I can close an eye on using LLMs for research only.

These days I feel I am the only one that hasn’t changed their way of writing software at all. I’d honestly switch careers rather than manage agents. But I’m currently out of work (contract ended) and wondering if there is still a sane place to work, or is it truly time to pivot to another career.

I don't touch LLMs for any purpose (when I can help it, they are frequently shoved in your face regardless); and up until recently was doing just fine. I was fired for this stance, not for performance reasons (officially stated that there was no concern with my output), only for political reasons (insubordination, refusing to use LLMs to do my work for me).

I have been seeing fewer and fewer job ads that ask for/require LLM use over the last couple of months, so I feel like the tide may be turning in our direction; but I am starting to feel the stress. This is the first time I have considered that I might have been "wrong" taking an absolutist stance against LLMs, though I think that's just the unemployment wearing me down.

I’ve learned for myself that I need to write code just to keep my brain in shape - same as going to the gym to be healthy.

Even though agents are building all of my code now, I’m starting every morning with a LeetCode problem. There is no practical use, but it make me feel good and sharp for the rest of the day.

For anyone coping with AI I suggest to give it a try, and stick to it for some time, just like with a gym.

We still have some product development software teams at my company that have not changed at all in the last year. We develop medical products in the U.S., so it is heavily regulated and processes change very slowly. We also require reviews from our legal team, so it takes a while for any new contract or agreement to be approved.

This is temporary though; other teams are fully embracing AI in their processes, and it's really only a matter of time until all teams use it. We have C-suite instructions to promote AI adoption. AI will slowly creep into software testing, verification, and eventually the requirements management and software development itself. I'm working on some of the pilot projects.

The code we write here is all human-generated. There’s been an initiative to explore LLM usage. So far it’s been in a reviewer role only, as in “here’s some things I found that you could fix”, and then humans discuss each one and determine if it’s valid and how to fix it.

Also one of my domains is in microcontrollers and using specific hardware that needs bit-banging drivers written for, so it’s unlikely any LLM has enough data to write those things itself.

There are so, so many. You just generally won’t hear about it here.

Where I work, I have chores than a claude subscription would knock out in minutes instead of months. But I don’t have permission to run something like that, so I don’t.

Weirdly our contractors can. They wrote it into their contracts. It’s not even fair, really.

They say they’re working on it, but some people might be pushing back.

I am working on an embedded device where AI coding has been contractually prohibited. I use it for learning about APIs and modern tools, but all of the code comes from my brain. The reason is that the issue of AI copyrights has not been resolved, and this product will be sold to parts of the government that care.

That said, their stance has been wavering, given the apparent inevitability that AI-washing of copyrighted works will be rubber stamped. As you say, damn near everyone has now embraced AI and infected their products with their output. The economic shock if this practice was found illegal would crater the stock market, so of course it will be made legal (at least for the big US models whose corps have been spending their lobbying dollars wisely).

Once the outcome becomes sufficiently clear, I expect the tide will turn, and I am not entirely sad about it. I would like to give them a spin for creating one-off tools, running security audits, tackling certain refactoring, analyze log files, and perform other adjacent tasks. But if I have my way, I will never delegate the core tasks of architecture, design, and development.

I can only say that because I am operating as an independent contractor, so I have a little agency to decide what I do.

It's now like coding using the socratic method. Knowing the right questions is now the hard part. And you need to know the context to ask the right questions. Yes, I'm not typing in as much code or reading as much code directly, but still figuring out how to produce a good solution.

Every four months or so we try getting hyped on agents for ~ a week. Up to this point the answer has always been a pretty solid "nope this sucks back to manual code." The agentically coded parts of the codebase that came from these trials are palpably more annoying than the rest. However, the current trial is going better than the last ones have (Fable 5.1 really is astonishing). On the other hand, each other trial started with a honeymoon so it's up in the air.

I kind of do. Not by will, but because as a consultant i'm most of the time constrained by the tooling and environment i can use at my client's. It's hard enough to build a reliable unit test pipeline or ci/cd, so even if i can use the agents built into databricks, i don't control the harness and i cannot build the necessary guardrails needed to use agents in a mature way. It sucks and i think it will hold me back on future projects.

I am a lead developer working in Cyber Security. We are just now integrating co-pilot reviews into our pipeline to help speed along the initial review process and weed out any surface level bugs. We still have no agentic coding, and I'm not sure we will for at least another year. I have advocated to modernize and keep up with modern tooling to no avail.

I still write code myself but I do use LLM. I use them for exploring ideas, explaining things, look up syntax, look up API/reference, look up algorithms, write tests, write duplicate code, write small functions that can be reviewed easily, do code reviews, find bugs, do performance review, do complexity analysis, etc.

The LLM generated code are correct about 80% of the time. Sometimes the generated code is overly complicated, has poor performance, missing a few assumptions, etc. that you need to review it carefully. It helps to have a solid and extensive set of tests.

Having a weird experience with this at work. We are being told to use it, but the amount of flip flopping on whether or not we can use ChatGPT or Codex or anything else has left a lot of us without any clear direction. One day it’s allowed, the next day IT security guys have blocked it. Local models are also not allowed. I’m carrying on coding everything myself, because I can’t keep changing my workflow when IT changes their mind every second day. Despite this, the directive from the CIO is that by next year 70-80% of code must be AI generated. Last I checked only 2% of the company was actually given a license for codex. So idk man

> These days I feel I am the only one that hasn’t changed their way of writing software at all.

If you mean using a text editor, a shell, and a makefile, then you're not alone by any stretch of the imagination.

