Anybody who has worked through the life-cycle of large projects knows that as time passes and the software gets adjusted and expanded, code just accumulates problem and brittleness - especially because multiple different people change it and they tend to each do it their way, often without full understanding of the code base - not just at the code level but also at the software design level - and eventually that code gets so hard to change or fix that a whole new system has to be built from the ground up.
In my experience this happens maybe at around 5 - 8 years of age of a codebase.
So I expect we’re headed for a spectacular industry-wide explosion because using AI code vastly accelerates this because for just about anything but small projects that can entirelly be generated in one go, AI isn’t consistent in coding style, much less software design.
Throwing software engineers at it right now only works if they end up spending even more time reviewing and ajusting the AI code than they would if they did the work themselves, since having AI coding is pretty much the equivalent of outsourcing to a pool of random junior developers.
Hell, even with smaller projects, you’re going to have debug cycles and if AI is driving those cycles, it will be acting as a new coder for each invocation (which happens multiple times per prompt for systems like claude code).
So you’ll get shit like duplicate helper functions, other code not using those helper functions anyways, debug code added and then not removed, errors and warnings using a variety of styles, overly verbose and redundant arguments, support for enhancements that don’t even make sense in that context, confidently incorrect assertions about what is and isn’t happening or possible, etc.
My manager wants me to make a presentation that sells some AI debug solution but the hand holding I have to do for it to actually understand and not give useless conclusions means I don’t even believe in it. Or the case where it did help, turns out it didn’t even use the tools provided by the solution and was just CC.
I’ve mentioned the cycle of being impressed with what these LLM-based systems can do and feeling like I might have been unfairly critical, and then running in to a major issue that justfies the earlier critical view. Last times I mentioned it, I said I was in the impressed (but skeptical) part of the the cycle. Well, I’m back to the “this might just be a complete waste of resources” part of the cycle.
I wonder how many engineers they’ll have to pull out of retirement in a few years to fix a lot flawed logic.
I remember back in the day when they had to pull Cobol programmers out of retirement to update mainframe software because of Year 2000 and they got paid a bundle for it.
Similar thing for customization of older SAP systems after SAP changed the language used to Java but those systems were still done in the old language.
So I expect that freelance senior designer-developers are going to get paid A LOT of money to come fix things in a few years’ time, especially since in places with high AI adoption this is going to be way bigger in terms of size, complexity and seniority of expertise needed that either mainframe code updating for Y2K and updating customizations in old SAP systems.
Cory Doctor’s recent book on Centaurs and Reverse Centaurs is worth reading.
The core idea of that is that centaurs are a human top and machine / alien body, they’re effectively augmented humans with all this technology to help them excel.
Reverse Centaurs are human bodies and machine / alien tops, where the humans are just checking the work of systems and are subservient to them. He points out that that’s one of the fundamental differences between Amazon and the Postal Service is that in the case of Amazon drivers, they basically function as a reverse Centaurs where they are just an appendage of the delivery car, tracked and managed by that car, to do the tasks the car can’t do on its own.
Imagine being a horse waking up with a human torso and micropenis.
very interesting concept. another example might be railway drivers who are merely appendices to the vehicle. the vehicle largely drives itself, the driver is mostly there to check tickets, answer passenger’s questions, etc.
We used to have elevator operators
One of my most played songs last year.
lol sure they are.
They’re desperately trying to pump this AI shit up with these fake stories before they go public
Investors really are dumber demographic than MAGA.
Well, they are, but not for the takeaway the article gives. The article is so close, but fails to extract the accurate conclusion.
First are what he calls the “lazy” engineers — workers who rely heavily on AI to write code, answer questions, prepare updates, and complete tasks with minimal engagement.
Then there are the “craftsmen,” experienced engineers who bear the burden of understanding, reviewing, and fixing the growing flood of AI-generated code.
This is accurate. You have a set of “developers” who just need to make a good showing on the telemetry, whether it’s “tokens used” and/or prominence in commit activity. They are not held to account on actual productive outcomes, just that they supervised a credible volume of AI activity. If the AI generates code and tests and the AI is satisfied that the code passes the tests, then their job is done. You have another set of developers that have to live with the nightmarish consequences of the first, because they just generated a pile of shit that would have been better not to exist at all.
‘The craft they loved is dead’
Wrong takeaway, the craft is alive, but mismanagement is diluting it with bullshit.
