IT/Software career thread: Invert binary trees for dollars.
- Thread starter Traak
- Start date
Khane
Got something right about marriage
- 21,501
- 15,389
I have never, in my entire career, struggled to get interviews. Until the last 2 years. It went from recruiters beating down my door until Covid, to only getting contacted once every few weeks, to absolutely nothing and now can't even get one to take a call or even attempt to work with me.
Dire is probably an understatement. I feel bad for every young person out there now who is about to graduate or is a fairly new graduate. I can't imagine what its like to pay the inflated tuitions that exist these days with the understanding that you were going into a field that was pretty much guaranteed to be lucrative (it was) and then having the rug pulled like this.
It's fucking bleak.
- 2
- 8,573
- 15,161
the stories are endless. Suffice to say here is a retarded pattern out of a million. can we connect to a mailbox with this shit ass app to scan for attachments to download them then pass to our file management system then upload to a servicing platform. why are we not automating this to send directly to the destination system? also all files are sent via email. ask them to stop handling shit manually and i get blue screen stares. and yes all the dev teams are indians. fucking morons everyone of them.In things that are in source control yes. The problem is that ancient Oracle DBs, MSSQL, and so on don't really have it. There isn't a legacy system really where you build out shit in SQL, store it in git and push to production and it then builds all of those things for you. If you did have that it was the result of forward thinking leadership and it was an in house solution. This is the problem my bae DBT exists to solve.
All of his jank is due to stuff that occurs around the edges in system architecture. Where access and controls are historically very limited. As in anyone can and does do whatever and has for a very very long time. When it comes to system architecture less than half of it really exists in source control or review as you would think of it. This is why network and platform people are big on stuff like Terraform, for example, where your config is all stored as code in git like any other thing. Dealing with all of that bullshit and trying to limit rampant retardation is pretty muchTomServo 's job.
Khane
Got something right about marriage
- 21,501
- 15,389
Also, something that is happening to me now...
I use AI all the time. It streamlines everything, its obviously very useful, and when you have to actually teach it new parts of your codebase so it isn't a drooling moron its kind of fun. And then you finish that, and you just send it commands to generate shit for you, and that is just so... unsatisfying after a while. My job feels soulless and boring now, I don't like being a developer anymore.
It has literally sucked any enjoyment I used to get from problem solving entirely out of the equation. Instead its like working with an idiot savant who is willing to learn but is pretty dumb if you aren't pedantic with your instruction.
I use AI all the time. It streamlines everything, its obviously very useful, and when you have to actually teach it new parts of your codebase so it isn't a drooling moron its kind of fun. And then you finish that, and you just send it commands to generate shit for you, and that is just so... unsatisfying after a while. My job feels soulless and boring now, I don't like being a developer anymore.
It has literally sucked any enjoyment I used to get from problem solving entirely out of the equation. Instead its like working with an idiot savant who is willing to learn but is pretty dumb if you aren't pedantic with your instruction.
- 2
Noodleface
A Mod Real Quick
- 39,377
- 17,826
My job is fairly niche so I still don't have trouble getting interviews, but the problem is these companies really think a senior engineer is going to spend their days grinding 500 leetcode problems. I can't tell you how many interview technical portions were straight leetcode medium problems. Sorry, but I'm not doing that shit, that's a young man's game. Ask me something real, I've been in the field 15 years.
On the topic of AI our VP sent out an email basically saying 'i heard AI doesn't work for you guys,.please tell us why. Also tell us who no longer needs a license.' I took the last sentence as a thinly veiled threat but whatever, I gave honest feedback.
The problem in UEFI is it just doesn't understand the underlying architecture at all. Sure, it can generate for loops and debug statements, but it doesn't understand the boot process of a system and how to write code for it. The higher ups think this will be solved if we just 'train it on the codebase' but it isn't that simple. There's a nuance that I'm not seeing AI capable of yet.
On the topic of AI our VP sent out an email basically saying 'i heard AI doesn't work for you guys,.please tell us why. Also tell us who no longer needs a license.' I took the last sentence as a thinly veiled threat but whatever, I gave honest feedback.
The problem in UEFI is it just doesn't understand the underlying architecture at all. Sure, it can generate for loops and debug statements, but it doesn't understand the boot process of a system and how to write code for it. The higher ups think this will be solved if we just 'train it on the codebase' but it isn't that simple. There's a nuance that I'm not seeing AI capable of yet.
- 2
TJT
Mr. Poopybutthole
- 46,395
- 126,680
Adding to that my team is building out a metrics suite on AI usage. It was primitive for 2025 in my opinion but 50% of your performance and thus raises, bonuses, or promotions was solely rated based on your AI use. The initial solution only scored you based on the chatgpt webapp. If you created a custom gpt alone you were a power user which was the highest. It was easily gamed but not many figured that out.
