How Riley Brown Runs a 1.7M-Follower Content Business With Codex
A YouTube creator walks through the actual AI skills, thumbnails, hooks, diagrams, B-roll, that run nearly his whole content pipeline.
Posted
3 weeks ago
Duration
Format
Interview
educational
Views
26.5K
630 likes
57 · 43
Big Idea
The argument in one line.
A creator with 1.7 million followers runs nearly his whole content pipeline, thumbnails, hooks, B-roll, diagrams, through custom AI agent skills he builds once by describing a workflow and refines by testing outcomes, never by editing the skill's instructions.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A YouTube or short-form creator who wants to speed up ideation, hooks, thumbnails, or B-roll without hiring a full production team.
Someone already using an AI coding agent (Codex, Claude Code) for creative, non-coding work who wants to see what else it can automate.
A creator curious about turning a one-off AI prompt into a reusable, named skill they can invoke by name later.
SKIP IF…
You want a step-by-step technical tutorial with exact prompts and setup instructions, this is a conversational demo, not a how-to guide.
You're looking for video-editing automation, the guest explicitly says AI isn't there yet for editing and he still hires human editors.
TL;DR
The full version, fast.
Riley Brown, who has grown to 1.7 million followers making AI education content, walks through the Codex skills he built to run his content business almost end to end. He treats every workflow he repeats, pulling B-roll logos, drafting a hook from a top-performing video's structure, generating YouTube thumbnails by scraping references and swapping his face onto them in an AI design tool called Paper, as a candidate to save as a named skill he can call up again. Hooks get built around a five-part framework (big, relatable, easy, new, safe), and skills get refined not by editing files but by testing them repeatedly and telling the agent what it got wrong until it stops failing.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Riley explains that everything except final video editing, ideation, hooks, thumbnails, research, runs through Codex skills he's built over time.
03:16 – 07:28
02 · Demo: animations and B-roll
He shows his Remotion plugin generating opening-frame motion graphics, and an internet image-puller skill that scrapes relevant logos and B-roll straight from a video's transcript using the SERP API.
07:28 – 13:24
03 · Demo: reverse-engineering a top YouTube hook
Using a YouTube researcher skill built on the SuperData API, he condenses a competitor's video into a reusable 'hook outline' and applies that structure to a new idea.
13:24 – 17:24
04 · The BRENS hook framework
He introduces BRENS, big, relatable, easy, new, safe, as his mental checklist for what makes an intro work, and explains that 'safe' only matters on videos over roughly 30 minutes.
17:24 – 22:01
05 · Demo: AI stack for diagrams, thumbnails, and posts
He shows an Excalidraw diagram skill built from voice-dictated rambles, a Typefully skill for drafting tweets, and a Notion-integrated skill that manages his video idea database from inside Codex.
22:01 – 25:41
06 · How he sustains a high content volume
He explains his goal of four long-form and five short-form videos a week, and argues that chasing views instead of curiosity turns content-making into a 'soulless game.'
25:41 – 35:01
07 · Demo: Codex and Paper for viral thumbnails
He scrapes roughly a hundred high-performing thumbnails onto a Paper board, then uses AI image generation to swap his own face onto the best layouts, spam-generating seven variants at a time.
35:01 – 41:48
08 · Setting up AI agents for content ideas
The conversation turns to cloud agents in Slack (his Chorus product and Anthropic's Cloud Tag), how he tests and improves skills without ever reading the underlying files, and the business behind the content.
Atomic Insights
Lines worth screenshotting.
Riley Brown treats any AI workflow he repeats more than once as a candidate to save as a named skill he can invoke again later.
He has never opened one of his own skill files; he improves them purely by testing outcomes and telling the agent what went wrong.
Codex stores everything locally on the user's machine, unlike ChatGPT, which is why it can be granted full control of the computer.
GPT-based models inside Codex refuse to pull a YouTube transcript verbatim, citing copyright, so he switched to Claude for that specific task.
The BRENS framework, big, relatable, easy, new, safe, is his checklist for turning any idea into a strong video hook.
