A new engineering service wants to charge you $10,000 to make your codebase smaller. The pitch, posted to Hacker News by a three-person team at odra.dev, targets a problem that’s spreading fast: AI-generated projects that work fine at first, then rot into something nobody can safely touch. According to Hacker News, the offer landed with a score of 185 and a headline that says it plainly: “We charge $10k a week to delete AI-generated code.”
The team behind it is Maciej, Kuba, and Krzysztof, three senior engineers who describe themselves as “not agents.” Their whole model is the reverse of what most AI coding tools sell. Instead of generating more code, they remove it.
The problem they’re selling against
The post describes a failure mode anyone who’s leaned on AI coding will recognize. “Adding a feature now takes days and breaks two other things,” they write. “This happens to every vibecoded project past a certain size: the agent stops seeing the whole picture and starts duplicating code instead of finding it.” Fourteen date formatters where you need one. A hand-rolled framework that should’ve been a library. Duplicated logic scattered everywhere.
How the service works
The process, as detailed in the Hacker News post, breaks into clear steps:
- Free analysis first. They look at your codebase for nothing. If they don’t think they can help, they say so and you walk away.
- A committed reduction target. You get a fixed price and a concrete goal, something like “100,000 lines down to 35,000, same functionality.”
- A safety checklist before cutting. They sit with you and write out what the app does, screen by screen, endpoint by endpoint. That document protects both sides.
- One week of focused work. Three senior engineers collapse duplication, swap custom code for libraries, and rebuild the unsalvageable parts clean.
- Guardrails you keep. You walk away with the smaller codebase, the QA checklist, and a set of controls: a CLAUDE.md file, lint rules, and CI checks meant to “slow the slop down” next time you build.
The pricing is the interesting part
Here’s what stands out. You pay in proportion to results. The target is set up front, and if they miss it, you pay less. Their own example: “Promised 50% and delivered only 20%? That’s 40% of the target, so you pay $4,000.” Hit or beat the target and you pay the full $10,000.
Lines are counted with scc, a standard code-counting tool, and only non-blank, non-comment lines count. The contract also bans code golf, so they can’t game the number by stripping comments or compressing your code into something clever and unreadable. There’s a two-week warranty on top: if they broke something that worked before, they fix it free.
Yes, they use AI too
The honest part of the pitch is that these engineers use Claude Code themselves, just “on a very short leash.” Their claimed edge is judgment: “thirty years of combined experience about what maintainable code looks like, and the agent doesn’t get a vote.”
Why this matters
This is a signal worth watching. A cottage industry is forming around cleaning up after AI code generation, which tells you something about where the first wave of vibecoding actually landed. The tools got very good at producing code and much worse at maintaining it. When people will pay $10,000 a week to delete lines, the cost of that gap is no longer theoretical.
There are caveats. This is a small team, not a scaled platform, and the whole thing runs on a personal conversation rather than a signup form. “We’re real people and we’d rather talk than exchange forms,” they write, pointing anyone interested to email and bring their repo.
Expect more services like this. As AI writes a larger share of the world’s code, the market for people who can safely take code away may grow just as fast as the market for people who add it. Full details are available at the original Hacker News post.