'Vibe Coding Cleanup Specialist' Is an Actual Job Title Now. One Shop Charges $10,000 a Week to Delete Code

A Polish software house put its rate card on the internet with a title that reads like a joke: "We charge $10k a week to delete AI-generated code." Hacker News gave it 304 points and 240 comments. Some of those comments are about whether the business is real. The more useful ones are about what the work costs, who buys it, and where the price breaks.

The offer is specific enough to argue with: one week, three senior engineers, $10,000, and payment in proportion to how much of the codebase they remove. Two weeks of warranty. The company is called Slopfix, and the founder, posting as zie1ony, showed up in the thread to defend it.

What is actually being sold

zie1ony described the job in the first comment: "A client calls, asks for a 'small refactor,' and sends you 100k lines of AI-generated spaghetti. And this is great! This is something we can work with."

Then the part that explains the pricing: "Removing the first 30% is easy. The next 30% is harder, and that is exactly what the price should be on: doing what others can't. We use coding agents too, of course, but as a tool, not as the driving force."

So the deliverable is a smaller codebase that still does what the old one did. The commitment is a reduction target agreed before the week starts, and the invoice scales with how much of that target they hit. That is unusual. Most consulting sells days. This sells a number the client can verify by running a line count.

zie1ony was blunt about why it isn't priced the way he'd prefer: "I'd charge for amount of tokens, but it would be harder to sell."

Who the client is

This is the detail that makes the business legible, and it came out under a question about whether clients would rather learn to build properly themselves:

"The typical client is a 40 years old ex-CTO, that had idea, verified it with Claude Code and got to the point, where the repo is too big. Maybe even rewrote it all from zero, once or twice, but end-up with the same outcome after adding more features."

Not a developer. Someone with enough technical background to get a product working and not enough patience to care how the server is implemented. zie1ony added a second trigger that turns interest into a purchase order: "Sometimes the requirement is a security review. And you pay much less if you downsize the repo" before handing it over.

That is a real buying moment. Nobody wakes up wanting a refactor. They want to pass an audit, close a round, or hand the repo to a developer who will otherwise quote three months.

Where the supply of work comes from

Two comments in the thread describe the upstream side of this market better than any market-size estimate.

The first is from thraway3837, who replaced a $120,000-per-year low-code platform with vibe-coded workflows: "Total cost of AI subscription per month: Less than $1000." On how the code gets reviewed: "We have senior engineers review the actual functionality and none of them have read any more than a few lines of code." Pushed on it later, they doubled down: "we don't even read the code anymore. In another comment I wrote: we just CMD+Q VS Code and it's not even in the recents/pin to dock, since what's the point?"

That is one company converting $120k of vendor spend into under $12k of subscriptions, plus a codebase nobody has read. It may work fine. It may also be exactly the repo that shows up in someone's inbox in eighteen months attached to the words "small refactor."

The second is what happens when the volume is public. A separate Hacker News thread covered a developer's audit of a site built by a YC executive who says he ships 37,000 lines of AI code a day. The findings, as quoted in the Fast Company writeup: 169 server requests totalling 6.42 megabytes on a single page load, 28 test files shipped to every visitor's browser, 78 JavaScript controllers loaded for features that aren't on the homepage, the logo downloaded in eight formats "including a completely empty 0-byte file that somehow made it to production."

arcticbull, in that thread, made the point that connects the two stories: "I am opposed to the idea that we've decided to go back to measuring work in terms of lines of code. It has always been the worst metric on earth as a proxy for productivity. Every line is a liability, and it always was."

If lines are liabilities and something is generating them at unprecedented speed, then a service priced on removing them is at least aimed at the right thing.

What the work pays the people doing it

bdcravens asked the question everyone doing freelance maths wants answered: "What your markup on their salaries? For the level of work you're promising, it sounds like they may be at market or below."

kristianc supplied the missing context: "They're all based in Poland, so it's probably fairly generous compared to European market." zie1ony's reply to that was one word: "true."

Lalabadie worked it out: a well-booked version of this lands around $105,000 a year per developer. That is a good living in most of Europe and an unremarkable one in San Francisco, which is the whole arbitrage.

The constraint isn't demand, according to the founder. It's the nature of the work: "knocking one refactor after another is super intense and we would burn out quickly. To me refactoring/reading/reviewing a code that you have never seen is one of the hardest thing to do for a software developer. If I have 4 such projects a month, I'd just hire more people and allow them to rest."

