Feeling weird about my career with respect to AI

The depressing thing about AI is that regardless of the long term viability of the strategy companies are taking right now, it's only bad news.

If AIs long term capabilities are overstated, we've wasted a whole shit ton of money, effort, and time forcing something to be used that will just result in tens of thousands of companies going bankrupt and millions losing their jobs.

If they're not overstated, human beings have lost the competitive edge to do be given any sort of actual, rewarding work. Solving problems is interesting. Making art is interesting. Writing code is interesting. The long term path of AI if everything pans out the way they say it will is that humans are really only there to provide some sort of intent to the AI.

User: "Make me a useful product"
AI: alright. Here are the 10 most profitable ideas I have. Which would you like to pursue?
User: whatever. The one that will make me the most money.

Any sort of actual interesting work is totally stripped away. That's a best case scenario if this stuff pans out too.

The whole thing makes me really cynical about the future, and has turned me into sort of a luddite.

In the industrial revolution, there was some refuge against those feelings. Namely, "yes, my sweat and manual labor are no longer valuable, but I, or my kids can become a skilled laborer instead"

When the information revolution and automation started eliminating skilled labor, it was "yes, my skill with my hands is no longer valuable, but I or my kids can use our minds to solve problems instead".

Now that it seems like our minds aren't valuable anymore, what's even left?

In my perfect world, not only would this technology not exist; it wouldn't be possible to exist.

I'm in a similar boat. My company has been all about AI for the last few years, but especially in the last year. It's not quite as shoved down our throats as it sounds like it is at your job, but it's not far oof. We had some training, we'e got a blank check to grab AI-powered IDEs and other tooling. We're using it fairly responsibly, with it being used to generate released documents, preliminary PRs, PR summaries, etc. Stuff that AI's really good at and is easily checked by a human. Devs also use it, but it's not mandated.

But then every now and then we get some blatant AI propaganda meme about how it won't take our jobs, but elevate them...and that just doesn't pass the smell test. I can absolutely see how in 5-10 years AI could (but certainly not should) replace entry-level developers. And it's terrifying because as you suggested, we could also get to a point where we're just glorified PR factories for the AI. And I'm not saying that's a sane choice, or that AI will actually be ready for that, but it's not developers that pull the purse strings in corporations--- it's the business dorks up top. And all they see is dollar signs with little regard for how their choices impact the people who do the actual work.

Years ago a Senior Dev warned me against using git UIs because they "make you lazy and forget the command line" and I kind of laughed it off, but he was right and I'm seeing that happen with AI. It's making me lazy and forgetful because I can just let the stupid robot do it for me.

At first it was a huge benefit because I could use it to quickly ramp up on new projects and start making impact immediately, but those accomplishments now feel hollow and I want to just...write code myself. I want to go back to the old days of nothing but my wits and an IDE. Even if my wits let me down from time to time!

This is one of the reasons I am wary to enter back into really almost any part of the tech industry that isn't physical in some way.

I want nothing to do with any institution that legitimizes use of those systems. Will not work with people that use it. Will not be in a position that requires it. (Don't ask me why, no need to rehash the pro/anti AI discussion for the thousandth time- part of it is personal and emotional in addition to political and social)

Mainly just want to say, even as someone that isn't a programmer, but has done adjacent tech work, I feel you

When I left my most recent position it was already starting to infect systems administration with coworkers using it in ways that caused them to have scripts they couldn't fully explain or would oddly fail and they couldn't figure out why, etc, and this was not in an industry where I would want people to take such an approach to extremely vital systems.

Though I will say- utility is secondary for me. I would still be against these systems even if they were 100x more reliable than they are now, in fact I'd oppose them even more strongly, because they would be an even bigger threat. Whether they work or are useful or not doesn't matter to me or at least does not move the needle of my opposition- things can be useful and also bad for us at the same time

I’m a software engineer and I’ve witnessed LLM usage explode at work too, though not to the degree you’ve described. I just want to say I totally understand how you feel. It’s tough working through years of school only to have the career you envisioned change in the blink of an eye. I’m sorry that all this nonsense is happening to you and our field right now. Programming is a wonderful craft and it sucks that your company is taking the joy out of it.

LLMs are almost certainly here to stay. However I don’t think this is the end of software engineers as programmers. The industry goes through cycles and right now we’re in a hype bubble. Even if we have to endure a few years of this nonsense, there will be a light at the end of the tunnel. I encourage you to find joy in programming outside of work, if you have the time and energy.

Again, I’m sorry this is happening and I wish you the best moving forward in your career.

Ignore the hype, give it time, the rapid pace will slow down eventually and along the way the software industry will figure out how to balance people and agents.