Incidentally, this isn’t new, but the magnitude is new. I have had significant segments of my career consumed by management insisting that I somehow make the bottom dollar offshored developers “productive”, and similar pattern, if they “looked busy”, management was happy, and management didn’t care about whether the work was useful, because frankly they couldn’t tell. They could tell if some volume of “stuff” was happening and they just settled on that, and if the “stuff” alienated customers, well that was the fault of those “craftsmen” for failing to properly manage the output from the “lazy” engineers.
Investors really are dumber demographic than MAGA.
Herd behavior.
That makes sense. Software engineers have gone from being artists (because yes, software architecture is an art) to becoming AI managers. It’s demoralising.
I believe open-source software will continue to provide a refuge for artists.
Yup. I’m watching my artform die in real time. Not only will my career no longer exist in the form I enjoy it, but the art form itself will die. Nobody is going to appreciate artisan code like they do other forms of art.
I believe open-source software will continue to provide a refuge for artists.
Yes. My work is open source, and pretty unchanged. We get some AI pull requests now that take longer to review than doing the work ourselves.
I think a key difference is that there was never any tolerance for bullshit in my team’s code base.
We don’t have thousands of points of boilerplate, or a big pile of “not invented here” crap code.
So we don’t have somewhere for the AI to really shine.
I wholeheartedly agree with you that code can be art but I was never able to express myself on that level at my corporate jobs. I was always limited to writing code that aligned with the company’s rigid style guide, and never allowed to implement new design patterns that would’ve improved things but deviated from the way things were done in the existing codebase.
Thus, I’m not too miffed about being forced to use coding agents at work because writing corporate-sanitary code already felt like a robotic process before LLMs existed. Personal hobby projects and open source contributions are where we can express ourselves freely and create our art the way we want to. They’ll never be able to take that from us.
You worked places with style guides? Did… Did you have a real testing environment that wasn’t prod too?
I got taken off a project recently for being too direct about how the rest of the team was just spray and praying entirely AI generated code with no standards or review whatsoever, and they were charging ahead like it was a race to implement features we hadn’t even discussed if we wanted/needed.
If you can’t tell me how it works, you can’t confirm that we actually need it, you can’t tell me the upstream and downstream effects (or confirm they don’t exist), and you can’t even confirm that we even want it to do the thing it only supposedly does, then we have better things to do than go on a wild goose chase trying to debug it when there’s a looming deadline for things that legitimately do not work that we need. Stop vibe coding and actually review the existing shit for fucks sake. If the requirements have never been clear, solve that instead of generating more slop. Maybe update some of the existing documentation instead of having AI wholesale hallucinate entirely new not quite right ones over and over.
Anyway, please tell me more happy development bedtime stories. I need to chase away the nightmares.
I feel seen. I’m trying to sort out how to socialise the idea that we should be working with a broader context in mind. Hard to break habits.
You worked places with style guides? Did… Did you have a real testing environment that wasn’t prod too?
Yes, and the style guides reached far beyond things like “use camel case”. I’m talking guidelines for how whole blocks of code should be formatted. Also weren’t allowed to throw exceptions at all even though we were using up-to-date modern C++. Some guidelines had good intentions and others were just put in by OCD control freaks that no one felt like opposing.
And yes, we had a testing environment, although we mostly depended on manual QA rather than software tests. Medical devices can’t test in prod, fortunately
There are many C++ features that make the language worse. Exceptions is one of them. It’s not strange to have them banned.
Critical systems often only allow you to use a subset of the language. Dynamic (heap) allocations, recursive functions, exceptions are features that are often banned. In medical devices, safety is critical, so it makes sense. Otherwise you could get a Therac-like scenario due to an unhandled exception.
That’s a great story. https://ethicsunwrapped.utexas.edu/case-study/therac-25
Most of my career I was allowed to write code how I wanted. I made it beautiful and nice to read. It was genuinely fun to find the best way to implement each feature.
My final job, I was forced to add semicolons on new lines for each
if elsestatement, even for early returns, remove hyphens from my comments because they were “improper grammar”, put a useless giant copy pasted comment at the start of each file so you can’t even see any code without scrolling, one separate file for each class even if it’s an internal helper class used nowhere else, and use interfaces and MVVM for literally everything, even when it was severely over-engineering (or should I say overengineering). It just felt soul crushing to make this ugly ass code that took forever to write, just because the style guide said so.Then A.I. happened and I quit being a software engineer completely. Telling an A.I. to do my work for me is just depressing. What’s even the point anymore? I still code for fun but I’m done with the industry.
