On your boss's ledger, you get more expensive every year. AI gets cheaper.

A while back I was helping a friend fix up his resume, and on the side I scrolled through my own raise history for the last several years. The first few years the line still climbed. The last two, it's basically flat.

My first reaction was that the market's just bad, wait it out.

Then I ran a different set of numbers, and it stopped looking that simple. The problem might not be the market. It might be a sheet I'd never really looked at.

First, what that sheet is.

Every company has one: a cost sheet the boss holds, line by line, showing where the money goes. Rent, utilities, servers, the cloud bill, everyone's salary and benefits. I'm not a stranger to this kind of sheet; running projects, you brush up against it. What the boss does every day is stare at that sheet and push each line to spend less and produce more.

On that sheet, you are one line inside "labor cost." Your skills, your experience, your overtime all get boiled down to the number on that line.

In the last two years a new line showed up, called "compute," or "AI calls." The company uses large models to write code, run support, process data, and that spend lands on this line.

Here's the part that stings: those two lines move in opposite directions. One is going up. One is going down.

Look at the AI line first.

The VC firm a16z ran the math in 2024: for a model of equivalent capability, the cost to call it drops by roughly 10x per year. Their example: GPT-3-level performance cost 60 dollars per million tokens in late 2021; three years later a small model hitting the same quality cost 6 cents. A thousandfold drop. Even if you only count the high end that arrived with GPT-4, it got about 62x cheaper in two years.

Now look at your line.

In IEEE-USA's annual survey, US tech salary growth in 2025 was 1.6%, the lowest in fifteen years. It was 2.9% in 2024, 3.5% in 2023, lower each year. Dig further and senior generalist developer roles were actually down 10% year over year. China doesn't have stats this granular, but you and I both know what the last two years have felt like.

One line gets ten times cheaper a year, the other goes up a point and change. The boss makes the most ordinary cost decision off that sheet, and the scale keeps tipping onto your side.

He doesn't even need to be targeting you.

At this point the move most engineers reach for is: I'll grind my skills harder, hard enough to be irreplaceable.

That's not wrong as a direction, but it doesn't get you off the sheet.

Grinding skills means climbing up the "per head, per hour" line. Climb as high as you want, you're still a number on that line. The ceiling above you is how that line is priced: the company pays for the time you occupy, and time is getting squeezed cheaper by the row next to it. There are only so many hours in a day, and the price per hour is under pressure. Both ends are pinned.

Skills matter, of course. They're what gets you a seat at the table. But pushing only on this line means you're fighting the direction of the whole line, not one bad year.

It's like running up an escalator that's going down. You're faster than the others, and you're still descending.

That IEEE data already says half of it: roles that involve AI or platform-building are going up, purely generalist roles are shrinking. Generalist skill isn't worthless. The same craft, hung on a different pricing method, comes out very differently.

Switching industries doesn't fix it. What you change is how you get paid.

Billed per head, per hour, per month, and you're on the line that's being squeezed. Billed by a finished thing, a system that runs, one complete delivery, one concrete problem solved, and you've moved to a different line. That line is priced by what the thing is worth, not by how many hours you put in.

This doesn't require quitting and betting the house. Start small: take a side gig, and when you quote it, don't write "X per day," write "Y for the whole thing." That one change, the client's reaction and your own headspace both shift.

From here, watch one number: of everything you took in this year, how much was billed by the hour or the month, and how much by the project or the result. Which way that ratio moves tells you more about which line you're standing on than whether you got a raise this year.