Four projects a month is the stated ceiling for a three-person team, and he described it as the point where he'd hire rather than push. yosefk's read on the price was shorter: "Too cheap, charge 3x more at least."

The pricing gets attacked from both directions within one thread

Under the same post, dreambuffer offered to do the work for $200 a week and left an email address. scritty-dev joked about charging $15k a week to fix whatever the $10k cleanup broke. hahahaa raised it to "$100k to tell them they never needed the code and delete it all."

The jokes are doing something real. A service defined loosely enough to describe in one sentence is a service anyone can claim to offer at any price, and the first thing that happens after you publish a number is that someone undercuts it by 98%. What protects the $10,000 is not the description of the work. It's the reduction commitment, the named engineers, and the fact that the client can measure the result.

Anyone selling a service off the back of a public post should read that $200 reply as the standard outcome, not as a heckler. We looked at the same dynamic from the buyer's side in who gets paid for your purchase: the price a client sees is rarely the price the work costs.

The objections worth taking seriously

Three of them survive scrutiny.

guluarte: "I don't think one week is enough to learn the complex business rules that some software needs to follow." dgacmu answered it in a way that also explains the target market: nobody is hiring a one-week engagement for a decade-old system with load-bearing corner cases. The vibe-coded tangles are newer, which is what makes the week plausible.

quuxplusone went after the warranty: "'Two weeks of warranty' jumped out at me. That's like 'you have two weeks to find the thing we broke, or else we aren't responsible for it.' In my experience, a good bug can hide for months more than two weeks!" zie1ony's answer was honest and a little unsettling: "these days bugs are not so scary anymore, so a client can vibecode a fix faster then getting me to fix it." Carrok pushed back hard on that, pointing at data loss and outages. Neither of them is wrong about their own category of software, which is most of the disagreement in this thread compressed into three replies.

And fwlr named the structural doubt: "my gut reaction to 'take an AI-inflated codebase and apply AI deflation to it' is something like 'that's akin to applying two rounds of lossy transcoding; the errors don't cancel out, they cross-multiply'." zie1ony did not dispute the premise, only the framing: "Our outcome is also AI-generated (or AI-assisted) code. It is not possible to rewrite 100k loc in a week."

The thread also had a running joke about the sales page itself reading like AI output. capiki, nathansoz and zeroonetwothree all flagged it. zie1ony took it: "I'm really bad at writing good long english texts, so I used Fable to help me. But the essence is mine." Which is either the funniest detail in the story or the least surprising one.

What the thread does not contain

No client. Not one comment from someone who paid the $10,000 and described the result. No revenue figure, no number of engagements completed, no before-and-after on a real repo beyond the generic case in the sales copy.

llm_nerd said it flatly: "I would be extremely surprised if they find a single paying customer." That claim is unsupported too, and llm_nerd was arguing partly with the anti-AI mood in the thread rather than with the business. But the absence cuts both ways, and it's the thing to hold onto. What we have is a published price, a founder answering questions in public for hours, and a large number of engineers agreeing that the underlying problem exists. What we don't have is proof that anyone bought.

We noted the same gap in the make $2k a month thread: the answers describing an income stream and the answers proving one are different sets, and the second set is always smaller.

What transfers if you sell services

Four things in this story are portable, and none of them are specific to code.

Price on the result the buyer can measure. A line count is crude, and zie1ony admitted tokens would be the better unit, but the client can verify a line count without trusting you. That is worth more than accuracy.

Pick the buyer who has a deadline. The ex-CTO with a security review coming has a date. The one who merely dislikes his codebase has a feeling. Only one of them signs.

Publish the number. The $200 undercut and the "too cheap" comment arrived in the same thread, which means the price is roughly where a public price should sit: low enough to attract a knife fight, high enough that someone tells you to raise it.

Say what happens when it goes wrong before you need to. The warranty question in that thread had no good answer, and it was the only question in 240 comments that could actually cost the business money.

The prompts in the panel handle those four in order. slop-scope-estimate turns "this repo is a disaster" into a week-shaped scope with a number and a list of what you excluded. outcome-pricing builds the payment structure with the arithmetic at a bad outcome, not just a good one. cleanup-client-finder goes after the buyer with the deadline instead of the developer who can do it himself. And cleanup-warranty-terms writes the clause that decides whether your fixed-price week stays a week, including the mechanism for telling your regression from the one their agent introduced on Friday.

The market for cleaning up after code generators is not a prediction. Someone published a price for it. Whether $10,000 is the right number is the part still being tested in public, and the person who set it is the first to say he might have gone too low.