Vibe Coding Problems: Insufficient Testing and Worktree Management
Vibe coding practitioners report that idea generation outpaces verification capability, with AI producing features faster than developers can test them. Issues include undetected worktree branches where AI Agent changes were made but never merged into main. Users have created automated rules to check for orphaned worktrees and enforce cleanup procedures for incomplete AI coding sessions.
Key insights
Testing Bottleneck in AI-Assisted Development
Users report that with vibe coding, ideas are no longer the bottleneck—verification is. AI tools generate features rapidly, leading to accumulating untested functionality. Unlike manual coding where development pace allowed thorough testing, AI-assisted coding produces features too quickly to test properly. Even Codex and Claude Code products contain frequent small bugs due to insufficient testing during development. Computer Use-based automated testing has also proven inadequate for comprehensive coverage.
Worktree Management Problem with AI Agents
A user discovered that AI agents sometimes complete tasks in separate git worktrees without merging changes back to main. Upon inspection, multiple orphaned worktrees were found containing completed but unmerged work. The user resolved this by creating agents.md rules requiring worktree cleanup upon task completion, including explicit merge-or-archive procedures with tagging for archived branches.
Community Workaround for Agent Task Tracking
Users developed reference prompts to detect orphaned worktrees, listing branches with modification summaries and session contexts. Clean-up prompts were created to merge valuable work and delete worktrees while providing clear recommendations. The agents.md rule mandates that worktree tasks must delete the worktree and branch after completion via merge or archive tagging, with subagent sessions responsible for cleanup of their spawned worktrees.