A learn-by-clip guide · built from your vidIQ account
Every feature you asked about — Record & Replay, coding, automation, and the rest — shown by someone setting it up and getting a result. No reading about it. Watch it work.
Hey Andrew —
You wanted to understand the Codex desktop app using nothing but video, pulled from your vidIQ creators, with a 50,000-view floor and a real setup → result demo for each feature. Here it is.
Two honest notes first. Codex is OpenAI's app, not Anthropic's — and "Record & Replay," the feature you led with, is OpenAI's (it shipped June 18, 2026). And of your 17 creators, only Nate Herk goes deep on Codex; the rest live on the Claude side. So I featured Nate where he's strong, and filled the gaps with the most-watched creators and OpenAI's own demos.
Every clip below cleared 50k views. Each embed is cut to the moment of the demo, so you skip the three-minute intros. Where a video teaches more than one clip can hold, I've listed its chapter timestamps so you can jump around.
— ClaudeEvery clip ≥50k views · ranked by views within each section
The desktop features at a glance
The Codex desktop app turned from "a coding tool" into a general agent that runs work on your machine. The headline capabilities, in plain terms:
Below, the big ones get a demo. Sections are ordered the way I'd actually learn this, and within each, the most-watched clip comes first.
01 Start here
Two minutes from OpenAI to set the frame: the app is a command center for agents — multitask with several in parallel, keep their work isolated in worktrees, build skills, and schedule automations. This is the mental model everything else hangs off.
Setup → result: the whole clip is the tour — watch agents running side by side, then a skill and an automation being set up. It's the highest-viewed Codex video on YouTube, so it's the cleanest possible orientation.
02 The feature tour
This is your "list of everything," demoed. Riley Brown walks the seven core capabilities of the app as a super-tool, each one set up live. The clip starts right at capability one, and the chapter chips below let you jump to any feature.
Setup → result: each capability is shown end to end. Jump to the one you want:
03 The headline feature
The feature you came for. You demonstrate a recurring task once — filing an expense, submitting a request — and Codex turns that single demo into an inspectable, editable skill it can replay later with new inputs. You control when recording starts and stops.
Why I led with OpenAI here: Record & Replay shipped June 18, 2026, so it's only ~10 days old, and when I first built this none of your 17 creators had covered it at 50k+. OpenAI's own clip is the cleanest end-to-end demo, well over the bar at ~111k views. Since then Matt Wolfe's quick walkthrough cleared 50k, so there's now a non-OpenAI option too (below).
Setup → result: the full two minutes is the loop — start recording, perform the task, stop, and watch Codex generate the skill, then replay it. This is the "setup and the result" you asked for, distilled.
04 How people code in it
Tech With Tim hands Codex and Claude Code the same build task and judges speed, the finished result, and code quality. The clip is cut to the build and the finished product — you see the actual coding, then what shipped.
Setup → result: the clip opens on the build phase and runs through the finished product. For the setup itself, back up to 0:57 (the prompt/task) and 2:38 (model settings); the detailed code review is at 16:57.
05 How people automate in it
An engineer on the Codex team shows the automations he actually runs: a morning commit pulse, upskilling Codex overnight, triaging Sentry issues with memory across runs, and keeping pull requests green. They run on a schedule and carry context forward — exactly what "automation in the desktop app" means in practice.
Setup → result: short and dense — each automation is set up and shown running. Chapters: 0:18 morning commit pulse · 0:47 upskill overnight · 1:48 Sentry triage · 3:22 keep PRs green.
06 The wild one
NetworkChuck hands Codex control of a real PC: he sets the permission mode, lets it clean out the Downloads folder, strip bloatware, and then watches six sub-agents build a Hyper-V lab. It's the most visceral "agent uses your actual computer" demo out there — and Peter Steinberger (now at OpenAI) shows up in it.
Setup → result: the clip starts at permission modes (12:22), then cleaning the machine (12:48), then sub-agents building the lab (15:44). For the safety model — how the sandbox isolates the agent — see 6:56.
And the calm, official version of the same capability — Ari Weinstein on how Codex works across your local Mac apps without seizing the whole machine, including how permissions work app by app.
Setup → result: opens on Codex working across local apps (2:06), then multiple apps at once (3:41), how it reads the screen via screenshots + accessibility data (5:14), and app-by-app permissions (8:35).
07 Browser control
Install the Chrome plugin in the app and Codex can work across your browser tabs in parallel, in the background, without taking over the window you're using. Two and a half minutes, straight to it.
Setup → result: install the plugin, then watch it run across tabs while the user keeps working. Pairs naturally with computer use above.
Heads up: the dedicated creator Chrome-plugin walkthroughs are all still under 50k, so the closest non-OpenAI option above the floor is Riley Brown's broader take on Codex driving the browser.
08 Control it from anywhere
This one is as fresh as Record & Replay — it went generally available on all paid plans on June 27. You set it up from the desktop app: open the sidebar, choose "Set up Codex mobile," and a QR code appears. Scan it with the ChatGPT mobile app and your phone is paired to that Mac. From then on, your phone becomes a remote for the agent — start new threads, steer work mid-run, answer its questions, approve or deny commands, review diffs and test results, even switch models, all while the real work keeps running on your machine.
What it is, and isn't: a control and monitoring surface, not "coding on your phone." Your Mac does the work and holds your files; the phone keeps things moving when you step away, and the host has to stay awake and online. As with the other brand-new features, none of your 17 covered it yet — so this is OpenAI's official demo plus Riley Brown's hands-on setup guide.
The result, in 60 seconds: starting, steering, unblocking, and reviewing Codex work from a phone while it keeps running on the computer. Watch this first for the feel of it.
Now the hands-on version — Riley Brown walks the QR pairing end to end, then actually controls his computer from his phone.
Setup → result: opens on setup (pairing your phone), and the payoff — controlling his computer from the phone — lands at 12:15. Jump points:
09 From your list 👈
Nate Herk is the one creator from your 17 who goes deep on Codex. In one sitting he takes it from zero to a working YouTube-comment intelligence system — skills, automations, a deployed dashboard, and browser-use QA. The clip is cut to the satisfying middle: designing the dashboard through deploying it live.
Setup → result: the clip runs from designing the dashboard to a live deployment. The full build is chaptered — jump anywhere:
10 The rest of the library
Everything here also cleared 50k and covers the desktop app — useful if you want a second angle on any feature. Sorted by views.
One more from your 17: How I AI — "How to write AI agent loops in Claude Code and Codex" (52k) covers Codex alongside Claude Code. It's the only other video from your creator list that touches Codex at all — the remaining fifteen are Claude-side, which is why this guide leans on OpenAI's own demos and the most-watched outside creators.