I'm a software developer in Germany and work for a small company.

I've always liked the job, but I'm getting annoyed recently about the ideas for certain people.....

My boss (who has some level of dev experience) uses "vibe coding" (as far as I know, this means less human review and letting an LLM produce huge code changes in very short time) as a positive word like "We could probably vibe-code this feature easily".

Someone from management (also with some software development experience) makes internal workshops about how to use some self-built open-code thing with "memory" and advanced thinking strategies + planning + whatever that is connected to many MCP servers, a vector DB, has "skills", a higher token limit, etc. Surprisingly, the people visiting the workshops (also many developers, but not only) usually end up being convinced by it and that it improved their efficiency a lot and writing that they will use it and that it changed their perspective.

Our internal slack channels contain more and more AI-written posts, which makes me think: Thank you for throwing this wall of text on me and n other people. Now, n people need to extract the relevant information, so you are able to "save time" not writing the text yourself. Nice!!!

I see Microsoft announcing that 30% of code is written by AI which is advertisement in my opinion and an attempt to pressure companies to subscribe to OpenAI. Now, my company seems to not even target that, but target the 100%????

To be clear: I see some potential for AI in software development. Auto-completions, location a bug in a code base, writing prototypes, etc. "Copilot" is actually a good word, because it describes the person next to the pilot. I don't think, the technology is ready for what they are attempting (being the pilot). I saw the studies questioning how much the benefit of AI actually is.

For sure, one could say "You are just a developer fearing to lose their job / lose what they like to do" and maybe, that's partially true... AI has brought a lot of change. But I also don't want to deal with a code base that was mainly written by non-humans in case the non-humans fail to fix the problem......

My current strategy is "I use AI how and when ->I<- think that it's useful", but I'm not sure how much longer that will work..

Similar experiences here? What do you suggest? (And no, I'm currently not planning to leave. Not bad enough yet...).

My teamleads too. They tried to sell a whole application that they vibecoded and marketik strategie was "fully ai generated!"

I am making similar experiences, but is is not as bad as you are describing it yet. We have a new member in the team who is not a developer by himself, but he has gotten the task to make our way of working more professional (we are mainly scientists and not primarily software engineers, so that's a good thing).

His first task was to create programming guidelines and standards. He created 8 pages of LLM generated text and example non sense code. He honestly put a lot of effort in it, but of course there are a lot of things in it that are wrong. But the worst thing is the wall of text. You are nailing it - it is my task now to go through this whole thing and extract the relevant information. It sucks. And I am afraid that soon I will need to review more and more low quality MRs generated by people who have little experience in programming.

Endure the next year or so, until it pops and there will be a massive need for senior devs for fixing the slop machine.

Fixing vibe code is a specialty that contractors will be able to charge a premium for here pretty soon.

Agentic use of AI didn't really work well enough until December of last year. The models and tools just improve that fast. Codex/claude (or opencode with the same top models) is what you'd need for it.

You still need to plan and define clear specifications for the model. Spend 80% of your time planning and breaking down the job into steps and it'll be pretty self-going from there.

Of course, this works best for common frameworks and solved problems or logical problems. React/node developers can easily 10x their output, and get it done better than they would by hand.

I'm working more with empirical development, so most of my time goes into studying environments and adapting to it. I get most benefit out of having agents read through logs and figure out what happened. It gets it right maybe half the time, but it's a good rubber ducky even when it goes wrong. I'd say it 2-3xes my output. But I can probably improve my usage, too.

But yeah, code review is where it hurts. If it's slop, it just takes so many rounds to get it right. Even when it's good, it's just so much code to review.

I have limited dev experience, but most Devs at my company utilise AI heavily for generic code generation.

For example, writing out the core functionality, and then having the AI propagate variables, error flags etc.

I like your point that it really should be a "Copilot". (Although ew MSoft.)

I think it's important to continue to experiment with new tools as they improve, and try to use them responsibly.

AI always produces extremely generic, slightly verbose phrases though. So it's important to generate in small enough chunks that you can use the structure it creates and then add the specificity and accuracy for your purpose. And cut out the waste.

I haven't had a similar experience yet, but maybe some if your collegues feel the same way? You could write a letter stating your concerns and let anyone sign who agrees and then send it to your manager.
Also, I'd like to add that under German law AI Output can not be copyrighted. You can only claim coownership or something. Maybe that could be interedting to your managers?

Managers using influence to inflate their importance is common. It sounds like they are hoping to head off innovation to protect themselves.

