This sounds pretty similar to all technology. I remember Slashdot threads about how IntelliSense was making software engineers dumber. I remember my teachers saying that allowing calculators on the SAT made kids dumber. I remember people saying that spell checkers would make people worse at writing. I remember people saying at the start of the pandemic that Zoom meetings would make everyone anti-social.
None of this really happened. I mostly use editor tools to automate away tedium that doesn't matter; I type "log.Inf<TAB>" and it adds "src/internal/log" as an import at the top of the file and types the o and open parenthesis for me. I have not forgotten how to do that myself, but it saves me a couple seconds. Calculators didn't really make people dumber, though I have to say that a lot of arithmetic I learned in elementary school did make me dumber ("touch math" was the killer for me; slows me down every time I do arithmetic in my head; I need some brainwashing program that deletes that from my brain). Spell check didn't make people worse at spelling; spelling things wrong still has a penalty (C-w to kill the last word and spell it correctly), so the incentive is to still to lurn how to spel wrds rite. Zoom meetings didn't ruin the corporate world; I personally found them very helpful for memorizing people's names with a high degree of certainty. You see it under their face for 40 minutes at a time, and you learn it fast. In real life, probably takes me a few weeks for people I only see once a week. So, honestly a benefit for me.
The current state of AI seems very similar to these technologies. I did a copilot free trial (and didn't renew it). With the free trial I think there were a couple things it was good at. One time I wanted a CPU profile for my app, so I just asked the AI to type it in. Open a file, check for an error, start profiling to the file, stop profiling, close the file. Would have taken me a minute or two to type in, but Copilot typed it in instantly. I also did something like "do the same as the function above, but for a YAML file instead of a JSON file". Again, super trivial to type in that code, it's really only one line, but the AI can handle that just fine in an instant. I don't really think it's more than slightly smarter IntelliSense, but without any access to the compiler toolchain, so it can sometimes just hallucinate stuff that doesn't exist.
I've found this to be kind of an interesting way to proofread documents and design APIs. Give ChatGPT a document you're working on, and then ask questions about it. If it gets the wrong answer, then your doc is no good! Similarly, ask it how to write some code using your new API (without showing it the API). Whatever it writes should be the surface area of your API. This avenue is pretty interesting to me, and it's not replacing humans, it's just a smarter rubber duckie.
Overall, I think we're in a little bit of a hype phase with AI. I look at it kind of like a dog that has read every book, if such a thing were possible. Pretty smart, but not quite human yet. As a result, it's not going to do well in the areas where people really want to apply it; customer service, insurance claims, loan underwriting, etc. But it is pretty good for asking questions like "does my document make sense" or "please find some boilerplate to cut-n-paste here, I am going to delete this before checking in anyway". Also not too bad at slightly modifying copyrighted images ;)