After 8 Months with Codex CLI, Here Is My Advanced AI Coding Workflow

Short summary

After 8 months using Codex CLI on real projects, Haijun shares a practical system for stable AI coding workflows: establish clear task boundaries, use Plan mode before implementation, provide visual references for frontend work, leverage Playwright for verification, and enable approval gates on risky changes. Key practices include MCP integration, sandbox modes, and automating proven tasks. The framework shifts thinking from one-off code generation to reliable, controllable engineering patterns.

• •Plan before implementing; use Plan mode to collect context and establish boundaries
• •Provide visual references (screenshots) for frontend tasks with Playwright-based verification across breakpoints
• •Use sandboxes, approval gates, and incremental changes for risky work; automate only stable, proven tasks

Generated with AI, which can make mistakes.