The second aspect is their harnessing of useful idiots. I think critical or practicing objective thinking is being pushed aside at the time when needed most. You probs have to tread carefully and maybe influence people to use their own minds not that of desperate managers. Alternatively set the useful idiots off in competition and watch the fireworks. Not helpful but it is fun.

There always some people are always looking for an easy fix, a way to rise quick, to be a yesman. But for the most they just wanna do a good job. Maybe appeal to the rationally of those people. Trust your brain and skill rather than cult practises.

Let me give you a little parable.

There once was a juggler, who could juggle with three balls all day. Then someone from the audience threw a fourth ball, and he kept going. Someone threw a glass, then a flaming torch, and he kept going, occasionally burning his hands. Seeing he could do it, someone throws a machete, and the juggler almost never cuts his fingers keeping all those things in the air. A chainsaw gets added, and an open bottle of bleach, and occasionally the juggler gets his hair caught or spills some bleach, but he keeps going. As he keeps going, people keep adding more and more things. Eventually it's too much, and it all comes crashing down, killing the juggler and several members of the audience and destroying all the objects in the air.

On the next street corner, a juggler stands with three balls. Someone from the audience throws in a fourth. He steps aside and lets it fall to the floor, happily juggling three balls.

I had a manager who pushed AI a lot. When he left, all the pressure to use it seemed to die down. So maybe it's just a couple of people creating this environment and if you can get away with avoiding them it's better.

The problem with AI code we saw is that often no human has actually looked at it. During reviews you won't check every line and you'll have to trust much of the code that seems to do obvious things. But that assumes it was written by a human you also trust. When that human hasn't reviewed the code either, you end up with code no one in the company has seen (and may not even know how it works).

If i saw a fully AI generated software i would cower in fear (for my computer safety) and run away

A small company this level of cooked and immature leadership may not have the resources to recover long term from the damage. Even if the bubble pops on them in a year.

I'd start looking for alternatives now, before it becomes urgent.

Remind them that copyright cannot be enforced on anything AI-written.

Murder.

Let him "vibe-code" himself into a problem, then tell him you can't fix the mess he done did.

I fail to see the relevance. OP is not talking about burn-out...

AI is inevitable in many fields but as usual people expect way too much from it. It's a tool, not a magic wand. I agree with you that it is useful and even powerful when used by somebody that understands when it is useful. But it is dangerous when wielded by somebody that doesn't. As others have said, let your boss vibe code themselves into a corner and leave them to vibe code themselves out of it. They will try to deflect that it is your job to solve it though so you better come up with a strategy to handle that. Be sure to have more people on your side in this venture.

Secondarily, install a chatbot with the instructions to derive essence as a bullet list from your boss wall of texts. If they make their life easier with LLM, so can you. If there are misunderstandings it's either all just ghosts in the machine or a failure from your boss to communicate clearly.

Is this free Software to use for everyone? Would that mean no Code at all was made by the Company? It's all stolen from everywhere.

For centuries, we spent less effort consuming content than it took to produce the content.

Good teammates and content producers understand that their content needs to have an intrinsic value and benefits beyond the mere existence of content.

If you make your team 10x slower by sending LLM generated 10 page content instead of a one liner, you are actively hindering the team.

Efficiency is not just production of content (slop or not), but the overall system. That's why the corpo speak has always been such a waste. Too many words to say nothing in a mandatory all-hands. Now the dial is up to 11 with the same time waste everywhere

Reading a wall of text to extract a simple concept which turns it to be wrong seems very appropriate for this thread, just perhaps not in the way they intended.

The best case for AI is that can heroicly solve problems that shouldn't exist in a first place in a well-managed company.

It can structure unstructured data, that should be structured. It can do simple coding where aboslutly no coder could be reached. It can sum up the red tape and produce red tape to meet complience requirement that were where created for the sake of the red tape alone.

If the work is important, it shoudn't be done by AI. If its busy work, it shouldn't exist in a first place. No one would need AI to summarize anything if everything was writen clearly.

And thats best case scenario when you don't count halucynations, errors, edge cases, company specyfic context - things that results in task taking more time after all the "oh it produced something and it looks good at a glance" cloud fall down.

If your company is getting a lot of value from AI than your company is trash.

Our internal slack channels contain more and more AI-written posts, which makes me think: Thank you for throwing this wall of text on me and n other people. Now, n people need to extract the relevant information, so you are able to “save time” not writing the text yourself. Nice!!!

I think this is one of your best bets as far as getting a real policy change. Bring it up, mention that posts like that may take less time to "write", but that they're almost always obnoxiously verbose, contain paragraphs that say essentially nothing, and take far longer to read than a hand-typed message would. The argument that one person is saving time at the expense of dozens (?) of people losing time may carry a lot of weight, especially if these bosses are in and read the same Slack channel.

