Vibe coding can feel fast—until your agent starts drifting from your stack, bloating context, or “fixing” bugs by rewriting everything. This tutorial breaks down how the speaker uses agentic coding in Cursor or Windsurf to push more work end-to-end onto an AI agent, while staying in control as the codebase grows.

• Walks through a concrete setup: choosing models and adding custom models via an overridden API key/base URL
• Shows why a detailed spec matters, and how they generate one separately before pasting it into the IDE
• Explains how Cursor/Windsurf “rules” guide tech choices, workflow, and scope to avoid unwanted rewrites or duplicate code
• Shares practical habits: managing context size, keeping requests narrow, favoring end-to-end tests, and committing/restoring checkpoints often