What people on Hacker News said about using Codex at work
64% say it worked.
Codex: OpenAI's coding agent.
Based on 452 first-hand reports from 338 people on Hacker News, posted between 6 Jul 2023 and 3 Nov 2026. Updated 3 Nov 2026. How this was made
EACH SQUARE IS 1% OF THE REPORTS
01 / WHICH MODELS AND TOOLS PEOPLE TALK ABOUTTALK, NOT RESULTS
How often each AI model is named in Hacker News comments, month by month. This counts talk, not whether it worked.
02 / HOW TO MAKE IT WORKWHAT THE PEOPLE IT WORKED FOR DID
How to make it work
What the people it worked for did, and what tripped up the people it didn’t, in their own reports. Tap one to read them.
03 / WHAT IT’S GOOD ATBY KIND OF WORK
What it’s good at
How often people said it worked, by the kind of work they used it for.
How much back-and-forth it took
04 / TOOL BY TOOLHOW OFTEN EACH TOOL WORKED
How each tool did
The share of each tool’s reports that said it worked. Open a tool to see its reasons and reports.
Claude Code and Codex are coding agents, and Cursor and Copilot work in your editor. Claude, GPT and Gemini are the models themselves, counted when someone names the model rather than a tool, for example using it through the API or inside Cursor. ChatGPT is OpenAI’s chat app.
Compare two tools
05 / IS IT GETTING BETTER?YEAR BY YEAR, MONTH BY MONTH
Is it getting better?
06 / THE REPORTSEACH ONE LINKS TO HACKER NEWS
All reports
Each report shows what someone wrote, with a sentence either side and a link to the whole comment on Hacker News.
No reports match.
Keys: j and k move between reports, / searches.
07 / HOW THIS WAS MADEWHERE THE REPORTS COME FROM
How far to trust this
These are reports people chose to post on Hacker News, found and sorted by AI models. They show what these people said happened at their work, not what happens to every developer.
How this was made