Reddit dev's ChatGPT + Codex workflow uses RRR method for efficiency

A Reddit user under the handle jaylanky7 has posted a workflow that combines two of the most popular AI tools—ChatGPT and GitHub Copilot (referred to as Codex)—to streamline software development and creative writing. The process begins with ChatGPT handling high-level worldbuilding and conceptualization, after which the user instructs ChatGPT to generate detailed, actionable prompts for Codex. To maintain quality and consistency, the user imposes strict instructions on ChatGPT: it must perform RRR (remember, research, review) before every prompt. This ensures the AI recalls previous context, researches relevant information, and reviews its output before finalizing. Additionally, the user sets explicit guardrails to prevent the model from straying off-topic or producing vague suggestions.

Once the prompts are handed to Codex, the AI coding assistant writes and tests code, generating reports that are fed back to ChatGPT for further refinement. The entire feedback loop reduces manual oversight and speeds up iteration. The user reports significant time savings, fewer headaches, and a dramatic reduction in typing—essentially letting the AI handle the bulk of repetitive work while the developer focuses on strategy. This approach exemplifies a practical multi-agent system where different AIs specialize in complementary tasks, with human-imposed structure ensuring reliability. It's a lightweight but effective alternative to more complex orchestration frameworks.