998: How to Fix Vibe Coding

In this episode of Syntax, Scott Tolinski and Wes Bos dive deep into the growing problem of 'Vibe Coding'—where AI-generated code lacks consistency, maintainability, and quality due to its reliance on subjective, non-deterministic outputs. They argue that while AI has democratized development, it often produces sloppy, duplicated, and overly complex code that breaks easily and is hard to debug. To combat this, they introduce a suite of deterministic tools that bring structure and logic back into the development workflow. These include code quality analyzers like Fallo (which detects dead code, complexity hotspots, and duplication), CSS-specific tools like Project Wallace and Clint, and agent-enabling tools such as Storybook MCP and Sentry’s Spotlight for real-time debugging. The hosts emphasize the importance of enforcing rules through linters, custom ESLint plugins, and pre-commit hooks to make AI behavior predictable. They also explore headless browser solutions like Agent Browser and Light Panda for reliable testing and automation, and advocate for structured task management with tools like Dex. Ultimately, they stress that the future of AI-assisted development lies not in trusting vibes, but in building guardrails—tools that are fast, enforceable, and integrated into the workflow to ensure code quality, maintainability, and reliability.