# Codex workspace basics: plan a task and inspect its output

Use the overview to find the working surfaces you need for a concrete job, with the supplied recording’s chapter limits retained.

[HTML](https://lukejbyrne.com/library/guides/o11e0pmbhrg/)
Content language: en
Author: Luke Byrne
Video published: 2026-04-29
Guide published: 2026-11-02

[Watch the video](https://www.youtube.com/watch?v=O11E0Pmbhrg)
Video language: English

## Put the task in a project context

The April description presents Luke’s move to GPT-5.5 in Codex and his personal comparison with Opus 4.7. The “95%” in the title is the overview’s framing, not a measured completeness claim. Begin with the job you want to do and the files or evidence needed to do it.

The projects-and-chats section begins at 2:10. A project gives the work a location; a chat carries the particular request and its follow-up. Keep the intended output specific enough to inspect after the agent works.

1. Choose the project and start a chat for one task.
2. Use plan mode to review the proposed approach.
3. Resolve missing requirements before continuing with implementation.

## Choose plugins for the actual job

At 4:03 the chapters turn to browser, GitHub, Gmail and Sheets plugins. The description also covers image generation and the side panel’s artifacts, sandbox previews and terminal access. Use these surfaces to reach evidence and inspect work, while checking the intended account or repository.

The original description links PixPretty: https://bit.ly/4txAnYR for visual workflows and says Scripti: https://scripti.ai was used to write and package the video. Those links preserve the source context, not a requirement to use those products for every Codex task.

1. Connect the service needed for the selected task.
2. Inspect the artifact, preview or terminal output that supports the result.
3. Compare the finished work with the plan and the original request.

## Keep parallel work understandable

The retained chapter at 6:49 covers awaiting-response status and parallel chats. Track which conversation owns which task and which one needs a reply before switching attention. A waiting status is a cue to inspect that conversation, not evidence that the entire project has finished.

Luke’s preference for plan mode and extra-high reasoning belongs to this recorded workflow. The supplied cut ends at 8:33, before the description’s 9:02 final-verdict marker. Use the available sections as orientation and assess your own result through the actual work returned.

## The takeaway

Find the project, review the plan, connect the needed tool and inspect the resulting work before moving on.

## Workflow resources

- [bit.ly](https://bit.ly/4txAnYR)

## Chapters

- [00:00 GPT-5.5 vs Opus 4.7 — why I switched](https://www.youtube.com/watch?v=O11E0Pmbhrg&t=0s)
- [02:10 Projects, chats, and plan mode](https://www.youtube.com/watch?v=O11E0Pmbhrg&t=130s)
- [04:03 Plugins: browser, GitHub, Gmail, Sheets](https://www.youtube.com/watch?v=O11E0Pmbhrg&t=243s)
- [06:49 Awaiting response + parallel chats](https://www.youtube.com/watch?v=O11E0Pmbhrg&t=409s)

## Sources and limitations

Written companion prepared 2 November 2026 from this exact published video description and its chapter sequence. The full recording was not independently watched or transcribed for this article, and the products and reported results were not retested. Checklists are editorial adaptations of the documented workflow. Model availability, interfaces, prices and offers may have changed since publication; use the retained original resources for current requirements. The description lists a final-verdict marker at 9:02, beyond the supplied 8:33 duration. It is omitted without retiming the four retained chapters.

[Original video](https://www.youtube.com/watch?v=O11E0Pmbhrg)