The Software Engineers Paid To Fix Vibe Coded Messes (404media.co) 50
"Freelance developers and entire companies are making a business out of fixing shoddy vibe coded software," writes 404 Media, interviewing one of the "dozens of people on Fiverr... now offering services specifically catering to people with shoddy vibe coded projects."
Hamid Siddiqi, who offers to "review, fix your vibe code" on Fiverr, told the 404 Media that "Currently, I work with around 15-20 clients regularly, with additional one-off projects throughout the year. ("Siddiqi said common issues he fixes in vibe coded projects include inconsistent UI/UX design in AI-generated frontends, poorly optimized code that impacts performance, misaligned branding elements, and features that function but feel clunky or unintuitive," as well as work o color schemes, animations, and layouts.)
And others coders are also pursuing the "vibe coded mess" market: Swatantra Sohni, who started VibeCodeFixers.com, a site for people with vibe coded projects who need help from experienced developers to fix or finish their projects, says that almost 300 experienced developers have posted their profiles to the site. He said so far VibeCodeFixers.com has only connected between 30-40 vibe code projects with fixers, but that he hasn't done anything to promote the service and at the moment is focused on adding as many software developers to the platform as possible...
"Most of these vibe coders, either they are product managers or they are sales guys, or they are small business owners, and they think that they can build something," Sohni told me. "So for them it's more for prototyping..." Another big issue Sohni identified is "credit burn," meaning the money vibe coders waste on AI usage fees in the final 10-20 percent stage of developing the app, when adding new features breaks existing features.
Sohni told me he thinks vibe coding is not going anywhere, but neither are human developers. "I feel like the role [of human developers] would be slightly limited, but we will still need humans to keep this AI on the leash," he said.
The article also notes that established software development companies like Ulam Labs, now say "we clean up after vibe coding. Literally."
"Built something fast? Now it's time to make it solid," Ulam Labs pitches on its site," suggesting that for their potential customers "the tech debt is holding you back: no tests, shaky architecture, CI/CD is a dream, and every change feels like defusing a bomb. That's where we come in."
Hamid Siddiqi, who offers to "review, fix your vibe code" on Fiverr, told the 404 Media that "Currently, I work with around 15-20 clients regularly, with additional one-off projects throughout the year. ("Siddiqi said common issues he fixes in vibe coded projects include inconsistent UI/UX design in AI-generated frontends, poorly optimized code that impacts performance, misaligned branding elements, and features that function but feel clunky or unintuitive," as well as work o color schemes, animations, and layouts.)
And others coders are also pursuing the "vibe coded mess" market: Swatantra Sohni, who started VibeCodeFixers.com, a site for people with vibe coded projects who need help from experienced developers to fix or finish their projects, says that almost 300 experienced developers have posted their profiles to the site. He said so far VibeCodeFixers.com has only connected between 30-40 vibe code projects with fixers, but that he hasn't done anything to promote the service and at the moment is focused on adding as many software developers to the platform as possible...
"Most of these vibe coders, either they are product managers or they are sales guys, or they are small business owners, and they think that they can build something," Sohni told me. "So for them it's more for prototyping..." Another big issue Sohni identified is "credit burn," meaning the money vibe coders waste on AI usage fees in the final 10-20 percent stage of developing the app, when adding new features breaks existing features.
Sohni told me he thinks vibe coding is not going anywhere, but neither are human developers. "I feel like the role [of human developers] would be slightly limited, but we will still need humans to keep this AI on the leash," he said.
The article also notes that established software development companies like Ulam Labs, now say "we clean up after vibe coding. Literally."
"Built something fast? Now it's time to make it solid," Ulam Labs pitches on its site," suggesting that for their potential customers "the tech debt is holding you back: no tests, shaky architecture, CI/CD is a dream, and every change feels like defusing a bomb. That's where we come in."
Up next (Score:2)
Re: (Score:3)
And after that: Extremely expensive and rare experts fix double-vibe messes. And probably do so by extracting a spec, throw away the whole mess and reimplement in a sane fashion.
Methinks this is just another crappy and idiotic way to make software cheaper than possible. And like basically all such past attempts, it will end up being very expensive. But most humans are stupid and greedy, cannot think strategically and cannot learn from history. Hence this crap will continue, probably until a time when we ha
Re: (Score:2)
Look! It's a vibe-coded version of the Godel Escher Bach "strange loop."