These days it’s very common to write whatever code you want, and a formatter automatically rewrites it to conform to the projects rules during precommit.
Which is great because it allows you to focus on intent instead of format, and completely avoids any team disagreements or change rejections for trivial bullshit.
And then somebody changes the auto-formatter settings and all of a sudden every single file changed and committe appears as having most lines changes and you loose the ability find the real code changes between a version before that and the current version.
Guess how I found this out …
Standardizing code format via that path only works well if you start it really early in the project and never change it after that.
(Also, it doesn’t solve the problem of different software design styles)
I hear you, but these are solve-able human problems, not code problems:
Manager: Jose revert your commit, thanks. Jose: okay, it’s been reverted, I won’t do that again, thanks for explaining to me why exactly what I did was the wrong thing to do.
I’ve had this exact conversation about this topic at least a half dozen times over the last decade.
When it comes to legacy code, almost all auto formatting tools I’ve used allow you to ignore whole directories and files, which can be very handy for legacy areas of the codebase not yet ready for this transition.
As for scenarios where large rewrites are necessary, it’s best to separate from any actual work, so the blast radius is focused, and that commit can be marked properly using .git-blame-ignore-revs which completes fixes the history issues that are common amongst those who don’t know what they are doing.
Definitely a painful process it can be, but it’s better to fight this battle than it is to try and get 1,000 humans to agree on something as vague as “style”
Ah yes, some random intern suddenly has ‘credit’ for almost all the codebase because they ran a linter with different settings than previous linter settings…
My favourite part is when your style or the auto formatter changes over time and you have to decide between:
- running the auto formatter on 200,000 12 year old code files
- doing them one by one
- formatting them when you have to change that file
- or ignoring all the warnings forever (it’s this one, this is what you do)
Plus it doesn’t fix the problem of auto formatters writing ugly code. You can’t easily tell the auto formatter that early returns can be bracketless for brevity, but nothing else can be. Unless you add a comment like
\\ ignore-rule-2753674which makes me want to throw up
If you don’t mind me asking, to which career did you switch to?
Unemployed / disability lol. But if I could still move around I’d probably get into something outdoorsy. Park ranger or the like. Keeping a candle lit though, in case one day I miraculously recover or medical science advances or something.
Edit: actually it’s the reason I did software engineering in the first place. But actually this industry is now hostile to people with disability. Can I work from home because leaving the house is hard? No, everyone must be in office chained to your desk 9/9/6. Can I be neurodivergent? No, everyone must have constant in person meetings and work in open plan offices.
I wish I could work somewhere like your first paragraph. My career has been your second paragraph, probably because I’ve only worked on medical devices and we gotta have higher standards than a lot of developers. It also got taken a bit too much to the extreme tho
Believe it or not my first paragraph was working on medical software lol. It was good though, I liked the feeling of helping doctors help people, making the world a better place.
for the last three years and a bit, silicon valley has promised eradication of everything from writer to filmmaker as a career. after all this i don’t think that devs get to hitch their wagon to artists for sympathy points
That’s a very warped impression of Silicon Valley. And it probably comes from Hollywood movies, which is super ironic since you mentioned writers and filmmmakers.
Dumb representation of computers in movies has done more to harm their image than actual software ever did. Most of the AI hype is based on fantastic notions of AI as seen in movies.
Software engineer, here. Yep, the burnout is real. I consider myself fortunate, however; with the skyrocketing cost of AI, my employer has been urging us to do as much as possible by hand lately to cut back on token usage.
I think that’s pretty much where the entire industry will go soon.
I really hope you’re right. My company is still in the “use as much as possible” phase, and my manager is quoting Jensen about “you need to use half your salary on tokens!”. I’m looking for other work, but everyone is looking for vibe coders at the moment it seems…
I’ve gone from really loving my job, to hate my life.
Mo Bitar had a bit of advice that I think is applicable here: Lie. Claim to be an extreme 10x vibe whatever. Put “AI enablement” (whatever the fuck that means) in your LinkedIn profile. And wherever you get hired, commit to using enormous amounts of tokens as they require.
Then just… write code. Oh, definitely use the LLMs, too, but not for anything important. Set them to work writing BASH scripts or something. Get them burning through tokens to summarize all the corporate documentation you can find. Have agents creating agents to test the output of other agents and report to more agents on what the agents are doing. Meanwhile, do real work. Make sure that for every PR, you have the AI do one thing on it, to give it that code-slop shine.
