped the clever thing and then maintained it for two years, watching each new person on the team walk into the same wall in the same week of their onboarding. From having picked the wrong abstraction in a planning meeting where you sounded genuinely convincing, and then living inside that decision for eighteen months while it charged you rent.

And a lot of it came from the other chair. Much of what I know about which correct thing is the right one, I learned across a table rather than at a keyboard — watching a decision I’d signed off on come back a year later wearing a different name. A support queue that never drained. A migration nobody had budgeted. A good engineer who quietly stopped volunteering for anything that touched that subsystem, and who left eight months later for reasons his exit interview recorded as personal.

That’s all taste is. Compressed consequence. Every fast judgment I make now is a slow, expensive thing that happened to me once.

Every fast judgment I make now is a slow, expensive thing that happened to me once.

Which brings me to the part I find genuinely uncomfortable.

The slow path is what built the taste, and the agent has just removed the slow path. Nobody will spend three years maintaining their own clever decision when the clever decision takes four minutes and the maintenance goes to something that doesn’t mind doing it. The output arrives without the consequence attached. And the consequence was the entire tuition.

I don’t have a clean answer to that. I notice I’m suspicious of anyone who does.

What I tell the seniors

When someone brings me this — and it is most of what I get asked about now — reassurance is useless. Reassurance bounces off a person who can already see the shape of the thing. What lands is the accounting.

Their judgment did not get less valuable. It got more valuable, and it happened fast enough that nobody re-priced it out loud, so it still feels to them like the thing being taken.

I try to be careful here, because the doom version of this is easy and mostly wrong. The work didn’t vanish. It’s coming back, if anything — but it’s coming back weighted toward the people senior enough to direct it rather than only produce it. What’s thinning isn’t the market for engineers. It’s the market for the part of engineering that was only execution, and that part was never where their value lived. It was just the part that was easiest to see.

What they’re grieving is real, and it isn’t their relevance. It’s the proof. The code was the part you could point at. It had a line count. You could show it to someone at dinner who didn’t understand any of it and they’d still see that you had made a thing that day.

They spent a career believing the code was the asset. The code was the receipt.

That’s a harder thing to carry into a performance review, and I won’t pretend otherwise. But it’s the true version, and people can work with a true version. They can’t work with a comforting one.

The lean

I pick the second one.

Not because I can fully defend it. I could build a decent case for the first, and if someone I respected pushed back hard I might fold on the lookup table. But something leans — built out of every decision I ever had to live with afterwards — and it points at the second tab.

It isn’t the typing that does that. The typing left a year and a half ago, and I miss it less than I expected to. It’s the lean.

And I paid for the lean without noticing I was paying. Every wrong abstraction. Every clever thing I had to keep alive afterwards. Every decision that came back a year later wearing a different name and a bigger number. It arrived slowly enough that I mistook the invoice for experience.

The machine gets to the edge of the decision, sets three good options down where I can see them, and waits for someone who has been billed.

I keep wondering who that is going to be in ten years.