Re:Up next (Score:5, Interesting)
I wouldn't say I do a lot of rewrites, but I would say I do have to.... do a lot of pretty difficult debugging.
Mostly, I think, because the style the LLMs use are often pretty foreign to my own.
Whether or not the company is saving money is an open question that only they know, really.
I'm still employed, I'm still expensive. The kids producing the LLM-genned code are very cheap, though.
About half the price as an entry level software engineer at our organization.
I wouldn't say they're equivalent to an entry-level, but they're also not terribly far from one. With an entry-level, I mostly fix problems of the "This code is sound, but your design is fucking idiotic" sort. With LLMs, the design actually tends to be quite decent, it's just riddled with small errors. Not "it doesn't compile" errors, but things using binary orders of magnitude for bitrates instead of SI.
Re: (Score:3)
Re: (Score:2)
80% of software in maintenance.
LLMs are capable of that as well.
Recompute.
Just off the cuff, if a kid with an LLM can do what I do in half the time, and it takes me half that amount of time to make it good, that's a 25% savings in time.
If you factor in dollars, then it's scary.
Whether or not we're building a house of cards that's going to collapse in on itself is entirely another question.
Fortunately right now, there's a real upper bound on the code complexity that even frontier LLMs can handle before they start going nuts and ma
Re: (Score:2)
80% of software in maintenance.
LLMs are capable of that as well.
Not really. Your code needs to be very simplistic for an LLM to even figure out what it does, let alone propose changes that make sense.
Re: (Score:2)
Not really.
Yes, really.
I use them on a regular basis, you don't.
Where maintenance broadly means adding features and keeping it running as platforms change- they can.
Your code needs to be very simplistic for an LLM to even figure out what it does, let alone propose changes that make sense.
It depends on size.
There's a complexity-context size trade-off.
It can handle a lot of complexity if it's looking at a small piece of code (tens of thousands of tokens)
There is significant degradation in performance and ability to handle the complexity past that, though.
The agentic tools that function really well succeed and figuring out what it can
Re: (Score:2)
Even if it seems to save some money (probably not THAT much in the end), it'll still cost them.
In 10 years, the Vibe coding kids will be middle-aged vibe coders, but the entry level engineers would have been senior level engineers. Eventually, once you were ready to retire or move to management, one or more of them would have been the new you.
Instead, now when you retire, they'll be swimming in a sea of middle aged vibe coders and nobody left will have a clue how to fix the horrors that they produce. They w
Re: (Score:2)
I bet most of them are already vibe coding to fix vibe code problems.
It's AI all the way down
Re:Up next (Score:5, Insightful)
I've actually fixed a lot of AI-generated code. I'm not buying that AI *can* fix vibe-coding messes. The problem is that, when AI suggests fixes, it introduces NEW dumb problems in its "solution."
Re: (Score:3)
This is not a problem with AI, it's the inherent design of the models.
Output is effectively a dart thrown at the dartboard, with a wide error distribution. Fixes are outputs thrown at the dartboard, with the same error distribution. It's a stationary process which must reproduce the constant variance throughout the iterations.
The outputs will come arbitrarily close to the target eventually but the number of iterations needed is exponential. In practice, the human asking for another iteration run out of
Breaking new: Professionals Fix Amateurs' Work (Score:5, Insightful)
Re: (Score:2)
dozens of people on Fiverr
Dozens of them! Dozens!!!
Businesses were shocked to learn that amateurs really do not know what they are doing.
Do you *really* think it was *professional businesses with business logic* that were paying people on fiverr to fix vibe coded messes? Its literally sales guys and PMs who want to do something without going through the legitimate business process. How is this different from before when those same fools would watch a youtube video from some kid who promised to help you build a website with python and javascript in less than an hour? Other than AI makes it 1) more accessible and 2) ea
Re: (Score:3, Informative)
Its literally sales guys and PMs who want to do something without going through the legitimate business process.
At one of my companies, one of the marketing managers decided to use his department's budget to buy desktops directly from Dell instead of going through IT to get his employees computers. His justification was it was cheaper to buy from Dell directly than request Dell computers through IT. Except he bought consumer models and didn't know why that was a bad idea. So they were not covered by the company's service contract with Dell and the IT department had to support them out of pocket. Also being consumer m
Re: (Score:2)
Re: Breaking new: Professionals Fix Amateurs' Work (Score:2)
How is this different from before when those same fools would watch a youtube video from some kid who promised to help you build a website with python and javascript in less than an hour?
