A few things I saw which give me questionable culture vibes. Almost makes me chuckle. I’m happily on the skeptical side of AI use in software and express concerns over it.
People first, AI second
Repo A exists in a company. Team works on it. Has absolutely zero documentation. Types are non-existent. Zero markdown files. The one README.md that is, contains invalid info.
Then comes LLMs. Teams adopt it. Lo and behold, motorway-length markdown files come to be, for the LLM.
Similarly, jira tickets would have no description. With integration of skills, we start getting some details. Good that we do, but that begs the question, “Why wasn’t it being done before for humans”? Why only start adding it for an LLM?
If you want to explore a doc-less repo and spaghetti code, using an LLM is a crutch. The team should have first fixed up the repo to be suitable for humans first, AI second.
AI isn’t there to fix fundamentally broken processes. That part is for humans to do.
Reading code is the hardest part
It is collectively agreed upon that reading code is harder than writing. Yet the slop machines spit out code like no other. Majority of the time it’s more complicated than it needs to be.
And yet, this gets easily solved by people saying “oh yeah I glance at the output to make sure it’s good”.
How? I kid you not, there will be instances where the slop output is going to have some things that you do not fully know. And believe me, you will not be going to the internet to confirm if it is actually correct or not. You will just test and if it works, push.
You either know, or you don’t. Writing code that went through your thought process makes sense. Glancing at slop output is not a trivial task and requires further cognitive load to validate it, which one tends to brush aside.
Agreeing with AI
There was a discussion I was part of where I proposed a certain way of doing something (with tons of arguments to validate my claim). I got pushback to do another way (with very weak arguments). Later on in the PR, their agent chimed in and said to do the thing my way. Then I was finally given that yeah, probably do it your way. You’re right.
That is a very low bar of quality to set (stolen from Andrew Kelley). An AI agreeing with me does not validate my arguments. The arguments themselves hold weight. They don’t need to be laundered through an AI to see what it thinks, and have its thinking be the deciding factor.
No ego with AI
Seniors with a bit of ego by default, pushback on arguments made by peers. But AI is not human, so the confrontation never takes place. “Wrong. fix it” at most. AI doesn’t give you arguments. A peer with valid arguments will keep pushing back. That interaction sometimes further irritates ego.
If someone dismisses peer arguments without valid reasoning, but then considers the same if said by AI, then that is a culture smell.
One might say, they have decades of experience. To that I say, they shouldn’t have then agreed with AI out of all things.
PR reviews
I haven’t used CodeRabbit and friends. It’s essentially one AI agent reviewing the PR, with its own user so it’s easy to see that it’s AI.
One thing I have seen, is beyond stupid.
People authenticating AI for their accounts, and then sending out AI reviews from their accounts.
Now I don’t know about you, but when I tag someone. I want them to go over my code and review. You know, human.
If I now add 2 reviewers, I now get 2 AI reviews.
Just use coderabbit or something if you really want AI in the PR.
But please, stop your account from being a laundering account for AI. I want human reviews from reviewers. AI can be integrated through a separate account.