Post

I don't use it but in my experience, every time a coworker uses copilot to create or comment on a PR it usually ends up adding an unnecessary and entirely wrong explanation of what the motivation/purpose of the change is.

I'll get it to do create short functions, like moving a file from S3 to Dropbox, see what methods it uses then go look up the actual definition for those methods and API calls etc. It's like picking apart a suggestion box.

I worked for a small/medium company whose CEO decided we needed to become the spearhead business in our industry. We were all tasked with brainstorming what that would mean and how we'd do this. I said that if we're going to structure ourselves around a metaphor, let's understand the metaphor.

I pointed out that a spearhead is all well and good, but for it to serve it's purpose, we need it to be driven forward by the rest of the spear, and at a target we want to hit. My way of saying: where are we going and what are we spearheading exactly? Understand that first, then we can talk 'how'.

But no. I was overextending the metaphor and being silly. Turns out what it meant was: throw millions into early generative AI R&D and wait years for a product that adds zero value, is a privacy nightmare and just burnt all the money.

Several years back I decided to stop copy-pasting code blocks as much as possible, I felt I never wanted to not understand any code that I worked on. That mentality and my complete rejection of AI coding have left me feeling like an old craftsman using hand tools to make good, honest, sturdy code 👨‍🌾

I have a policy of never copying anything from sites like Stackoverflow, even if I'm using it. It's better to write it yourself, then you have to think about every line and why/if it's actually needed in my case.

The reason why I'm not terribly worried about AI's impact on job security as a software engineer is because I see a lot of future work opportunity in cleaning up this dreck after the fact. Not that that's the line of work I'd *prefer,* mind you.

Given most orgs don’t bother doing it today, they’d rather apologise with a “we take security seriously” and another round of ceo bonuses

I've done my share of greenfield dev over a 25 year career, but I've mainly been a BAU/maintenance developer, working on 5-15+ year old codebases. If I stick at it I see myself being as valuable as the the Y2K COBOL devs in a few years time. AI can't even begin to debug and refactor like a human.

The problem is that they already have a lot of production code they don’t understand, and they’re never willing to address that base problem. COBOL coders get paid less than data scientists.

I do quality assurance on training and education resources, and while I'm confident my job is essentially safe fixing AI written content, some of the examples I've seen so far are so bad I didn't even know where to start. Just deciphering what it's trying to say in the first place can be a challenge

We've had actual orders come down from the board of execs that we should be using it more. Thankfully someone lower down had the idea to put together a working group to actually investigate where and to what extent AI might be actually useful to employees.

It's always like it's writing to a word count, maximising amount of words without saying anything of substance. At least the human errors I'm used to tend to be common things like they keep using passive voice or they don't know how punctuation works, I can work with that

This is exactly what they did with writers and copyeditors. It's an extention of the same grift. Editing generated slop always requires a full rewrite, but employers expect to pay the same rate as a polish. The end result is the same as with failures of automated labor: depressed wages for all.

A mark of a good developer is not writing more code than needed. Generating tons of fresh code rather than understanding what's there spells trouble.