Independent visual research / Public editionReviewed 04 Sep 2026
Motion fieldnotesOpenAI film study
About this study

Presenter demo / 02:04 / Published 2026-06-18

Record & Replay in Codex

A recorded publishing workflow is followed by a generated skill and a second upload that reuses it. The repeated task makes before-and-after comparison concrete.

Watch official film ↗18 screen examplesAll 100 films
Representative frame from Record & Replay in Codex
OpenAI · representative sample 00:50.000 · Source moment ↗

Useful to borrow

Establish a manual workflow, expose the reusable artifact, then replay the same kind of task with new inputs. Keep the external application's completion state visible.

Do not infer

The film compresses a multi-step upload; its duration is not a product latency measurement. Do not suggest your product has record-and-replay without verifying it.

Look closer

22 shot groups.

Read across from the observed content to its treatment and possible adaptation. Enlarge the frame for detail.

Approximate study intervals; horizontal scrolling reveals all columns on small screens.

Moment / sourceReference frameWhat is on screenFraming / editUseful adaptation
Shot group 01
≈ 00:00–00:19
People / environmentWatch from here ↗
Sample 00:10.000
Presenter introduction

A desk presenter introduces the workflow before any screen capture.

Single steady wide shot.Let the supplied talking head establish the problem.
Shot group 02
≈ 00:19–00:25
App / browser screenWatch from here ↗
Sample 00:24.000
Codex composer

A Record & Replay request is entered into the new-task composer.

Wide app view moves attention to the prompt.Show the initiating request before the external task.
Shot group 03
≈ 00:25–00:28
App / browser screenWatch from here ↗
Sample 00:26.000
YouTube upload / file picker

An upload dialog and local file window establish the source media.

Overlapping native windows show the handoff.Keep input selection visible when it explains the workflow.
Shot group 04
≈ 00:28–00:30
App / browser screenWatch from here ↗
Sample 00:28.000
Publishing spreadsheet

A spreadsheet supplies publishing information.

Brief full-browser reference view.Show the source of task data rather than implying it appeared automatically.
Shot group 05
≈ 00:30–00:34
App / browser screenWatch from here ↗
Sample 00:32.000
Upload metadata

Title and description fields are populated in the upload dialog.

Stable browser framing through form entry.Group related fields into one readable action beat.
Shot group 06
≈ 00:34–00:37
App / browser screenWatch from here ↗
Sample 00:36.000
Subtitle import

Subtitle controls open a native file picker.

Modal-on-modal sequence retains the browser behind it.Preserve a meaningful file-import step.
Shot group 07
≈ 00:37–00:40
App / browser screenWatch from here ↗
Sample 00:38.000
Subtitle review

Timed caption rows appear beside the video preview.

Two-pane editor demonstrates the imported result.Show imported content in its destination, not only the picker.
Shot group 08
≈ 00:40–00:43
App / browser screenWatch from here ↗
Sample 00:40.000
Publishing visibility

The upload flow reaches its visibility choices.

The final form step stays legible.Do not omit a material privacy or publishing choice.
Shot group 09
≈ 00:43–00:49
App / browser screenWatch from here ↗
Sample 00:46.000
Recording completion

The conversation shows a done message and recording-processing status.

Return to the original task context.Make the handoff back to the assistant explicit.
Shot group 10
≈ 00:49–00:57
App / browser screenWatch from here ↗
Sample 00:50.000
Generated skill artifact

Skill documents and an edited-files summary appear in the response.

Close conversation crop emphasizes inspectable outputs.Show the reusable artifact before claiming the process is reusable.
Shot group 11
≈ 00:57–01:04
People / environmentWatch from here ↗
Sample 01:00.000
Presenter bridge

The speaker returns between the recording and replay examples.

Same desk framing separates the two demonstrations.Use a chapter transition instead in the silent presentation.
Shot group 12
≈ 01:04–01:08
App / browser screenWatch from here ↗
Sample 01:06.000
Replay inputs

A file window is used to select the next upload's source files.

Input selection is shown against the empty composer.Make the changed input obvious in a repeated workflow.
Shot group 13
≈ 01:08–01:12
App / browser screenWatch from here ↗
Sample 01:10.000
Skill invocation

The composer shows attached inputs and a request referencing the upload skill.

Prompt and attachment chips share the frame.Connect the earlier artifact to the next action.
Shot group 14
≈ 01:12–01:16
App / browser screenWatch from here ↗
Sample 01:14.000
Replay task state

The submitted task appears in a conversation with an output rail.

Brief established app state before browser activity.Retain a visible start state for an automated sequence.
Shot group 15
≈ 01:16–01:19
App / browser screenWatch from here ↗
Sample 01:18.000
Replayed upload

The browser opens the next file and reaches the upload details dialog.

External application remains identifiable.Show the real destination rather than an abstract progress animation.
Shot group 16
≈ 01:19–01:23
App / browser screenWatch from here ↗
Sample 01:22.000
Metadata and thumbnail

The new video's description and thumbnail appear in the upload form.

Successive form states within one browser view.Choose a visibly different result to prove the new input was used.
Shot group 17
≈ 01:23–01:27
App / browser screenWatch from here ↗
Sample 01:26.000
Replayed subtitle import

A file picker leads to the subtitle import options.

The same task stage returns with new material.Match framing across before-and-after workflow stages.
Shot group 18
≈ 01:27–01:30
App / browser screenWatch from here ↗
Sample 01:28.000
Imported captions

The subtitle editor displays the new video's caption rows.

Preview and timed text provide result evidence.Include a validation view after an automated import.
Shot group 19
≈ 01:30–01:34
App / browser screenWatch from here ↗
Sample 01:32.000
Visibility confirmation

The replay reaches the final visibility and save step.

Stable dialog framing through the closing action.Show the consequential last action without hiding its settings.
Shot group 20
≈ 01:34–01:38
App / browser screenWatch from here ↗
Sample 01:36.000
Channel content result

The content list shows the uploaded video and a visibility notice.

Result view replaces the upload dialog.End the demo on the actual saved item.
Shot group 21
≈ 01:38–02:01
People / environmentWatch from here ↗
Sample 01:50.000
Presenter conclusion

The desk presenter carries the closing explanation.

Long static human shot after the result.Keep this time with the talking head; do not stretch UI for filler.
Shot group 22
≈ 02:01–02:04
Title / identityWatch from here ↗
Sample 02:04.000
End identity

A white identity ending closes the film.

Centered wordmark and symbol.Use the presentation's own final card.

Reference frame

Enlarged reference frame

Watch this moment on the official channel ↗