{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreievohpfvaonospkp63y5q5ckfsbglhulfp4vvxy46toirr3ncskwu",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mowc54jfy5j2"
  },
  "path": "/t/using-chatgpt-for-planning-vs-codex/1382876?page=2#post_33",
  "publishedAt": "2026-06-23T00:36:24.000Z",
  "site": "https://community.openai.com",
  "textContent": "SLYM3R:\n\n> So, the winning workflow has been:\n>\n> ChatGPT Pro / Extended for product and architecture planning.\n>  Codex Plan Mode when tactical decomposition is useful.\n>  Codex for implementation.\n>  ChatGPT/human review before accepting the output.\n\nI’m working in a very similar way. TBH I don’t think Codex on it’s own could have made the progress that has been made so far - although I am relying on ChatGPT to produce the prompts for Codex, and ChatGPT has been very cautious.\n\nCodex has been useful to identify, group and rank potential issues in the codebase as I progress through the PHP versions, and Codex has been used to do bulk code edits, directed and checked by ChatGPT. Singular/unique issues have been resolved just using ChatGPT, because there’s not much point in handing it off the Codex. ChatGPT is also doing all the git commits, tagging, bundling and project logging. ChatGPT is also doing the smoke tests and I now also have a significant Playwright script that runs over 100 wbesite functionality tests.\n\nI’ve now got the site close to reasonable level of confidence on PHP 8.3 (from PHP 5.2). It has been a lot of tedious work, but I suspect the work could have been automated a lot more - the longest Codex runs have been no more that a few minutes at a time.\n\nHowever, I have no right to complain given that I couldn’t possibly have done this work on my own",
  "title": "Using ChatGPT for Planning vs Codex?"
}