Past that I'd just let things go as they are, and take every opportunity to point out when AI made a problem, or made a problem more difficult to solve (while downplaying human-created problems).

enjoy the ride, look for a new job, so you're not surprised when the buisness dies due to declining quality or a customer lawsuit suing for damages.

We had a dev drop a combined total of 8,300 lines of readme files into the code base over a weekend. I want to nuke all of them, my boss suggests reviewing and updating them.

I try to push on the maintenance aspect. Developing something new is easy, and my company does do that, but the group I'm in is primarily doing maintenance on existing software. Bug fixes, feature additions, etc. If we generate applications entirely using LLMs, none of us will be experts on the applications we push to the customers.

They push corpo buzzwords like "responsibility", but who takes responsibility when no one has done the work to begin with? It feels like a liability nightmare, and the idea of sitting there cleaning slopcode just isn't very appealing to me.

Your entire comment echoes my thoughts. Things aren't exactly improved by the idea of adding LLMs to the review process either. Gods.

The point was that the more you keep compensating for other people's dumb moves, the greater the damage when it all inevitably comes crashing down.

In other words, just do what they ask, get them to sign off on it and watch it crash and burn in an unmaintainable, unsecured mess

What do you think im doing from that company?

No :)
Yes :)
And its not for purchase because nobody wanted it and they got ghosted from another company lol

Do better than the AI

Oh, they fear the slop. :D

That last line belongs in a horror novel

I have a similar situation at work. But I'm a designer. Previously I have worked for a company that did research in ai, from a product standpoint where I also had a boss who thought ai can do anything and blamed the workers for not succeeding.
Now Im beginning to see the same thing here in my new company. The boss, product manager and a few of the developers (actually those with less ai experience) are driving this motion that ai will just replace everything. I tried to tell them about my experience, but they didn't listen to a designer.

Now this pressure is also coming to the UX team. A few days ago the boss asked me to use ai to reskin a whole app. It was assumed by him that that's a simple thing.
Why is it so hard for them to understand it doesn't work like that?

Learn how AI works, what it is good at, and how to build trust to minimize bugs.

You will be less efficient than other coders using it responsibly if you're not using it. It WILL be another tool you're expected to know.

Tldr; Be cautious and use the LLM to increase your test coverage and start with low risk basics.

Switch employer.

(hey ich bin bei dir das kann nicht einfach jeder trivial aber aus diversen Gründen habe ich das bereits 2x in den letzten jahren)

Give it a honest shot, learn the limits, help your team establishing best practices. Depending on the language, framework and vertical it can be mor e or less autonomous.

It always needs good specs, clear steps, and lots of testing. Some of which can be automated by the AI itself. Make sure you write yourself or strictly review (am besten im vieraugenprinzip) the code that touches anything related to money or safety.

It’s all about finding that happy medium. I’m a developer myself that uses it, mainly to write the “mundane” stuff, but I always double check its work. I also have it check mine, and ask for any alternatives it may see - a lot of times they’re helpful, AND I learn them for future use. So maybe just try to push the balance rather than the all-in?

Tell him that vibecode is barely good enough for a prototype (a demo) and watch if he fires you. Either way is a good outcome.

Any idiot can write code. "Vibe coding" is just the new pasting code from stack overflow. For that matter, a lot of LLM generated code probably came from stack overflow.

Your value as a developer is not in your ability to rapidly pump out code. Your value is in your ability to design and build complex systems using the tools at your disposal.

As an industry, software engineering has not yet been forced to reckon with the consequences of "vibe coding." The consequences being A.) the increasing number of breaches that will occur due to poor security practices and B.) the completely unmanageable mountain of technical debt.
A lot of us have been here before. Particularly on the tech debt front. If you've ever been on a project where the product team continually pushes to release features as fast as possible, everything else be damned, then you know what I mean. Creating new code is easy. Maintaining old code is hard.

Everything starts out great. The team keeps blowing through milestones. Everyone on the business side is happy. Then, a couple years into the project, strange things start happening. It's kind of innocuous at first. Seemingly easy tickets take longer to complete than they used to. The PR change logs get longer and longer. Defect rates skyrocket. Eventually, new feature development grinds to a halt and the product team starts frantically asking, "what the hell is going on?"

A question to which maybe one or two of the more, senior devs respond, "Well, uh, we have a lot of technical debt. I mean A LOT. We're having to spend tons of time refactoring just to make minor changes.
And of course, unplanned refactoring tends to introduce bugs.

