anyone else feel like their brain is turning to mush since fully adopting cursor/claude?

Posted by StatisticianFluid747@reddit | ExperiencedDevs | View on Reddit | 21 comments

i feel like i'm shipping 10x faster but retaining absolutely nothing. before AI, if i spent 3 hours debugging a weird caching issue or evaluating database trade-offs, that knowledge lived in my head. now I just paste the error, spar with the AI, accept the fix, and move on. the output is there, but my actual thinking just evaporates into the chat logs.

the worst part is the amnesia. every morning feels like 50 First Dates. i spend like 15 mins just re-explaining my architecture and past decisions to the AI so it doesn't give me generic slop. i have this massive rules file where i try to write down "i prefer explicit error handling" or "we rejected redis for this", but it feels like a full-time job just keeping my AI updated on how i actually think.

is anyone else feeling this weird identity crisis of just being a "prompter" now? how are you guys keeping track of your actual architectural decisions and context without spending hours writing manual notes in obsidian that you'll abandon in a week anyway?

[-]

CardinalHijack@reddit

I've been thinking about this a lot, because I feel something similar, but I don't think its quite as bad as you think. I also would highly doubt that if you spent 3 hours on something, that that knowledge would live indefinitely in your head (I forget stuff im not doing repetitively within a few days).

Before I had AI to help with engineering, my workflow was as follows:

- Need to solve another problem which I've actually solved before, but forgotten how to solve. Be it use a certain method ive used before, or set up something ive set up before, or implement something ive implemented before or a method available to me in the language I use.

- Having forgotten how to solve it, the next step of the workflow would then go down a path of either looking at a bookmarked stack overflow post, or opening up a certain file to remind myself (but taking a while to find said file because id forgotten which it is) or looking at docs. I knew what the solution was if I saw it, but I couldn't quite remember it off by heart.

- Once I've refreshed my memory on how to do it, begin lifting certain bits of code, writing some new parts to solve said problem

AI, for me, has essentially removed that need for me to remember. I dont need to go and find my book mark, or trawl through the code to find the example of something, or refresh my memory on an available method by looking through the docs - AI now fills this gap for me in seconds.

Initially I thought I was bad. It felt like I wasn't applying myself. But I don't think thats true now.

If AI can give you 5 solutions in 5 minutes, and you know which to pick, and you implement its solution in a total of 10 minutes in my opinion this is orders of magnitude superior than you solving the same problem yourself in 3 hours.

I think the strength of a senior dev comes in knowing which is the right solution, rather than spending hours to find it.

UnderstandingDry1256@reddit

Hours lol? I remember high level stuff, somewhat you can easily draw on whiteboard.

Keeping notes is so 2024-ish.

I don’t care about deep details, like which args this endpoint gets etc. Just ask cursor whenever you need this knowledge. It will keep up with the real state of the code rather than permanently outdated notes.

Code is the best documentation, and now it is stronger than ever before.

TehBuddha@reddit

Then Claude goes down for an afternoon and you can do absolutely nothing

Close your laptop and have a short day then :)

You optimize workflows for everyday routine, not some extra rate emergency.

MisterIndecisive@reddit

Not really AI, is far from perfect. It still chucks up nonsense, and you have to review the code and have make sure tests make sense. Maybe it's because I just use as pair programmer rather than one shot and deploy

Sheldor5@reddit

at this point just write all the code yourself

there still is no proof that AI speeds anything up, in fact it has a lot of hidden costs (increased review time, more bugs, more post merge work, codebase becomes a mess more quickly ...)

Nah personally I find the pair programming element a big benefit as helps me think through stuff better. Speed is better and code review by colleagues isn't proving to take more time, less if anything. No doubt other people can use it more effectively but right now I have found a balance to improve productivity and retain good quality if I don't let it go completely wild.

do you feel more productive or do you have actual measurements you can compare?

because studies show that people using AI feel more productive while taking up to 30% longer in reality

mechkbfan@reddit

Makes me about 20% faster and still turns it to mush because you're not using critical thinking

FriendOfEvergreens@reddit

Some times yes, sometimes no. It's nice to have an "expert friend" on topics I want to learn more about. In some ways my general creativity has increased a lot.

For some problem solving stuff though I definitely see loss. There's a default jump to LLM that I purposefully avoid. For first thoughts, the anchoring effect of the first idea you think about/hear is real, its important to keep that as your own thoughts first IMO. There was a time when I was about to make a relatively simple decision but I asked Claude without much critical thought - I realized this was happening and have made it a habit to not to do it.

In terms of small coding stuff, I don't really care as much. I agree with the worries about losing hard skills, but even local models are so good now that I no longer fear the "token cost rugpull". I think assisted coding is here to stay permanently, so its more about keeping your decision making skills/architecture top of mind than the low level stuff.

AggressiveResist8615@reddit

That's my secret, it's always been mush

Ok_Alps9776@reddit

reminds me of debugging spaghetti code

experienceddevsb@reddit

This flair is only allowed on wednesday, saturday (UTC). Please repost on an allowed day. Intentionally trying to circumvent this rule will result in a suspension. See: https://www.reddit.com/r/ExperiencedDevs/comments/1rfhdrg/moderation_changes/

wowbaggerBR@reddit

Yes, I am exactly in the same boat. And I don't have any answers because execs are pushing me to use AIs.

tacosdiscontent@reddit

Luckily no, since I don’t do offload everything to ai. I do however prompt most of the stuff and every iteration is only one or more functions and after AI adding them, I instantly review and adjust if needed. If you run 3 agents in parallel then you are cooked

eoz@reddit

I simply write code with my own brain

that's also my take after trying AI

african_or_european@reddit

I started by just letting the models go nuts and build whatever they wanted as long as it met my spec. But as I use it more, I am finding all of the cracks and I've started keeping a much tighter rein on what I allow.

I'm trying to work out a system where I can think about the system more at the architectural level and less about the specific implementation details, but I'm not sure if the models are not there or the process I'm developing isn't.

I definitely see a path but it's pretty rocky at this point.

Independent_Corner18@reddit

Totally getting you on this. I have been using Claude for sometime now, I'm not feeling the Artistry of hand-building my projects.

I think it depends on how to use it too. Delegating everything to IA will definitely makes me feel 100% dependent of it and I can't take on the project without it afterwards, which leads to sacrificing the joy of the work. I do think it's saner to take time to write down your design to IA, then ask it to enhance it. Debugging/reviewing small IA outputs is definitely more manageable and enjoyable than letting whole systems generated at once.

It comes down to the way IA is used.

Altamistral@reddit

There are already studies about this is psy journals. Using AI regularly replace your reasoning and your cognitive skills suffer. Basically you stop learning new skills immediately and then slowly you also start forgetting what you know, similar to how managers become post-technical after a year or two.

lugia4k@reddit

Unfortunately, it seems that's how it's going to be in the future for us.

I'm wondering how will the interviews be in a few years... are we supposed to just grab their claude subscription and speed run a project?