AI is removing the middle class of software engineering

The title of the linked-to blog post is true, and the contents are true, they just have no explicit, cause-and-effect relationship. The title has been the desire of the management class for at least 30 years that I'm aware of. Microsoft promised managers that software engineering would pretty much go by the wayside with the introduction of Windows NT circa 1994. There was next to nothing about this in the official, printed, bumpf that went with NT, but in person, Microsoft salespeople promised managers huge programmer productivity gains, 5x-10x, and NT would make all programmers plug-compatible. Managers could just hire and fire downtrodden Windows code monkeys. This turned out to be marketing gibberish, and Microsoft lies, but back then, managers believed it.

In 1998, I worked for US WEST, one of the "Baby Bells". The Communication Workers of America struck, so suddenly, I was in management, and had to Loop Assignment. As soon as the strike was over, I wasn't in management and I had to give back my corporate AmEx card. The manager class is really uneasy about sharing power with a new group, they already have to deal with lawyers, and they hate that. The last 30 years of having to listen to software engineers as a category, and take them seriously, has been traumatic. The title of this article describes the end result of many, many attempts to stuff the genie of a new non-worker category of employees back in the lamp.

You don't have to believe me on any of this, but that doesn't make it less true.

You were never expected to understand every service and every database. But at least someone did and would explain it to you.

I think to companies, this is a potentially desirable outcome of LLM-driven software development; instead of knowledge, understanding and accountability about the software system being located in one or more employees, who have to be retained, listened to, and/or have replacements trained, LLMs potentially offer the perspective of storing this knowledge digitally, turning it into intellectual property that is entirely owned by the company.

The main problem with this idea is that coding tools as they exist today are sycophantic lackeys entirely under the control of developers who, as the author points out, are incentivized to game the worst kind of productivity metrics.

This is just another example of the optimism bias that causes people to construct a comforting narrative where AI takes everybody else's job, but not their own.

I think the article is probably right, but also a lot of very smart people are getting into programming because of 'vibe coding' and learning at a tremendous rate.

Programming had a pretty high bar of entry. I certainly remember many decades ago wanting to give in so much fighting basic syntax errors on my projects (self teaching myself from books).

It took a lot of persistence to break through that. "Vibe coding" lets you learn very differently, everything is "too easy" until you end up with a slop mountain. But I have noticed curious and smart people really picking up the concepts quickly. It's like learning the opposite way down.

I've no real idea how this will turn out long term though.