Introducing the blue-red framework for effective viable coding

Moving from vibe coding to viable coding with AI agents.

Everyone’s vibe coding now, using Cursor, Claude, and GitHub Copilot to generate code faster than ever. But does it work for heavy-duty software? In their presentation at All Things Open, Nnenna Ndukwe, Principal Developer Advocate at Qodo AI, and Itamar Friedman, CEO and Co-founder of Qodo AI, demonstrate why vibe coding went from exciting invention in early 2025 to “rest in peace” by year’s end, and introduce the blue-red framework that makes AI-generated code actually production-ready.

Nnenna live demos vibe coding by refactoring a manual MCP server to use Fast MCP. She prompts an AI agent in her IDE and watches it analyze code, make changes, run tests, and create a pull request in minutes. But Itamar stops her before she ships to production. The generated code dropped essential logging and argument validation. The JSON interface changed, potentially breaking other services. Tests passed, but what about clients consuming this service?

This is the problem with vibe coding for production systems. Code generation is great for prototypes, but real software includes maintainability, compliance, security, and architecture decisions. The blue-red framework treats AI assistance as a team sport with distinct roles. The blue agent generates code in plan mode first, figuring out implementation before touching anything. Then it switches to code mode. The red agent reviews everything, checking requirements, identifying missing validation, flagging breaking changes, and providing specific suggestions.

Better planning makes review easier. When the blue agent has clear requirements and creates a detailed plan, the red agent verifies implementation matches intent. This catches the common problem where AI agents miss early instructions after context window summarization. Running review workflows in the CLI before creating pull requests lets developers catch issues before anyone else sees the code.

Read more: 5 forces driving DevOps and AI in 2026

Key takeaways

• Vibe coding works for prototypes but fails for production systems requiring maintainability, security, and reliability.
• The blue-red framework separates generation (blue agent: plan then code) from review (red agent: requirements and quality checks).
• Better planning enables better review, catching dropped validation, breaking changes, and missed requirements before production.

More from We Love Open Source

• Want to get into AI? Start with this.
• 5 forces driving DevOps and AI in 2026
• Deep dive into the Model Context Protocol
• The secret skill every developer needs to succeed with AI today
• Open source won, so why are we still fighting?