In the short to medium term, provided you're a good engineer (or even just above average), you won't become redundant. It's people graduating now, and over the next few years, who are going to have a rough time. And a lot of them will figure out how to adapt because they have no baggage to bring into the process of learning a completely new technology that almost no one has experience with at this point.

Whatever you do, don't avoid learning how to use coding agents. There is no future where they aren't an integral part of software development.

You mentioned using Claude Code... there are lots of ways to use it that don't involve just telling it what to do. For example, let's say you're writing core functionality for a new feature and you realize there are some helper functions you're going to need that aren't in the spec. Highlight relevant code and tell the agent what you need, then go back to writing the interesting code while it whips up some boilerplate. If by Claude you mean Opus 4.5, and it has codebase patterns and conventions in context, 80%+ chance it gives you what you need with no issues while you're busy making progress on the important bits.

I get your frustration, and the frustration of most everyone else who's posted so far. None (well almost none) of us asked for AI, and even if we had we wouldn't have asked for it to be unloaded onto society via a trillion dollar firehose of hype and FOMO in service of capital and little else. But being angry at it is akin to being angry at the weather. Unless you're going to move to a different climate, find a way to enjoy it.

Hi OP,

I'm in the same boat. With similar years of experience, but working in data science and machine learning. I can relate to your feelings.

Best of luck, OP. I pray that we can look at this post in the future and laugh at my paranoia.

I have a lot of concerns about AI too, but with respect to programming, here is a different perspective.

When I first became a programmer in the late '90s, I had a few books from college that I took with me everywhere. These books had algorithms and patterns. One of the books was "The C Programming Language", which showed the implementation of various low-level functions. I would refer to the books as needed.

A few years later, when java and then c# came out, programming became somewhat different. There were more high-level languages and things were moving a bit faster. I didn't ever have to figure out how to write a sort algorithm anymore, I just had to find which library had the best one. At this time, my colleagues and I would frequently go to bookshops and buy new books. We bought a lot of those O'Reilly books that have the animals on the covers. These books would usually be good and accurate for 6 months to a year. Things were speeding up though and you had to keep learning the new libraries etc.

A few years after that, things were speeding up again. Javascript was becoming a big thing. We stopped buying books to keep up, we would just go to sites like Stack Overflow (nee Experts Exchange).

Now we aren't searching the internet for answers. We are using copilot in vscode or intellij.

But it's really all the same thing, just sped up. It's slightly different in that copilot can kind of write some code for you, but that's hardly different than pasting a block of code from a book or from Stack Overflow. It's just happening faster. Also, a lot of the world thinks AI is more powerful than it is and that it can replace people. That's probably the dangerous part.

Early in my career I was kind of jealous of engineers who came before me when things were lower level. They often had to carefully write very low level code in assembly or whatever. It seems more satisfying than just calling someone else's libraries. I guess that is kind of what you are thinking, but magnified.

I feel the same way.

It used to be you could pick a niche that you’re good at and enjoy and make that your career, like I was really good at making micro services that actually made sense.

But now everything is just AI. From top to bottom all engineering is about knowing your models and AI tech and working with that. Everything Ive made in the past 3 years is proprietary to AI. Used to be I could build from the bottom up, no third parties, but now I’m just an AI engineer. The latest API architecting standards are old news, now I have to know the latest AI standards. I was going to build a whole career off my knowledge before and now I’m basically a mid level SWE again with 3 years experience in “AI” which would be fine if I went to school for something like machine learning but I didn’t, I wanted to design data pipelines. I was a data architect.

4.5 is a fair amount for software engineering - you're certainly not entry level anymore.

Personally, I don't think it changes all that much. I already didn't do that much actual coding anymore. The more you go up in level, the more your responsibilities shift from writing code to executing projects holistically. It's a common joke that you go from writing code in an IDE to writing code in google docs as you develop as an engineer - that is to say, you spend more time defining objectives, scoping projects, and aligning other people on what you're going to build than actually heads down, code writing.

In that respect, tools like claude code just reduce that ever smaller sliver more. Which doesn't change all that much. In the back drop of "I'm measured by the amount of impactful work I can execute", I don't particularly mind using any tools that speed things up.

I guess all that is to say, you were going to eventually move away from doing all that much hands on coding regardless, if you were to progress your career.

My company is rebranding to have AI in the name shortly

Most companies that rebranded to have dot com in their name didn't do well. The few that did well were founded with dot com in the name from the get go (Amazon.com, Flowers.com, Bookings.com.)

have been forcing us into 2+ hour long AI trainings once or twice a week

What sort of training is relevant in AI these days? Classic AI is largely dead to most companies. LLMs are so new the best training is to simply roll your sleeves up and get coding.