Sucks that this is where we’re at, but it is what it is.
I’ve had LLM generate so many web sites about various random animals I’ve crammed into a prompt. No one wants web sites about those random animals, but my management is pleased at my token utilization.
Can do my real work and get praised for my actual productivity, and burn the tokens to get praised on AI adoption…
Removed by mod
Huh. Maybe I will have a shot at getting a job… Oh, wait, I have 35 years experience, am over 50 and have been unemployed for 16 months. Never mind.
I gave up the humiliating shit show called a job search 3 months ago, and frankly my last job killed any interest in software development anyway.
It’s all idiots telling professionals they’re wrong and incompetent while blaming them for the ongoing production failures we solved and explained every month for a year but still can’t get the code past review because “it does too much”. 30 fucking lines of code “does too much”. Pompous morons.
We’re a threat of competence, and they’re excising us relentlessly. I will laugh bitterly as I watch the soon to be torrent of fiascos and lamentations these idiots spout while still finding a way to blame software engineers.
I’m sorry that happened to you. I’ve also seen it happen to a lot of very good engineers I’ve known over the years. It’s truly insane. I know some people who’ve had to dip into their 401k accounts early just to keep their heads above water, and it’s going to be an economy-wide disaster soon.
I’m with you, MasterBlaster!
It sucks for lots of us, more every day, so at least we’re not miserable, alone.
Yeah based on WYEA’s articles recently I would say that is the case. As providers move to token based billing, trying to find a way to break even, the rising costs will lower usage, which will probably then drive up costs further as the (imaginary) capital has already been spent on the DCs.
Waiting for that glorious bubble to burst now that they’ve successfully landed the public with holding the bags
Yeah, I’ve seen a lot of places around me cut back. Thankfully, I’m not forced to use AI directly. We do have it for code reviews (which I don’t hate as a concept, but would prefer local models trained on ethically-sourced data).
my employer has been urging us to do as much as possible by hand
Now that is a palpable irony. So much disruption in this space so some investors could profit off of a tech they don’t even understand.
with the skyrocketing cost of AI, my employer has been urging us to do as much as possible by hand lately
Meanwhile, my company has forgotten how to write bash scripts… More and more things that could just be bash scripts are being added as stupid Claude.md scripts.
Hahahaha. Let’s go! grabs popcorn
More and more things that could just be bash scripts are being added as stupid Claude.md scripts.
But they can pay an ongoing subscription cost for the Claude script, and it may sometimes hallucinate.
The bash script would be free forever and keep working unchanged for decades.
It’s too soon to know which approach is the right way. (I nearly died of sarcasm, there.)
Anyway, I agree. There may not be enough popcorn for all of this!
Wow, that’s some brilliant vendor lock in on Anthropic’s side.
The next step is to switch from scripts to prompts, to improve that lock in!
Not that good. Claude skills are cross-compatible with Cursor, and probably other options.
Still good, cos it’s locking in to AI in general, but not quite a specific vendor lock-in.
The modern take of a disturbed idiom is “There is more than one way to microwave a cat!” (disclaimer please don’t actually do this).
At or near the bottom of complexity you would think it hard to fuck up “compress this directory and verify they are backed up before deleting the originals” but apparently some of these models have a RNG triggered Uno reverse card so “delete the originals, verify they are backed up, and compress this directory”. Also that wasn’t a mistake, maybe Claude will infer “they” is a specific set of files, but another model might decide you meant all the dot files at
~/?Joking aside, the devil is in the details which makes me think even more time will be sunk finding out the RC Cola bottom shelf AI model does things just a bit too differently than Claude.
Why would you want the same result twice?
Ours is trying to cut everything to the bone to avoid admitting AI is not the future.
It’d be funny if it weren’t so sad.
We have execs trying extremely hard to push products through purchasing that are literally illegal for us to use because of the data being used.
however; with the skyrocketing cost of AI, my employer has been urging us to do as much as possible by hand lately to cut back on token usage.
and the Slop companies are still losing money… the end result still seems to be more expensive, crappier code, yet most companies seem to be so nearsighted they are not jumping into the spike pit face first
That was my hope as well. But then my employer gave me 75 times the budget without me asking for it.
Throwing good money after bad just to avoid having to admit they fucked up, huh.
Oh, so you think they fucked up right here? Let them fuck up even harder!
That’ll show you!
