Maintainability of vibe-coded application without b***t

Vibe coding is no longer a meme - it's how 25% of YC W25 startups shipped their codebases, and 84% of developers plan to use AI tools in 2026. But the hangover is real. Recent analyses show vibe-coded code contains almost 2x major issues, 3x security vulnerabilities, and requires almost 2x more time for debugging, than human-written code. The industry is drowning in "AI spaghetti" - code that (hopefully) works but nobody can maintain - while "AI experts" and shiny new AI generators try to sell the story that their setup allows them to vibe code whole operating systems in one day, with a few prompts, without issues.

Using examples from startups and my experience leading vibe-coding teams, I will break down the most common failure modes of vibe-coded applications: duplicated logic, shallow abstractions, fragile architecture, context loss across files, weak tests, accidental security debt, and the silent transfer of review burden to senior engineers. Then, drawing from the research on development methodologies for agentic processes, I will propose a practical, no-BS framework for maintainable vibe coding at both the individual contributor and manager level.

Attendees - individual contributors and engineering leads - will leave with a practical framework they can apply immediately to build faster with AI without turning their codebase into AI slop.