A practical guide to AI-assisted development that distinguishes between pure vibe coding (accepting AI output blindly) and disciplined AI-assisted development. Covers five core practices: writing specs before prompting, context engineering over prompt engineering, the Plan→Execute→Verify loop, test-first development, and security review. Also walks through a hands-on demo using Mistral Vibe, an open-source CLI coding agent, showing codebase understanding, feature planning, subagent-based verification, and PR creation via custom skills. Includes anti-patterns like endless error loops, comprehension gaps, and session drift.
Table of contents
Turn trace reviews into production eval metricsHow to vibe code: A developer’s playbookSort: