Andrej Karpathy coined "vibe coding" as a throwaway thought. "Fully give in to the vibes, embrace exponentials, and forget that the code even exists." The industry built a movement around it. Now the data is in: experienced developers using AI tools take 19% longer on real tasks, but believe they are 20% faster (METR, 2025). The gap between those two numbers is where your technical debt lives.
Why does AI code ship faster but break sooner?
The numbers tell a different story. AI-assisted code produces 1.7x more major issues than human-written code, including 2.74x more XSS vulnerabilities (CodeRabbit, Dec 2025). AI-coded projects accumulate technical debt roughly 3x faster than traditionally written ones (HackerNoon, 2025). By year two, maintenance costs surge 4x compared to traditional development (BayTech Consulting, 2025).
The output looks fine on demo day. It breaks on day 90.
96% of developers say they do not fully trust AI-generated code. But only 48% actually verify it before committing (Sonar, 2026). That is not a quality problem. That is a verification gap baked into how teams work. And 38% say they skip reviewing AI code entirely because it takes more effort than reviewing human code (IT Pro, 2026).
The pirate and the architect
Dan Shipper nailed the team structure question: every team needs a pirate and an architect. The pirate ships fast by vibe coding. The architect turns that surface area into a reliable, structured system. You need both. But you need to know which one you are hiring.
The problem is obvious. Every tech headline celebrates the pirate. The 10x vibe coder who shipped a product in a weekend. Nobody writes about the engineer who spent three weeks stabilizing it afterward. Hiring pipelines reflect this imbalance. Companies still optimize for speed signals: LeetCode throughput, portfolio volume. Not judgment signals: how someone reviews AI output, why they chose one approach over another.
Ole Lehmann put it bluntly: "The dirty secret of vibe coding: most AI-written code is going live without anyone actually reading it."
What happens when nobody reviews the code?
Forrester predicts 75% of enterprises will face moderate-to-high technical debt severity by 2026, driven by AI-fueled rapid development (CFO Dive / Forrester, 2024). Software engineer vacancies are up 30% in 2026 with 67,000+ openings globally (Metaintro, 2026). Some of that demand is new work. A lot of it is cleanup.
Meanwhile, 42% of all committed code is now AI-assisted, expected to reach 65% by 2027 (Sonar, 2026). The debt is compounding. And the junior developers who would have developed judgment by fixing small bugs? Companies are cutting those roles to "save costs" while generating debt only experienced engineers can fix.
That trap does not fix itself.
Your interview reveals which type you are hiring
Sierra AI eliminated coding interviews entirely. They now test for initiative, ownership, judgment, and system understanding. Zac Gott rebuilt his interview from scratch after realizing they had been hiring for coding ability when they needed judgment.
I think these teams have it right. Vibe coding is not the problem. Vibe coding without verification is. Your hiring process needs to distinguish between the engineer who ships AI code blindly and the one who reads it, tests it, and iterates on it. When vibe coding debt is a line item on your balance sheet, the interview scorecard becomes a financial prediction.
Fairground captures the process, not just the output. When a candidate uses AI in a screener, the scorecard shows whether they verified the result, caught the errors, or just accepted the first response. Start free, 100 credits. No credit card. No sales call.
Related: What Skills to Test in AI-Ready Interviews
Get started with Fairground in just few mins.
Plug and Play. Works well with your existing ATS.
100 Free Credits
