5,000-Line Untested PR: How Vibe Coding Is Tearing Teams Apart
A 5,000-line untested AI-generated PR exposes the engineering discipline crisis in the Vibe Coding era.
A Hacker News post captured a growing tension: a developer's business partner submitted a 5,000-line Pull Request generated entirely by AI and never tested. The post explores how "Vibe Coding" — using large models to rapidly generate large amounts of code — creates serious collaboration problems when abused. The core issue is that AI drives code generation costs toward zero while review and verification costs multiply, creating a severe responsibility imbalance. The article argues AI isn't the culprit; the problem is skipping basic engineering gates and offloading all verification to collaborators, eroding both efficiency and trust. Four remedies are proposed: own what you ship, control PR size, treat tests as non-negotiable, and align as a team on AI usage norms.
A Real Team Dilemma
A short post recently appeared on Hacker News that struck a nerve across the developer community: a developer complained that their business partner had submitted a Pull Request with over 5,000 lines of code changes — entirely "vibe-coded" with AI and, crucially, never tested by the person who submitted it.
The post itself was brief, but it cut straight to one of the most pressing pain points of the AI-assisted coding era: as code generation becomes nearly free, the cost of code review and quality assurance multiplies. This isn't just a technical problem — it's a problem of teamwork and trust.
What Is Vibe Coding?
"Vibe Coding" is a concept that has gained traction in the developer community, coined and popularized by AI researcher Andrej Karpathy. It describes a new style of programming: instead of writing code line by line, developers describe their intent to an AI in natural language, letting large models (via tools like Claude, GPT, Cursor, etc.) generate large blocks of code. The developer essentially "goes with the vibe" — accepting and tweaking the AI's output rather than authoring it directly.
The appeal is obvious — production speed is extreme, with hundreds or even thousands of seemingly functional lines appearing in minutes. But vibe coding's risks are just as pronounced:
- The person generating the code often lacks a deep understanding of the details
- Code may "look right" while hiding subtle bugs
- Large-scale changes are difficult to review manually
- Code enters the collaboration pipeline without test validation
The "5,000-line untested PR" in this story is a textbook example of what happens when vibe coding is abused.
Andrej Karpathy formally introduced the term "Vibe Coding" on social media in early 2025, describing how he used this approach to build a complete small application in just a few hours. As a co-founder of OpenAI and former AI director at Tesla, his endorsement helped the term spread rapidly through the developer community. The maturation of tools like Cursor, GitHub Copilot, and Claude Artifacts further lowered the barrier — developers can now describe requirements in natural language and get hundreds of lines of complete code in one shot, rather than just inline completions. This is fundamentally different from earlier "AI-assisted coding": in that model, the developer led and the AI filled in the gaps. In vibe coding, the AI drives the code structure while the developer plays the role of "intent expresser" and "output acceptor," with far less line-by-line comprehension involved.
The Core Problem Isn't AI — It's Process
It's worth emphasizing: AI-generated code is not inherently the problem. The real issue is that the person involved skipped the most fundamental quality gates in software engineering — testing and self-review.
The Shift and Imbalance of Responsibility
In traditional collaboration, the PR author bears the responsibility of making it easy for reviewers to approve. A well-prepared Pull Request should be tested, runnable, and scoped to a manageable set of changes. When someone simply pushes raw AI output to their partner without any review, they are effectively offloading the entire cost of understanding and validation onto the reviewer.
Reviewing 5,000 lines of code is an enormous undertaking. The reviewer faces an impossible choice: spend hours reading and testing every line (extremely costly), or rubber-stamp it and accept the risk. This asymmetry — where code generation approaches zero cost while code review remains expensive — is quietly eroding the collaborative efficiency of many teams.
In economic terms, this cost asymmetry is called an "externality": the party generating the action (the PR author) doesn't bear the full cost, which is instead passed on to a third party (the reviewer). In software engineering practice, large engineering organizations like Google and Meta have long imposed code review standards — such as the rule of thumb that a single PR should not exceed 400 lines — precisely to cap review costs and maintain quality thresholds. When AI increases code generation speed by 10x, without corresponding constraints, the reviewer's burden can scale up linearly as well, making them the new bottleneck in team collaboration.
The Fracture of Trust
The combination of "business partner" and "didn't even test" in the original post reveals more than a technical complaint — it signals disappointment in the working relationship itself. When one party throws unverified work at another, it's fundamentally a drain on collaborative trust. In small teams or startups, this kind of fracture can be especially damaging.
Maintaining Engineering Discipline in the Age of AI
In the face of the productivity temptations that vibe coding offers, teams need to establish new collaborative norms to manage the risk:
1. You Ship It, You Own It
Whether code is written by a human or generated by AI, the PR author is responsible for its correctness. At minimum, that means: it runs locally, the core paths have been tested, and you can explain every significant change. "The AI wrote it and I'm not sure how it works" is not an acceptable excuse.
2. Control PR Granularity
A single 5,000-line PR is never healthy, full stop. Breaking large changes into multiple small, focused Pull Requests not only makes code review easier — it also reduces the risk of introducing bugs. This is basic software engineering common sense, and it shouldn't be abandoned just because AI speeds up the writing.
3. Tests Are Non-Negotiable
AI can help you write code, and it can help you write tests. Having AI generate corresponding unit tests and actually running them is the minimum standard for validating vibe-coded output. Five thousand lines of code without tests is, at its core, 5,000 lines of unknown risk.
4. Establish a Team Consensus on AI Usage
Teams should reach a clear agreement on how AI-assisted programming is used: which scenarios allow AI to generate freely, which require human oversight, and what checks must be completed before submitting. Setting norms upfront is the only way to prevent this kind of collaboration conflict.
Productivity Tools Require Engineering Constraints
The reason this brief Hacker News post resonated so widely is that it reflects a growing pain the entire industry is experiencing. AI has made "writing code" easier than ever before, but the demands of "delivering software responsibly" have not gotten lower — if anything, they've gotten higher.
Vibe coding is a powerful productivity tool, but it cannot become an excuse to dodge engineering accountability. The most effective teams aren't the ones using AI to mindlessly pile up lines of code — they're the ones embracing AI-powered speed while holding fast to the time-tested engineering principles of testing, code review, and scope control. The tools may change, but the commitment to quality should not.
Related articles
Insufficient Source Material to Generate a Valid Article
The provided source material is a single unrelated tweet with no AI or tech relevance — insufficient to support a complete, valid technical article.
Insufficient Source Material to Generate a Valid AI/Tech Article
This source material is a tweet about the ages of Underworld members — unrelated to AI or tech, and insufficient to support a full article.
Insufficient Material: Unable to Generate a Valid AI/Tech Article
The provided material is a condolence tweet about a San Diego mosque attack — unrelated to AI/tech and too limited to generate a valid technical article.