IT/Software career thread: Invert binary trees for dollars.

Deathwing said:

Generally, I agree. I honestly don't know how to square those two longterm as this is now my 4th attempt trying to type up this reply. I feel like we're living through the runup to 2008 again and not enough people give a shit about what storage prices are doing as a barometer.

And you're not being rude, not at all. Recently, I feel like I'm the black sheep in some these tech threads. I was expecting a far more bombastic response.

FWIW, I've started dabbling in Claude Code on work's dime. It wrote a script for transition bugs out of an outdated system and into gitlab. Not really had that eureka moment. Yet? Idk, maybe I'm too stupid to figure out how to integrate it better into my workflow. I just spent a whole day pouring through logs to try to figure out why clang++ couldn't find basic headers like <any>. Turns out, emacs had installed libgcc-14-dev but not gcc-14, so the headers were incomplete. When clang++ ran, it picked gcc-14 over gcc-13 for system headers and barfed because of course it would.

How do I integrate AI into that kind of workflow? That's like 50% of what I do.

Click to expand...

"I'm getting this error, how do I fix it" is what I ask Claude a lot.

Well Gemini is not using anthropic models, so yes it's AI but it's not great.

By the way today we finally ran my code and it worked the first time on a live server. Just bonkers.

It sounds a bit like you're not even attempting to use AI at all because of preconceived notions Deathwing

Deathwing said:

I was posting serious sized chunks into google, does that count as using Gemini? I usually scroll right past the AI response but I was getting desperate. I don't recall them being very helpful.

The tricky part of this is that I was trying to match a production environment, so the common solutions were all useless because they suggested system modification that wasn't going to happen.

In the end, it was my distaste of emacs that I uninstalled it on a whim and noticed apt saying some dev package could now be autoremoved that finally clued me in. Afterwards it was just 10m of figuring out what -I flags to supply to the commandline.

Click to expand...

Yea this isnt remotely the same thing. Those ai summaries in Google usually are true generic slop.

I started using Copilot in VS Code with model switching. At the time, the models were somewhat throttled, but after trying multiple options it became clear Claude was generally stronger, with GPT/Codex trailing in some cases. Around Nov or December 2025, these tools crossed a thresholdfrom browser chatbots to genuinely useful integrated development assistants

In order to get better efficiency and the most out of it, you'll need to to put work in at the beginning by either creating a specialized skill for what you are doing or even just a .MD file for it to reference before each session. You can do it without, but it will usually make the same mistakes and require you to feed it the necessary info every time you start a new session. I try my best to remember to tell it to add critical lessons learned when Im done with whatever I was doing.

As of now, I cannot really say which is better between claude and gpt as its basically an arms race between the two. Sometimes gpt will solve an issue for me pretty quickly that claude will get stuck on and vice versa. The good part about that is you can usually have the other review a plan or code as a 3rd pair of eyes and it will call out gaps or problems you or the other agent missed.

Its also really not difficult to utilize or use.

Khane said:

Its also really not difficult to utilize or use.

Click to expand...

That's the thing I don't understand. People talk like AI is hard to use skill or is this huge barrier. It's pretty damn simple to use when you break it down.

The entire reason Helpdesk 1 exists is that 2/3rds of the population can't remember which sticky note has their outlook password. Do not underestimate how bad people are at computer

Deathwing said:

But I said I'm using it. I listed the reasons for my opposition and only one person responded to that. It was a valid response, but didn't really address my points.

I'm pretty sure I'm misusing it or under utilizing it. I think that's orthogonal to my opposition. That was the main reason I made that post. Opposition to AI was being classified into 2 camps: ignorant or hostile. I think there's a significant third camp: useful but at what cost?

Click to expand...

I think questioning the ethics of utilizing so much AI is perfectly valid. Think about a place like nvidia where jensen said if your senior devs aren't using 250k tokens they shouldn't be employed. 250k tokens is more than we have allocated for our entire couple hundred developers. We get between 100-700 tokens a month. In the grand scheme of things what I'm doing vs what Nvidia is trying to push is night and day.

Being cognizant of what you're using is ideal, I just hope that you embrace it for your careers sake. This feels more than just a dot com boom.

dragonbr said:

I really feel like the ones opposing it fall into 2 categories. The first being people that used early iterations of chatgpt, saw the glaring slop issues with the lack of integration, and then formed their opinion and never looked back.

The second set are people that have seen what the latest models are capable of, and their survival mode activates. They convince themselves that its still bullshit due to the fear of change and the threat of becoming obsolete.

It's still by no means perfect but in comparison to what agents were capable of even just a year ago it's made massssssive jumps.

It definitely still requires guidance and hand holding. If you are giving it garbage prompts and or dont have a solid understanding of what the finished product should look like, then you are probably going to get garbage out. But even that issue is somewhat being solved with planning modes. After using that mode for a bit it still surprises me with how well it knows what questions to ask me regardless of the subject matter.

Click to expand...

Some developers here have complained that it takes away the "fun" of coding. Which I find kind of bizarre. I write code for the end result...what is created. Not for the act of coding itself.

ShakyJake said:

Some developers here have complained that it takes away the "fun" of coding. Which I find kind of bizarre. I write code for the end result...what is created. Not for the act of coding itself.

Click to expand...

I am in this camp, I like coding because its a creative process. AI kind of (not completely but kind of) removes that creative process.

It's kind of like if you spent 20 years being an author and suddenly the world decided now you have to be an editor because authors are passe.

Khane said:

I am in this camp, I like coding because its a creative process. AI kind of (not completely but kind of) removes that creative process.

It's kind of like if you spent 20 years being an author and suddenly the world decided now you have to be an editor because authors are passe.

Click to expand...

In another thread here on the forums, I described my idea of creating a fantasy 8-bit "super games console" using off-the-shelf components that would have existed in the mid-80s. I then realized I could write an emulator for such a thing. However, doing this properly would require extensive knowledge of CPU timings, Z80 assembly, C++ for the emulator, and the right libraries. Hand-coding the whole thing would have taken forever. For me, the real enjoyment comes from being able to take any idea and quickly realize it with AI. That's where the true power lies - not in the long "journey," because doing it the traditional way would literally have taken years.

TJT said:

What? A single prompt in Cursor is like 20k tokens. Even when asking a question about anything. For $20 you will have like 10M tokens. Minimum.

Click to expand...

I am now realizing all tokens are not created equally. I'm using windsurf/Claude credits