Vibe coding is quietly building your next legacy crisis
Anthropic’s CFO disclosed that over 90 percent of the company’s code now comes from its AI systems. Tasks that once took hours finish in 30 minutes. The productivity metrics look great. The problem is what those metrics don’t measure.
What the Research Actually Shows
Research on Fortune 50 companies found that AI-assisted developers introduce security flaws at ten times the rate of traditional coders. Forty-five percent of AI-generated code contains OWASP Top 10 vulnerabilities. Independent studies put the rise in technical debt somewhere between 30 and 41 percent after AI tool adoption.
He et al. analyzed 806 open-source repositories that adopted Cursor AI. Their difference-in-differences study, accepted at MSR 2026 and available on arXiv, found a 41 percent jump in code complexity and 30 percent more static analysis warnings. Velocity climbed at first. The debt stayed.
GitClear’s 2025 analysis of 211 million changed lines showed an eight-fold increase in code duplication. Augment Code reported that large language models embed 10 to 30 unstated assumptions per feature, and human reviewers cannot match the generation speed.
The Developer Experience on the Ground
The 2025 Stack Overflow Developer Survey, drawing from more than 49,000 responses, found that 66 percent of developers spend more time fixing AI code that seems almost right. Forty-five percent find debugging AI-generated code more time-consuming than ordinary work.
Kin Lane, an API evangelist with 35 years in technology, told LeadDev:
“I don’t think I have ever seen so much technical debt being created in such a short period of time during my 35-year career in technology.”
Adam Shea, director of AI go-to-market at TEKsystems, described a specific failure mode: AI tools drive duplicated code and phantom dependencies, where the AI invents connections between systems that don’t actually exist. A University of Texas study put hallucinated dependencies at 5.2 percent in commercial models and 21.7 percent in open-source ones.
The Nuance from Google’s DORA Report
Google’s 2024 DORA report added some balance. AI lifted code review speed and documentation quality. It also trimmed delivery stability by 7.2 percent. The Harness State of Software Delivery 2025 found developers reporting more hours lost to debugging and vulnerability fixes than they gained in feature delivery.
OX Security’s report described AI coding agents as behaving like an army of junior developers: they favor monolithic structures, avoid refactoring, and leave 20 to 30 percent of codebases with phantom bugs.
What the Counter-Examples Show
Some high-stakes organizations took a different approach. The US Army built auditability into AI-generated ammunition ordering systems used in disconnected environments. Merck applied similar controls to its clinical supply chain. Both extracted specifications from legacy code, turned them into visual models, then let supervised agents generate reusable components. Development time fell to roughly 25 percent of traditional methods while traceability held.
The pattern across the research is consistent: AI inflates code volume three to four times while quality metrics slide. Teams that treat AI as a spec-driven partner, keeping requirements explicit and humans reviewing assumptions before generation, report fewer surprises. Teams that chase prompt velocity alone are building faster roads to future maintenance pain.
The COBOL parallel is worth taking seriously. Decades of rushed shortcuts became tomorrow’s constraints. Line-by-line translation into modern languages just moves bad process into new infrastructure. According to TechRadar, the same strategic flaw applies to vibe-coded codebases: fresh legacy dressed in contemporary syntax.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
