Relatedly, if we take the conclusion at face value, that companies who adopt AI end up growing headcount way faster than those who don't adopt AI... doesn't that mean those companies now have to pay for more people... and a massive token bill... as compared to their fast-growing peers that didn't? Doesn't this mean they're each doing less with more?

The effect shown in the study is that costs of grew along two dimensions without demonstrating a benefit. Not that there isn't one, just that the study didn't measure one.

I'd love them to index to growth in something other than the number of people in what is otherwise a cost center e.g. revenue. That would help us find out if this is good or, as a cursory read may appear, double-bad.

It would be interesting to compare as you say the median wage change, aggregate wages paid and change in business metrics (ideally revenue and/or profit margins) before and after adoption.

It seems like everyone's trying to have it all ways: AI will lead to more efficiency but also more jobs, more revenue for businesses and more massive bills to pay for the trillion dollars we've put into AI so far. That doesn't... add up. Where's the profit go?

"Help me balance this budget" meme but "oops! all candles."

We've gone from lines of code in a codebase being a liability, and headcount being bloat, to deploy the slop cannons and load up on cannoneers. I suspect we'll learn... something... I just can't quite put my finger on it. Ah well, who wants to go play with Fable-but-coding-tasks-get-routed-to-Opus?

To get onto Hacker News and other social media sites for marketing exposure?

The current narrative is that AI is displacing jobs.

Similar to the current post there’s just not enough data to be conclusive.

Bingo. Bullshit work grows to accommodate productivity gains.

Is it easier to implement a design now? Then let’s redesign every quarter.

Is it easier to refactor the whole codebase? Then let’s rewrite in whatever new hotness.

This is not different from web frameworks 10-20 years ago. Is it easier to make a website using Rails? If so, why did I see an explosion in devs-per-project after I started using Rails? Because of the BS.

One of the study authors, Ara, tweeted that they control for that by comparing early adopters against firms who haven't yet, and built like-for-like control groups with similar pre-adoption growth trajectories. (This is a rough quote).

It's not entirely clear to me, but I believe this is 10% over the control. The 10% is therefore relative to "regular headcount growth".

The claim is and always has been that automation increases net jobs. There's never been an automation revolution that worked otherwise.

The counter claim that LLMs were special for some reason has never been supported. And a lot of its proponents have other axes to grind, like UBI, Socialism, Communism etc. It would suit their worldview for this time to be different, and so thats the message they push.

Inside can be as well. Please manually review these 5000 lines of slop I generated while watching YouTube and verify the unreadable readme due to all the bloat and icons is correct with the code.

If that is not soul crushing I would not know what is.