Stefan Mischook
Stefan MischookMar 31
Tech

Vibe Coding Problems are Popping Up

10 min video5 key momentsWatch original
TL;DR

Companies aren't broadly blocking vibe coding, but they're tightening controls on it due to security vulnerabilities and low-quality AI-generated code, while experienced developers using AI tools see 5-10x productivity gains.

Key Insights

1

Non-developers hit the 80% problem where vibe-coded apps stall at completion because they lack proper system-level design and software development principles to finish the remaining 20%.

2

10% secure solutionsOnly 10% of AI-generated solutions are actually secure in tests, with 69% of security professionals finding major issues in AI code, matching early PHP's vulnerability patterns.

3

5-10x productivity gainsExperienced developers using agentic development tools report 5-10x productivity improvements because they understand how to structure applications and direct AI effectively.

4

Companies like JP Morgan are mandating AI usage for developers and tracking it in performance reviews, making AI proficiency a non-negotiable skill.

5

platform-level blockingApple and other platforms are blocking vibe-coded apps with uncontrolled code generation, while open source projects are rejecting AI-generated pull requests due to quality concerns.

6

design principles matterThe key to effective modern development is mastering software design principles, design patterns, and refactoring—not just knowing how to prompt AI for code.

Deep Dive

The 80% Problem: Why Vibe Coding Fails for Non-Developers

Stefan explains that non-developers using vibe coding typically hit a critical wall at approximately 80% completion, where they get stuck and cannot finish their applications. The core issue is that without formal software development training, these developers don't understand system-level design, proper code structure, or how to resolve complex integration issues. As Stefan notes, "if you're halfway there, you're nowhere"—the last 10-20% of development is always the hardest part, and non-developers lack the foundational knowledge to push through. Even worse, apps that do appear to work initially are often unreliable and filled with security vulnerabilities.

Security and Quality Concerns Driving Company Restrictions

Companies are not broadly banning vibe coding but are tightening controls on specific methodologies due to serious security and quality issues. Government security agencies and researchers found that only 10% of AI-generated solutions are secure in testing, meaning 90% of vibe-coded apps are not production-ready. Additionally, 69% of security professionals discovered major vulnerabilities in AI-generated code. This pattern mirrors early PHP, where untrained developers created working but fundamentally insecure applications. The response includes stricter code reviews, more guardrails, and platform-level interventions like Apple blocking app updates from tools like Replit that generate uncontrolled code.

Expert Developers Achieve Massive Productivity with AI Tools

When experienced, well-trained developers use agentic development tools and AI-assisted coding, they report 5-10x productivity improvements. Stefan recounts meeting a former colleague building an AI product who demonstrated this dramatic efficiency gain. The critical difference is that expert developers understand how to structure applications properly, use design patterns, and direct AI effectively to generate quality code. Companies like JP Morgan are mandating AI usage and tracking it in performance reviews, while Meta aims for 75% of code written by AI in some teams. This signals that AI coding proficiency is becoming a fundamental developer skill, equivalent to knowing React for front-end developers.

Open Source and Platform Backlash Against Low-Quality Code

Open source projects are implementing defensive measures against vibe-coded contributions, with some banning AI-generated code outright or autorejecting pull requests because maintainers cannot distinguish between genuine code and AI translation. Teams are increasingly adopting tiered policies: allowing AI for prototyping and internal tools while requiring manual understanding and ownership for core systems and sensitive data. As Stefan observes, this isn't always public policy but is becoming standard practice. The backlash reflects frustration with AI-generated code flooding systems with low-quality contributions that are difficult to maintain and understand.

The Non-Negotiable Role of Software Development Principles

Stefan emphasizes that the key to effective modern development is mastering software design principles, design patterns, and refactoring—not just knowing how to use AI tools. Developers must understand granular code structure, separation of concerns, and system-level architecture to vibe code effectively. For serious systems handling sensitive data, "blindly vibe coding is not going to be the solution." The contrast is stark: trained developers leverage AI as a powerful tool within a solid foundational framework, while untrained developers use AI as a substitute for understanding, resulting in unreliable, insecure applications. This reflects a timeless principle: the best developers write simple, understandable, updatable code.

Takeaways

  • If you want to use vibe coding effectively, learn foundational software development principles, design patterns, and system architecture—AI alone won't compensate for missing knowledge.
  • AI coding is now a mandatory skill for developers; companies like JP Morgan are mandating its use and tracking adoption in performance reviews, making it non-optional.
  • Vibe coding works well for prototyping, throwaway implementations, and internal tools, but should never be used for production systems handling sensitive data without expert review.
  • The 80% problem is real for non-developers: expect to get stuck when completing applications without proper training in how systems actually work together.

Key moments

0:45The 80% Problem Explained

You're vibe coding, you're rocking and rolling, your apps are coming up really quick, and you got about 80% of your app built. I love vibe code. And you can't get that last 20% to work.

3:00Security Reality Check

Research shows that only 10% of AI solutions are actually secure in tests. That means 90% of the vibe coded apps not production ready.

4:40Expert Developer Productivity

He's between 5 to 10 times more productive depending on the type of work he's doing but he also pointed out that he has, he knows what he's doing right, he knows how to structure his applications.

5:20JP Morgan Mandate

Companies like JP Morgan are literally mandating AI usage for developers and tracking it in performance reviews.

7:00The Non-Negotiable Foundation

The key to being a great modern developer today, no question, you have to do AI, but the best developers write very simple, understandable, updatable code unless you're trained in software development principles.

Get AI-powered video digests

Follow your favorite creators and get concise summaries delivered to your dashboard. Save hours every week.

Start for free