Because before, most of them would give up after the coke wore off having accomplished exactly nothing. Now, they have managed to actually get a large quantity of pure slop shoved into a production environment which is a lot more difficult to clean up.
Re: (Score:2)
Indeed. That nicely sums it up. The most important skills any professional has is that they have a keen understanding of their own limits. Amateurs lack that and make a big mess. Add that software as a discipline is both unfinished and more difficult that most other engineering and does not have adequate liability requirements, and you get a nice explanation for the insecure, unreliable and expensive mess we routinely get in that space.
Re: (Score:2)
To no one's surprise, people with knowledge and experience often have to fix what people without them did. Businesses were shocked to learn that amateurs really do not know what they are doing.
IANAP so I could be talking out of my ass; but I think that If it's properly used, "vibe coding" - followed by REAL coding - might be useful.
It seems to me that sometimes one of the hardest things about programming is nailing down a functional spec, which often changes after coding has begun. Vibe coding allows for quick-and-dirty test apps that can pinpoint flaws in thinking and can be changed quickly without investing time in making the code stable and reliable. Once you have something with the right feat
Re: (Score:3)
Re: (Score:2, Informative)
To no one's surprise, people with knowledge and experience often have to fix what people without them did.
Well... Some people [wikipedia.org] will be surprised.
As expected (Score:2)
Genuine progress in AI is being made and experts are finding good uses for some of it.
Pundits, hypemongers and lying salesweasels tell a fantastical, fictional story about "vibe coding".
Clueless executives believe it with predictable results.
Re: (Score:2)
One thing LLMs can do well is search on steroids, including some adaption, but minus reliability and often minus references to the source (which is a massive problem, because suddenly your code may belong to somebody else...). Hence they are powerful tool without adequate safety features and that makes them "advanced experts only".
Coding a project with a genuine understanding of the whole is not something that LLMs / search can do, not matter how high-powered. But there are tons of people that cannot do it
Re: As expected (Score:5, Insightful)
I've said thing multiple times through my career:
- computer do what I say
- computer do what I mean
- computer do what I need
These are usually 3 different things. With more junior devs only delivering the first version.
The current AI agents are only doing the first one while a competent experienced software developer will do the later one. It does require to ask questions and clarification. AI will at best do something similar to "what I mean", but since it has zero common sense is pretty much incapable to actually grasp the meaning and the needs of the requests.
Not a new angle (Score:3)
I far preferred other clients - almost by definition, these people are cheap, and the sorts of asks they could come up with could be utterly absurd. ("No, I cannot build you an eBay replacement with built-in voice calling for $500.") But sometimes you can't be picky.
Re: (Score:2)
I've worked on projects like that, things that were first written by amateurs and then they thought they could bring in professionals to clean it up and turn it into something robust. Years into the project, they were still dealing with the consequences of the bad original design. You can't take a mess and retrofit a coherent architecture into it. It's almost always better to start over, writing something that's properly designed from the beginning.
Step 3: profit? (Score:2)
Step 1: pay AI for "code" ...
Step 2: pay actual dev to fix "code"
Step 4: give CEO bonus!
AI is a Multi Trillion Dollar Tulip mania (Score:2)
The more interconnected technology enables us to become, the bigger the possibility for us to make a fool of our collective selves. The AI craze is just a repetition of the endless hype cycles we have been through in the past many centuries. It was the printing press that enabled early hype cycles, then radio and TV. Now it's internet and social media.
I have friends who haven't had to work since the dot-com bubble went bust. They caught wind of the dot-com hype, made the most out of it and bailed out in tim
Re: (Score:2)
Yes, since the dotcom bubble burst, IT professionals have found it difficult to find work. All that internet stuff has become so irrelevant that you might as well learn to herd sheep. Oh wait...
If you aren't part of a startup that promises things they don't know how to achieve yet, you'll be unaffected by any bubble bursts. Every large company from before the dotcom bubble is still a large company today.
Re: (Score:2)
A lot of large companies have been acquired and no longer exist as such. AOL for example, to cite just one.
Even Sun, who called themselves the dot in dotcom, no longer exists. Oracle absorbed them, but much of the technology - Solaris and Sparc - did not survive. The reality is that many companies come and go, and this is no less true of large tech companies.