The product team gets an expression on their face like Wyle E. Coyote as the shadow of a falling ACME anvil closes in around him.
At this moment, they have two choices.
Option A.) develop a plan to mitigate the existing tech debt and realign the dev teams objectives to help prevent this situation again by focusing on quality over quantity. Option B.) ignore the problem and try to ram feature development back on track by sheer force of will.

Only one of these options will achieve meaningful outcome and it's not "B". Unfortunately, in my experience that's often the chosen option. The product team does not understand that while Option A impedes feature development, it's only temporary. Option B impedes feature development permanently.

We're going to see a very similar cycle with vibe coding. It just takes time to materialize. Personally, I think the tech debt for vibe codes projects will be compounded due to the sheer verbosity of LLM's and the fact that no one actually understands a vibe coded project well enough to fix it.

That said, these issues are rooted in hubris and ignorance. Failure to appreciate the "engineering" part of software engineering. This is not something you alone can change.

The AI hype is going to disappear. Probably sooner than later. Just like every other tech hype cycle before it. But, LLM's are probably here to stay so we have to make the best of it. I don't usually use LLM's for code generation. There are better tools for that already.
I do use them frequently for research. Honestly, using an LLM with search incorporated is often a lot faster than scouring dozens of websites to figure out how to do something. You still have to take the information with a grain of salt as much as you would with anything on the Internet because LLM's have no understanding of the text they spit out and will feed you incorrect information without missing a beat.

If I were you, I would focus on quality over quantity.
Closing tickets faster is pointless if you're introducing a bunch of new bugs. If your bosses don't know that already, they will learn it soon enough.

Vibe coding has its place (prototypes, simple webui), and use of ai for debugging, adding simple features can be effective. Even for more complex stuff it's getting better all the time; but one needs to keep an eye on what it's doing and refactor when things go off the rails (and ai can be a good tool for refactoring).
But ai messages that just make it look like someone is contributing are infuriating. I've responded by booking a meeting with the person who pasted ai slop to review the message because 'I want you to go over that and explain it to us'. And then when they try and start going over irrelevant crap I ask why it's relevant. And when they respond that actually it's from chatgpt and they don't know, if they don't immediately indicate how they realize now it was useless to paste stuff like that, I make sure they know my opinion (in a nice way). I haven't had the same person do it twice.

Our internal slack channels contain more and more AI-written posts, which makes me think: Thank you for throwing this wall of text on me and n other people. Now, n people need to extract the relevant information, so you are able to "save time" not writing the text yourself. Nice!!!

My last software dev employer did this, except with the "voice recording" feature. Instead of composing messages in text in a text chat (because that takes too long), he'd hit the record button and just start talking it out, then send the recording. Easy! Then the team had to download and listen to ~5 minutes of verbal diarrhea, pausing and rewinding for twice that long in an attempt to glean something useful from it. This particular kind of delusion existed before AI.

Soon? It's been on my resume for over a year.

This is where AI could be useful. Transcribing and summarising the voice recordings.

That's going to be a problem, almost like a money laundering scheme. AI can spit out content that's 99% derived from copyrighted content but is itself free of copyright.

I am not a software developer, but isn't AI code often really hard to maintain, like making changes and improvements down the road?

Maybe you could provide examples of that?

I don't know, I guess I use vibe coding as a shortcut for a one off script or tool that I need. But anything that is going to be maintained long term, I feel like we still need proper developers

We're using LLMs at the company I work at and it seems very useful in many cases but sometimes it still doesn't work. I'm a bit worried about the aspect of the code rotting by LLMs generating stuff based on existing code.

My mindset has shifted a bit, now I'm more focused on making stuff easy to find and easy to figure out patterns to use so that the codebase becomes easier to work with. There's some horrible code in the project and the LLM absolutely sucks balls at it but if it's a clean routine job such as making a table with update dialogs and actions to manipulate the data the success rate is >95%.

So yeah, don't trust it, treat it like a junior dev that got straight As in school and has never considered security. Code reviews are now where it's at.

Two words: malicious compliance.

"I don't have time to read through that much bullshit."

Maybe phrase it a little more kindly, but that's what I'd try at the very least. "I have other priorities at the moment" could work too.

It's hard to be a contrarian in these kinds of positions (I've been there, and it didn't end well), so I wouldn't be too outspoken, but at the same time, try to innocently point out the issues with approaches like this. I would just try to point out the flaws in this approach, the same that we would for any other kind of programming fad - without making it seem like it's an agenda, of course.

For example, any time teams are looking for feedback - code review, retrospectives, etc. - just point out the flaws on why vibe coding is a bad idea and bring it up casually when the time comes. It doesn't hurt to be honest as long as you don't come off as being an ass about it.

Adapt