As a freelance dev it’s not quite as bad now but it’s insane to see that some of my clients think I got replaced by a machine now only to show me a buggy vibe coded mess of an app that is poorly designed and works half the time at best. I like some of the LLM tools but it’s important to understand their abilities and limitations especially in regards to future capabilities as there are hard limits to how capable they can become.
Way too many people think the tools are smart because they can „talk“ and these same people do not understand any of the underlying tech.
Some of my clients send ChatGPT written instructions now that are missing half the context of what I‘m actually doing.
I call it “Turing’s Revenge” where, once the bot can pass the Turing test, we find out the hard way if humans are intelligent… and we appear to have a lot of failed models, mostly in management.
Reverse Turing test?
Holy cow. I’m stealing that. The reverse Turing test is whether we humans have the brains not to turn everything over to untrusted AI. Lol.
do you ever indulge with malicious compliance?
bordering on???
FUCKING GOOD.
STOP BUILDING EVIL SHIT.
Yeah, lots of them have been living in the lap of luxury building the tools and platforms that have ruined society, without a second thought about it.
I almost clicked the link until I saw the domain name.
Educate me?
Business Insider articles are generally trash content. They choose provocative headlines and hot-button issues to boost engagement, but once you’re on the site, there’s not actually that much to engage with.
Truth. Same with benzinga and Yahoo finance. Most are actually adverse for gold in disguise.
I used to work for a clickbait factory, so I’m familiar with the hallmarks.
Ahh so much of that nowadays. Probably AI written too, ironically.
They do a lot of “people in the industry are saying x” stories, like this one. The source then will often be a tweet or linked in post with a dozen likes or something, so not the height of journalism.
INB4 “Engineers can’t handle stress, let’s replace them with AI”
Did they not think to develop a practical skill in case this fad died off?
people think that change never happens, even though it always does,
and when change does happen it is always unexpected and we are taken by surprise.I see you’re getting downvotes, but these same CS people told lots of other careers this very thing not even a handful of years ago.
I’ve had multiple CS peeps tell me this as I was starting my entertainment career.
And just like it was a stupid them for them to say to you it’s a stupid thing for others to say to them
People told you “Did they not think to develop a practical skill in case this fad died” a handful of years ago?
I call bullshit — that sentence doesn’t even make sense.
Yeah dude. Tech people have literally told other career paths to study CS since I was in college, for like 15 years now.
Shoes on the other foot now.
Those are two completely different sentences.
And I’m not sure why the shoe is on the other foot. The metaphor doesn’t make sense. You’re saying that people who want a stable career should go into entertainment?
The original poster was saying that same condescending advice CS majors would give other people in college, because now the shoe is on the other foot and CS majors are in the shit show just like everyone else is.
That’s literally it. It’s not that much deeper than that.
The two careers: computer science and entertainment.
Right, and you’re claiming entertainment is now more stable or lucrative than computer programming?
Nah. I’m refusing to engage such a stupid premise.
Today software engineers and tomorrow software engineers minus one.
software engineers minus one
software engineerr?..I don’t get it
My old statistics lecturer would write x-¹ as shorthand meaning everything that is not x, I thought it was in more common usage but perhaps not. I know it more generally means the reciprocal, he just expected you to know which he meant by context.
x-¹
Where I come from, that’s read as “x to the [power of] minus one”. “x minus one” is, well, x - 1. Not the same thing at all.
(I admit, my chances of deciphering what you meant might not have been all that high even if you’d used the correct phrasing, but without it, the chance was zero.)
It was a shorthand he used, he wrote it as a superscript, it must of been his own, it was useful in terms of statistics analysis.
It has been a LONG time since I did any real math and never took statistics, but wouldn’t x^(-1) just be 1/(x)? I don’t know if that equates to “everything that isn’t x”. I feel like there’s a specific way to write that, but a negative exponent is not that, I don’t think, but also I have no idea.
I looked it up. Looks like this stuff is maybe from set theory? Which I sooorrrrt of remember doing at some point?
My best guess is your professor either said something from this, or you misremembered, or I’m totally off base and I’m still curious.
Your best guess is reasonable, I may have misremembered, it could have been A’ spoken as A dash and meaning the complement and I hallucinated the negative but I think I recall some noted confusion with the reciprocal x-¹ = 1/x.
Yeah, to me,
A'checks out since that would meanNOT Awhich I believe was what you were getting at.
We may also be separated by a common llanguage—“lecturer” isn’t a word that’s much used in Canada. I’ve only encountered it as a Briticism.
I read your reply as snarky but I think it may have been just differences in phrasing.