I only mentioned a couple that I worked at, but there are countless examples such as DEC.
Re: (Score:2)
Is it bad for the company or their employees when someone acquires them?
No, the problem with popping bubbles is, that all the small startups lose and some investors and traders make a large loss. But the large companies and the technology survives. The problem is not with the tech, but with the stock markets.
Re: (Score:2)
Yes, it's bad for a lot of employees, because acquirers rarely want to keep 100% of the business going. Many employees are let go.
In the case of Sun, it was a distress sale, so those employees were SOL - there was no appreciated stock price to benefit from, for those who held on to their shares - which I fortunately did not. And look where Solaris and SPARC are today - nowhere. So no, many companies did not survive, and their technology did not survive.
Bubbles are indeed bad, but that's a separate issue fro
Re: (Score:2)
I agree, some may lose the job. But in the end, the industry survives. And the people who lose the job short term belong to the people whose expertise is needed when the merged company hires again the next time when the merge is finished and they are growing. Being skilled in a new tech is always good, even when you next job does not need that exact skill.
Re: (Score:2)
Just Java was an inspiration to many.
Oracle bought it up.
And snuffed out all the joy.
Re: (Score:2)
My thoughts exactly.
not going anywhere (Score:5, Funny)
"Sohni told me he thinks vibe coding is not going anywhere..."
A claim with multiple meanings, all true.
Fraud is one of the world's oldest professions.
Hire the expert to vibe code (Score:1)
Cut the middleman.
The expert will be both faster and spot the mistakes before it becomes complicated and expensive to fix them.
Of course, the vibe code fixers currently make more by waiting for amateur vibe coders to create the real mess, so they have more work. But in the long term vibe code fixers will be no longer needed because of better AI, but people good at vibe coding will.
Re: (Score:2)
Or have the vibe coders check their stuff for errors and inefficiencies before committing it. Except that vibe coders were hired to fill in once the AI was installed and all the actual devs were let go.
So, ignore my suggestion and just let your company drown in the AI cesspool.
Re: (Score:2)
we would surely like to have more than an amateur
The amateur/professional distinction isn't very good at predicting job performance. The ark was skippered by amateurs, the Titanic by professionals. At any rate, if wages (cost) is proportional to coding skill, the primary goal of Vibe and other AI nonsense is to minimize it. Noboby is going to hire real talent just to have them sit around only doing code reviews.
On the other hand, this does make a good business case for hiring in a consulting firm to do code reviews and fixes. The overhead for the talent
At least based on the names in TFS (Score:2)
At first glance, it appears all the people laid off by the big Indian off-shoring companies might now be pivoting to Indian vibe code cleanup companies.
"Poorly Optimized code" (Score:2)
"Siddiqi said common issues he fixes in vibe coded projects include inconsistent UI/UX design in AI-generated frontends, poorly optimized code that impacts performance, misaligned branding elements, and features that function but feel clunky or unintuitive,"
Seems to me that "poorly optimized code" is a bit of an oxymoron.
Re: (Score:2)
Seems to me that "poorly optimized code" is a bit of an oxymoron.
How is that an oxymoron? Code that works does not always mean it works well or reliably. For example, some AAA game titles in the last several years have launched and ran poorly on the hardware. It sometimes takes a few patches before those games are playable. In the realm of video games one legendary code optimization is Carmack's Reverse [wikipedia.org] which was used to generate realistic shadows. id software's John Carmack discovered the technique independently of William Bilodeau and Michael Songy from Creative Labs.
The best scam (Score:1)
I once worked for an environmental cleanup organization. Then we merged with a conglomerate full of polluters. In the gallows we joked that we were being paid to clean up our own messes.
Does the original AI train from the corrections? (Score:2)
This is a good thing (Score:2)
unit testing (Score:2)
Professionals fix amateur work, news at eleven (Score:2)
This has always been the case: Amateurs screw up, professionals have to fix it later. This happens not just with vibe programming (anybody remember "programs" written in MS Access or Excel?) or even just in software development. Amateurs build huts that then collapse or leak, or do electric wiring that then causes outages or worse.
The part that has changed is companies thinking they can get way (again) with amateurs doing the work that would require a professional, because the former are using AI now, and t
The Vibe-Coded Their Own Website (Score:2)
I went here: https://vibecodefixers.com/ [vibecodefixers.com]
And got this: "Internal Server Error"
Any chance they vibe-coded their own website?