# ChatGPT Dots vs Codex vs ChatGPT Work: which to open, how a Dot hands work to Codex, and what 93 real cases show

Codex writes and tests code, ChatGPT Work runs a delegated task, and your Dot owns an ongoing job and hands coding to Codex. Where each task runs, what counts toward Codex usage, and fixes from 93 real cases.

When to open Codex, ChatGPT Work or your Dot, how a Dot starts and follows Codex tasks (local, cloud, existing threads), why Codex usage can stay flat or hit zero, and seven handoff failures with what got them moving.

- **Open Codex directly** for a single coding change you're going to watch. It is the shortest path, and it skips a layer that can fail. Kamimoto's Dot went 20 minutes without noticing that Codex had failed at a node install on his Windows PC; typed straight into Codex, the same instruction worked.
- **Use ChatGPT Work** for a one-off or scheduled task that isn't mainly code: a research summary, an analysis, a document. A Dot can also start these for you.
- **Give it to your Dot** when the work keeps going after the first answer: a bug queue, CI that has to stay green, several repos, or several Codex threads you want to follow from your phone. Kurt Bell went to bed with a bug and a data issue and woke up to tested fixes and a morning report; Eduardo (@martiano) went out for dinner while his Dot merged 16 PRs and waited for CI; 余佳恆 (@tachichocolate) gives the Dot tickets, task splitting, acceptance checks and merges, and leaves Codex the implementation and review.
- **Ask your Dot for status, not Codex** when you're away from your desk. Kevin Ly asked his Dot "Dex" from his phone where a project stood, and it checked his connected Mac and listed what was done and what was left. Kurt Bell asked on a walk and heard that 1,148 tests had passed.

Source: https://gptdot.net/dots-vs-codex/