If you mean your skills stagnated 10+ years ago and have not advanced since then, then... yeah, you might be the only one ;).

I have said it in some prev threads but the majority of my workplaces code is written by ai today, and it's not looking good.

All implemention decisions are largely decided by claude these days.

We have so many useless tests, so many repetitive code, Every change, even few lines long will trigger 25-50 additional files changes just because everything is so horribly written.

I have heard this was the case before ai, but it got so much worse.

At the end, I believe it comes down to 3 things: care - which before ai you had to have at least some care. Strong leadership - the ai psychosis spreads like a desease in our csuite. Ownership - there is none, every interaction you have about something starts with "but claude thinks that or says this"

I wish we could go back.

We have groups in our company that do not have access to LLMs yet, due to company beuracracy. Also, token usage is capped, so developers need to be strategic about their usage. This could of course all change. We are swithing from Codex to Github Copilot due to pricing (I think).

Thus far I still am, but I'm in a very weird and atypical work situation at the moment.

Weirdness aside, I'm continuing to write code as I used to, despite the overall company _cultural_ push towards AI, given the lack of direct orders, and pray my output is good enough that the "powers that be" leave me alone.

I am also in the job search process, and am also still noticing plenty of listings that don't mention AI.

I don't code much in my position (State of California), but I'm pretty sure the public sector/government is a solid bet. Government has a major reluctance to use (and depend) on techniques and tools that haven't been proven out for like 5+ years. I suspect that most government coders' practices aren't at a 2021 best practices level yet.

In my company they give each one claude code.

The transition of winning an argument change to “claude said…” which are fucking annoying facing this shit in daily basis.

Code base becoming a blackbox where god and AI know what its like, every function have 4 line of paragraph historically describing changes from past to the future.

Im thinking of resigning because i cant bear this shit anymore but seems like its not better in other place.

Its time for me to retire and change career IG. Im sick of this becoming middleman between my manager and the AI

I use Codex extensively, but mostly as a sounding board for architectural questions. I'll also have it do a lot of boilerplate work, but damn, I don't get these people creating multiple Claude or Codex accounts to burn tokens all day. I still love writing code!

Yeah, as much as I can.

I like building and tweaking computer stuff, and managing agents is a completely different carrer than the one I signed up for. I guess corporate interests fell out of alignment with me with this one, definitely thinking about my career more these days.

>I can close an eye on using LLMs for research only.

Yes, this is the only way I use them. As a better search tool for the collective knowledge, and to do arbitrary queries on a complicated code base.

If there are any mechanical changes, like renaming something or such, I might use an LLM for it as well.

Interviews require you to code without any LLM help. 9/10 require this still. Once inside, I don’t think they write any code themselves.

I work at a significant computer security software vendor with 5-digits of employees. They provide AI tooling and training and encourage people to use it, but it's not like "use this or we will fire you and kill your children" like you hear from other companies. I use it a couple times a week, mostly for code review and sometimes boring boilerplate-type tasks, but it hasn't been a major overhaul in how I work. People here who obviously outsource their brain to the AI get negative reactions to that behavior.

If the company does try to turn the screws and force me to use it more, I'd rather exit the company (and maybe the industry entirely) than become a slop slinger. We'll see how it goes, but for now at least, I think this company has been handling it well.

Yes, I do. I don't see an advantage to using LLMs in my main project. I do see dangers: losing the overview.

A colleague does use it extensively on another project, and it's near impossible to follow at some points, even though it's a small project. There's a lot of code duplication, that's for sure. And there was a complete lack of security, as the project started as a simple nodejs based demo, and then sprawled out. I fixed that the old fashioned way on the outside (sessions, auth, and a CRUD API).

I do use LLMs occasionally, for converting code (something they're really good at), and trying to find bugs, that kind of thing. It is really impressive, and it's fucking up education.

As the meme says: start eating healthy now, because your future doctor will have vibed his way through med school.

Graphics programmer who still codes by hand here. I use LLMs for search, and for rubber ducking. I would not disrespect the years I've put into this craft by outsourcing my thinking. Nor do I want to wake up in ten years and realize I have willingly made myself stupid through overuse of these tools.

There are some "boring" enterprises that "lag behind" (subjective) the alleged trend. It just needs to have a sceptical independently-thinking CTO I guess. Which kind of goes against the typical people filling such roles.

I'm pretending to drink the koolaid, faking interest in agentic coding while I figure out something else to do. It's okay for personal projects but professionally, I honestly hate reviewing slop, being sent 10+ page documents full of junk that nobody has read.

Unemployed.

Dow/classified projects can still be pretty old school. Our company hasn’t really done much on the AI front yet although I’m pushing hard to change that.

I ended a 10 year contract when the new clownshoes CTO started hyping the slop. I would rather stare at a wall than 'manage' an 'agentic team'.

I do not have an answer to your original question but I do have a comment that I have realized it myself quite recently. May be the analogy is not good enough but here it goes: no one would go back to punch cards etc, we have come a long way. And GenAI is such a tool - there is no going back.

What kind of programming do you do? If it's enterprise programming and it's basically CRUDs all the way down, yeah, LLM are pretty good at it and you need to find a balance if you don't enjoy it anymore.

All this bitching does feel a little off to me though, I get it, but I don't think you guys get how bad it sounds from the outside.

LLMs are not perfect, even in enterprise programming. Even in my case, where a few CRUDs may manage 100s of millions of dollars, and the code LLMs produce is actually good enough, for sure comparable with mid developers, most of them, you do need to develop a system to keep everything on track, I've found my balance and honestly? It's good and I enjoy it.

Go into embedded critical systems, that's probably the last thing they will get good at.