This quarter we are developing a high octane one that measures copilot and cursor use. Lines of code generated, most used models, lines rejected, in line code suggestions accepted, and how much makes it into your git branches and thus into production code. Probably more than 50% performance based on this for 2026.
As in you could have been a rock star at your job but if you didnt use AI to do it you were rated as an under performer. Its that serious.
Bet your asses all of your companies are doing this or worse.
This quarter we are developing a high octane one that measures copilot and cursor use. Lines of code generated, most used models, lines rejected, in line code suggestions accepted, and how much makes it into your git branches and thus into production code. Probably more than 50% performance based on this for 2026.
As in you could have been a rock star at your job but if you didnt use AI to do it you were rated as an under performer. Its that serious.
Bet your asses all of your companies are doing this or worse.
- 1
- 1
Noodleface
A Mod Real Quick
- 39,377
- 17,826
Well I don't really code anymore so I'd be fucked. I don't even know how they'd measure my productivity
- 17,621
- 8,643
I still code, not as much as I'd like, too much managerial shit. But just this week I refactored a python function that was throwing multiple linter errors for complexity. You know, too many lines of code, too many branches, too many parameters, etc. Usually I ignore those but in this case I think they're right, it's shit code.
This is the bulk of my work, refactoring existing code. How do you guys use AI and then trust the output enough that it ends up saving you time? In this example, I can't imagine an output where I'm not going through it with a fine tooth comb and testing each function it touched.
This is the bulk of my work, refactoring existing code. How do you guys use AI and then trust the output enough that it ends up saving you time? In this example, I can't imagine an output where I'm not going through it with a fine tooth comb and testing each function it touched.
Khane
Got something right about marriage
- 21,501
- 15,389
Khane
Got something right about marriage
- 21,501
- 15,389
I don't trust anything, I look at the output and then tell it to refactor its own output until it spits out something that actually works.
Yes, this is faster than doing it myself. A lot faster.
But you also don't use AI (or at least shouldn't) to crawl an entire repo or branch and try to refactor the whole thing. I don't use AI that way at all. I use it like I code, modular, one piece at a time.
- 1
- 17,621
- 8,643
What do you in the case of external dependencies that are resolved at runtime? I can't get my IDE to resolve those, so forever they remain red squiggles. In my original example, many of the code paths that I ended up touching were referencing external dependencies.
How do you identify the boundaries of "ok, it should be able handle this much"?
Khane
Got something right about marriage
- 21,501
- 15,389
How did you resolve it without AI? Do that exact same thing, but prompt AI to do it (or at least attempt to do it) for you. One problem at a time.
AI that exists today is not intelligent at all, but its got a lot of stored data and can access it (like memories) way faster than we can. It doesn't know everything, and its often not quite right, but if you understand the problem you are trying to solve, and can actually solve it yourself, you can use AI to solve it the exact same way you would, just faster.
I think a hurdle lots of folks might have with utilizing AI is thinking there are special commands or secret features or a certain way to use it. Its conversational, have a conversation with it like you would a co-worker. Ask it if it knows how to address a problem you're having. Just don't try to send all your action items to it all at once.
- 1
- 17,621
- 8,643
I just resort grepping the entire sandbox. Good ol grep and find.
I'm not sure what it would take to get over my personal hurdles. I know at the very least my employer would have to pay for it. No way these are remaining cheap long term. Nor would I trust myself currently to know how to set it up without leaking proprietary code.
That's not even touching moral and ethical quandaries.
TJT
Mr. Poopybutthole
- 46,395
- 126,680
So here's how it works right now. Its a scoring algo. Based on AI usage related inputs. You can imagine it like this:
5x+2y+6z+9.1a=100
With 100 being "100% utilization of AI." For 2025 the above was a combination of the amount of messages you sent to AI, the amount of messages you send to custom agents, time spent, and if you created and used your own custom agent. Custom agents were weighted HEAVILY to such an extent that if you made a custom agent (takes like 2 minutes to do in chatGPT) with nothing really special about it you then hit 70% utilization just from that.
This year's is going to be way more complicated. Gaming it by code generation bloat wont be what gets you to "top performance" this year I am certain of that. The real marker is going to be how much of it makes it into feature branches that are merged to prod code bases. Which is also tracked. The engineering team already uses Cursor like 30% more than Copilot. I am happy its my team building all of this shit so I am certain to max it out though!
Just is what it is.
You are resolved to a code proofreader and tweaking minor things or examining complex logic where it has the most problems. Then changing that.
Workflow is definitely 3x-5x faster than what it was 2 years ago though and that's a fact.