I want to solve problems and think and write code, not talk to an AI and become a full-time code-reviewer.

The need to think logically is not going away, but I fear the need to write anything more than psuedo code is mostly going away.

Early on in the dot com boom, things moved a lot more slowly, and so I spent most of my time enhancing other peoples code. You think reading code generated by an LLM is soul sucking? Try reading code hand written by one human, and edited by ten others, back when VI & SCCS was state of the art.

Those were the days. Working on new stuff, meant a one-three hundred page requirements document and 6-18 months to code, test and QA the beast. Except perhaps those weren't the days. Because the guy who wrote the specs specified the how, not the what or the why. Worse, he was in no way qualified to define the how. Even more worst, he had not spent enough time thinking about the why to even begin to specify what.

Perhaps I prefer the days I learned to write pure hex? Punching machine language in one hex code at a time. Those were the days. Where bits made bytes but nibbles turned me on (old machine coders joke.)

Ultimately I think the other posters are correct. Change is sadly inevitable. Client Server. HTML. Javascript. Frameworks. Agile. Scrum. Cloud. AI. Things change. You will change. The change is just happening crazy fast right now. Eventually AI might get so good you never want to go back to coding by hand. You might become so senior, your days will be filled with useless meetings, architectural discussions, code you don't want to review, or simply figuring out why that one enterprise customer who is loading 100 million transactions a day is not getting sub hour performance like they demand.

GL.

I hear this pessimism about AI all around, and I don’t quite get it (from my perspective)

I use AI often for my work (and outside), but I enjoy the help. I use it as a kind of rubber duck and the same way I’d take to a peer to figure out the solution to a problem, or iterate over architecture, or even code. But ultimately, I’m in charge and I write the code (based on some AI suggestions sometimes, and based on our discussions).

I find this much more pleasant, the AI becomes just a (very fancy) tool rather than I becoming its agent. I really see it the same as a bicycle: you can still walk if you want but by using a bicycle you go much further much faster with less effort. Someone still has to steer the bicycle, shift the gears, décide which route to take, and watch out for cars and pedestrians and trees etc.

I think you should use the AI to do better work than you otherwise could. It’s upsetting to me when people produce mediocre work. There are always constraints. You don’t have infinite time or money. But don’t use AI to produce way more at lower quality. Use it to produce a little more at a higher quality. You can spend that efficiency budget on both output and quality.

Use AI to write tests (always review them). More and better test cases than you normally would have because most people don’t like writing them. Use them to raise potential issues with your last commit. You’ll have to discount 80% of their suggestions but it doesn’t take long to pick through the list. Use them to find where in a library something is implemented. Use them to get you unstuck when debugging.

For me it’s not too upsetting to use these tools when I can see them help me do better work. The users are happier and so am I.

I want to write a blog post about this, but briefly, I think video games are a useful metaphor for speculating about the future of programming.

Traditional programming is like a first-person shooter. Sure, you might have nice tools, like maybe an auto-aimer or a really big gun, but you’re fundamentally driving one character. Or if you’re multi-tasking then it’s like a turn-based RPG where you directly control all the members of your party.

There are also games like Lemmings or the Sims or RimWorld where you have somewhat indirect control over multiple characters, by giving them tasks or controlling their environment. They might interfere with each other and won’t do quite what you want and that’s part of the challenge. Fortunately you can restart the level if you need to. This is what it’s like to write software using coding agents. I am writing software with one coding agent and I can report that it’s fun and educational. It probably helps that it’s a personal side project. I’m still wary about running more than one at a time; it seems like running around spinning plates?

There are also RTS games where you control a small army in real time and frantically scroll around giving them orders. People are trying to write orchestrators to make software development like an RTS, but this is currently a crazy science project. Maybe it will be practical in a year or two. It seems stressful to me; I prefer turn-based games.

Zooming out a bit more, there are games like Sim City or strategy games where you’re managing large populations of NPC’s (which may or may not be explicitly modeled). There’s no equivalent to this yet, but maybe it will happen if they can get the coding agents to coordinate well enough?

Writing code in assembly is still needed in certain niches. I wrote a bit of assembly as a kid and took a course in college where we wrote assembly. Even back then, it was taught as something you should understand rather than something you’re likely to do much of at work.

Similarly, I expect that hobbyist programmers will be able to play the programming game at whichever zoom level they like and people will do some software development at each level as part of their education. Commercially, I expect that there will be lots of demand for people who are comfortable managing coding agents and cleaning up their messes. It’s a different game, but it’s still software development. You are giving the coding agents tasks by giving orders (essentially, writing bug reports) and attempting to control how they do it by editing AGENTS.md and other documents that the agents refer to.