Vibe Coding Is Here—and It’s Out of Control: Amateur AI Coding Is Wrecking Real Software, And Veteran Programmers Are Cashing In Fixing It
In a future where software is built from vibes, AI models generate code based on mood and style rather than exact specs. The results can run, but they feel off—polish, UX, and reliability lag behind the pace of generation. Even the tech giants are embracing AI coding, claiming sizable portions of code are AI-generated: Sundar Pichai said Google has up to about 25% AI-generated code; Satya Nadella put Microsoft at roughly 30%. The risk is palpable. One vibe-coding misstep wiped out a small business’s database. Meanwhile, a salvage economy has emerged: on Fiverr there are more than 230 gigs dedicated to fixing vibe-coded projects. There is even a dedicated site, VibeCodeFixers.com, with more than 300 veteran programmers ready to salvage projects. The term itself entered the scene after Andrej Karpathy coined it in February.
In This Article:
What Is Vibe Coding? A Quick Definition
Vibe coding is the practice of letting an AI model write code guided by vibes—intent, style, and user experience—rather than precise requirements. The idea is speed: prototype ideas quickly and iterate. But the outputs are often functional yet lacking polish or a cohesive vision. Interfaces may look off, and logic may drift from the intended path. The risk is real: AI models hallucinate, misinterpret tasks, or produce code that performs oddly or breaks as the app grows. Even large teams often find that the speed gains are offset by hours of debugging and verification.
Meet the Fixers: Who They Are and How They Profit
Hamid Siddiqi, a programmer, now offers to “fix your vibe code” on Fiverr. He explains, “I started fixing vibe-coded projects because I noticed a growing number of developers and small teams struggling to refine AI-generated code that was functional but lacked the polish or 'vibe' needed to align with their vision.” Siddiqi says clients come for help with everything from horrendously optimized code to botched AI-generated UIs. He’s been offering vibe coding fixer services for about two years, starting in late 2023, and he currently works with around 15-20 clients regularly, with additional one-off projects throughout the year. The fixation isn’t limited to freelancers. Ulam Labs, a software firm cited by 404, says on its site, “we clean up after vibe coding. Literally.” An entire ecosystem has formed around the niche, including VibeCodeFixers.com. Its founder, Swatantra Sohni, told 404 that over 300 veteran programmers have signed up. He notes that most vibe coders are product managers, salespeople, or small business owners who believe they can “build something.”
The Industry’s Paradox: AI Is a Tool and a Trap
AI coding can speed development, but it introduces new failure modes that require constant oversight. A recent study found programmers using Claude were 19% slower and used less than half of the AI’s suggestions, highlighting a counterintuitive inefficiency. Meanwhile, big players continue to lean in. Google’s and Microsoft’s claims that a sizable share of code is AI-generated underscore the stakes—and the pressure to salvage and polish AI-created work. This is where the fixers fit in: they bridge the gap between ambition and reliability, turning rough AI outputs into usable software. Yet the tension between promise and reality persists, as the industry debates the true value and cost of AI-assisted coding.
Costs, Caution, and a Path Forward
There are real costs to vibe coding. AI usage fees can add up, and in some cases it may be cheaper to start from scratch rather than patch a botched piece of software. Sohni notes that vibe coders burn through AI fees in the final stages when adding features that break the app. Despite the risks, the market persists. Vibe coders are still in demand, and some firms—like Ulam Labs—continue to fix and clean up after vibe coding. Andrej Karpathy’s February coinage of the term highlighted a growing awareness that the phenomenon isn’t going away. Today, more than 300 veterans on VibeCodeFixers.com and 230+ Fiverr gigs attest to a broader, ongoing shift toward human-led salvage of AI-produced software. Ultimately, AI will be part of software creation, but human oversight remains essential. The vibe coding trend shows that speed without reliability is a dangerous mix, and skilled engineers are increasingly needed to guide AI toward real, usable products.
