All videos
0:00 / 0:00
ai

Claude Code and Codex Quietly Learned to Watch Video

Mark Kashef8 August 2026Watch on YouTube

Part of series

Ep. 13 · Claude Code Free

View the series

What you'll learn

  • Claude Code and Codex can now process a raw MP4 file without a plugin or skill and understand what happens in picture and audio.
  • Video processing arrived silently through the record feature, without an official announcement from Anthropic or OpenAI.
  • A video works best in a lower resolution, because higher resolution demands more processing power in the context window.
  • The Codex record feature is not available in parts of the EU and does not capture audio, which can be solved with an own Loom recording.
  • The record feature can be bent from skill creation into a live audit of a manual workflow, producing an SOP and API advice.

Frequently asked questions

What can Claude Code and Codex do with video since the recent update?
Both coding agents can process a raw MP4 file without a plugin or skill and understand the picture and audio. The presenter demonstrates pointing to a Loom link from which the model pulls the video itself.
Why does video processing work better in a lower resolution?
The model breaks the video into frames that travel through the context window together with the audio. Higher resolution demands more processing power, so a ten to twenty minute video fits best in low resolution.
Is the Codex record feature available everywhere?
Not everywhere. In parts of the EU the feature is not enabled, according to the presenter. An alternative is making your own screen recording with Loom and feeding that to the model.
How can the record feature serve as a process audit?
By explicitly asking the model to make no skill but watch the manual workflow for thirty minutes and then produce a standard operating procedure with advice on how an API can automate steps.

Topics

multimodal aiclaudevideo processingai automationproductivity

Sources

What is known about this topic outside the broadcast, and where it says so.

Description from the channel

Master AI: https://www.skool.com/earlyaidopters/about Work With Me: https://www.promptadvisers.com A few weeks ago Claude Code and Codex both shipped screen recording features, and the part nobody announced is that the same work quietly made both of them better at processing raw video. In this video I break down three ways to use it out of the box, dragging in a raw MP4, pointing it at a bare link, and hacking the record feature to watch me work over my shoulder and audit my process. Along the way I turn a Loom of my website into a full improvement plan, clone an award winning 3D site from a narrated recording, and start rebuilding a SaaS platform I pay for every month. You used to need Gemini for this. Now it is sitting inside the subscriptions you already have. --- 0:00 - the upgrade nobody announced 1:00 - drag and drop a raw MP4 into both 2:00 - how it actually watches (frames + audio) 3:00 - the flipbook trick and the resolution tip 4:00 - use case 1, website improvements from a Loom 6:00 - community 6:30 - use case 2, cloning a 3D website from a recording 8:20 - use case 3, rebuilding a SaaS platform for myself 9:30 - hacking record and replay to audit my workflow 12:45 - the SOP it handed back 13:00 - where this leaves you #ClaudeCode #Codex #AI #Anthropic #OpenAI #AIAutomation #AIAgents #VideoAI #Loom #Productivity #AITools #Claude #Multimodal #Workflow