Codex: Build Your Full AI Marketing Team (Agents + Skills)

Codex as the new marketing command center

Riley opens with a blunt stat: 95% of the content and marketing tasks he does on his laptop now happen inside Codex. He calls it OpenAI’s “super app” because it merges chat, docs, spreadsheets, app generation, browser use, and computer control into one place, with plugins and skills as the real force multiplier.

Skills, plugins, and the slash-vs-@ mental model

He pauses to explain the interface because the whole video depends on it: slash commands call a specific skill, while @ mentions a plugin bundle like Gmail, Calendar, Computer Use, or Vercel. His framing is simple and useful — plugins are bundles of abilities, while skills are instruction files that teach the agent how to work the way you want.

Grounding content in YouTube instead of generic AI taste

The first major tactic is “grounding,” which he describes as giving the model a useful reference point instead of relying on OpenAI’s idea of what good content looks like. With his YouTube Researcher skill, he has Codex pull transcripts via Superdata and write in the style of creators like Theo or explain concepts in Andrej Karpathy’s voice, and he’s visibly delighted when the outputs feel eerily accurate.

Turning Readwise into a live second brain for ideas

The next workflow grounds the agent in his own saved knowledge using Readwise CLI. Riley shows how bookmarked tweets and notes become a searchable idea bank, then asks Codex to turn the last week of saves into 30 short-form concepts — and when he notices missing source links, he just tells the skill to always include originals from now on, editing the workflow in plain English.

From one good prompt to a daily 8 a.m. automation

This is where his operating philosophy gets practical: do the task manually, refine it until the output is right, then convert it into a skill or automation. He demos that with a Readwise workflow that now runs every morning at 8:00 a.m., creating a fresh idea document in-chat so the system starts feeding him before he even asks.

Excalidraw and Paper for visual thinking, with sub-agents doing the prep

For diagrams, he uses Excalidraw to build low-text visual outlines for videos and Paper for higher-fidelity, animated canvases. One especially fun moment: he stops a slow run, tells Codex to use sub-agents for YouTube and Readwise research, and watches two named helpers spin up in parallel before the final diagram lands about 10 minutes and 50 seconds later.

Remotion and Hyperframes for launch-video polish

Then he shifts from diagrams to motion graphics using Remotion and Hyperframes, both enabled as plugins inside Codex. He shows a phone-demo animation where he can ask for a zoom at 8 seconds, a red gradient at 10 seconds, or a full 360 spin on exit, and ties it back to actual use in a recent 120,000-view video where animated overlays helped frame the whole story.

The ‘mini app’ idea: when both the human and the agent use the same tool

His favorite emerging pattern is a gen-media app he vibecoded with FAL APIs and a local database, which took about 40 minutes to assemble in its first version. The point isn’t just image generation — it’s that the agent can create assets directly inside the app, then Riley can jump in for the last 10%, like turning one generated Riley-on-a-tiger image into a more cinematic thumbnail with white text and a dimmed background.

Email, Buffer, and the pitch for Chorus

He closes on practical operations: a Gmail + Calendar workflow that summarizes sponsorship outreach, ranks companies by fit, and proposes meeting times, plus a Buffer skill that turns recent Codex work into draft content ideas. The final pivot is a soft launch for Chorus, his own cloud agent product, which he positions as a virtual computer that “never dies,” can run these skills persistently, and even plug into iMessage and group chats.