The power user’s guide to Codex: parallelizing workflows, planning techniques, advanced context engineering tips, automating code reviews, and more | Alexander Embiricos
How I AI
2 DAYS AGO
The power user’s guide to Codex: parallelizing workflows, planning techniques, advanced context engineering tips, automating code reviews, and more | Alexander Embiricos
The power user’s guide to Codex: parallelizing workflows, planning techniques, advanced context engineering tips, automating code reviews, and more | Alexander Embiricos

How I AI
2 DAYS AGO
In this episode, Alexander Embiricos, product lead for Codex at OpenAI, dives into the practical applications of AI-powered coding, demonstrating how both novice and experienced developers can harness Codex to streamline software development. From real-time code edits to large-scale application builds, the conversation explores the workflows that are redefining engineering efficiency.
Embiricos showcases Codex’s integration with VS Code and terminal environments, enabling natural language-driven code changes like adjusting game mechanics or creating new features. To manage parallel tasks without conflicts, he highlights the use of Git worktrees—allowing isolated development across multiple branches. A key workflow involves creating detailed implementation plans using PLANS.md, which Codex processes to generate structured development steps, as demonstrated in building the Sora Android app in just 28 days. He distinguishes between rapid 'vibe coding' for prototypes and deliberate planning for production systems. With GPT-5.2, Codex delivers faster, more accurate code by improving reasoning speed and problem-solving. Automated code reviews powered by Codex are now embedded across OpenAI’s development cycle via GitHub, Slack, and Linear, boosting pull request volume significantly. However, challenges remain in balancing automation with human oversight, particularly around latency and user trust. The focus is on designing intuitive, transparent interfaces that enhance—not replace—developer agency.
04:19
04:19
Codex processes parallel natural language commands to edit code instantly
07:14
07:14
Use Git worktree to isolate Codex changes and prevent conflicts
07:34
07:34
Git worktrees allow one Git instance to track multiple copies of the codebase
11:10
11:10
Codex can translate input placeholder strings in isolated Git worktrees.
14:54
14:54
Effective planning is critical when using Codex for complex engineering tasks.
15:37
15:37
Codex is thorough and good at handling complex implementation tasks.
20:57
20:57
Codex enables massive acceleration on both learning and execution in engineering workflows.
24:35
24:35
GPT-5.2 fixed a bug that was previously unfixable by earlier models
28:08
28:08
Codex provides better critiques than human engineers in code review
32:02
32:02
Users are reluctant to type prompts, creating a barrier to effective AI interaction.
34:35
34:35
Developers are limited more by thinking speed than typing when using AI tools.
44:31
44:31
Being polite to AI helps preserve your own humanity and prevents rudeness toward others.