“A full software engineering teammate”: OpenAI product lead on getting the most out of Codex | Alexander Embiricos

What It Covers

Alexander Embiricos, OpenAI product lead for Codex, demonstrates practical workflows for using the coding agent from basic setup through advanced techniques like parallel work trees, automated planning, and GitHub code review integration.

Key Questions Answered

• •Work Tree Parallelization: Use git work trees to run multiple Codex instances simultaneously on separate branches without conflicts, enabling parallel exploration of different implementation approaches while maintaining clean separation of concerns and independent code review paths.
• •Planning for Complex Tasks: Copy OpenAI's planning specification into a markdown file and reference it when prompting Codex for major features. This structured approach produces thorough 120-line plans with milestones and implementation details, particularly effective for 30-60 minute tasks requiring architectural thinking.
• •GitHub Code Review Automation: Enable Codex automated code review in repositories to catch bugs proactively without human prompting. The system only flags high-confidence issues to protect developer attention, and engineers can reply directly asking Codex to fix identified problems within the same thread.
• •Context-Rich Prompting: Always include the why behind requests, not just the what. Describe the desired outcome and constraints rather than prescribing exact solutions, allowing Codex to leverage its understanding of the codebase to determine optimal implementation approaches that humans might miss.