The 'safe' element of a hook only matters for videos longer than about 30 minutes, where viewers need reassurance their time won't be wasted.
He writes and films his intro last, after the full video is shot, because tangents often change what the video actually ends up being about.
His YouTube thumbnail process starts by scraping roughly a hundred high-performing thumbnails and pasting his own face onto the best-performing layout.
He generates seven thumbnail variations at once by spam-pressing enter in an AI image tool called Paper, then picks the best one manually.
Style references in Paper work by pointing the AI at an example image and saying 'make it look like this' rather than describing a look in words.
He dictates almost everything with WhisperFlow instead of typing, including 10-minute voice rambles that become the outline for diagram slides.
His publishing goal is four long-form videos and five short-form videos a week, though he admits he isn't hitting that pace yet.
He deliberately avoids letting packaging and projected views dictate what he makes, calling that approach a 'soulless game' that leads to chasing trends he doesn't actually want to make.
He runs cloud-based Slack agents rather than local Codex automations because he doesn't want his personal computer to have to stay on to run them.
Revenue comes from brand sponsorships, mid-roll ads, and AI-agent consulting, not from selling a SaaS product directly to his audience.
He predicts browser-controlling AI agents that stay signed into a person's apps will be common within the year.
Takeaway
Turn every repeated AI workflow into a saved, named skill
WHAT TO LEARN
The real lesson isn't which AI tool Riley Brown uses, it's that any workflow performed more than once becomes worth naming, saving, and improving by testing outcomes rather than reading the underlying instructions.
01How Codex runs the content business
Nearly every stage of content production except final video editing can be handed to an AI agent, from ideation to hooks to thumbnails.
Local AI agents like Codex can be granted full computer control because everything stays on the user's machine instead of syncing to the cloud.
02Demo: animations and B-roll
A motion-graphics plugin can generate a video's opening animation from a single prompt describing the desired flowchart or graphic.
A dedicated image-search skill can auto-detect relevant logos and B-roll from a video's own transcript instead of the creator hunting for assets manually.
Multiple narrow skills can be chained in a single request, for example pulling logos and then generating a graphic in the same prompt.
03Demo: reverse-engineering a top YouTube hook
A strong intro can be reverse-engineered by feeding a competitor's transcript to an AI research skill and condensing it into a reusable outline structure.
Different AI models vary in their willingness to pull a YouTube transcript verbatim, so the choice of model can matter more than the prompt.
Scraping an entire YouTube channel's transcripts for research can take as little as 30 seconds when an agent splits the work across sub-agents.
04The BRENS hook framework
A hook gets stronger the more of five elements it hits: big, relatable, easy, new, and safe.
The 'safe' element only matters on videos over roughly 30 minutes, where the viewer needs reassurance the runtime is worth spending.
Naming a clear, simple methodology inside a hook, a six-step method, nine tips, satisfies the 'easy' element and increases the promise's credibility.
05Demo: AI stack for diagrams, thumbnails, and posts
Diagramming tools can turn a rambled, spoken brain dump into a structured slide outline that frames the rest of a video.
A style guide isn't necessary when the AI tool supports visual references: pointing at an example image and saying 'make it look like this' replaces written prompts.
Routine drafting tasks, queuing social posts, writing emails, are worth delegating even when each one only saves a few minutes.
06How he sustains a high content volume
Letting projected views dictate what to make is described as a trap that eventually produces content the creator doesn't actually want to make.
Spending more AI-driven research time per video is a quality lever, not just a speed lever, since the agent can fact-check and study pacing while the creator does other things.
A publishing cadence goal (four long-form and five short-form videos a week) is stated as aspirational, not as a claim he consistently hits.
07Demo: Codex and Paper for viral thumbnails
A thumbnail process can start by collecting a hundred high-performing thumbnails and applying the creator's own face to the best-performing layout.
Generating several AI image variations at once and picking the best one manually is faster than trying to perfect a single AI-generated image.
An AI design tool with a shared visual board turns thumbnail design into an iterative, image-referenced process instead of a single one-shot prompt.
08Setting up AI agents for content ideas
Cloud-based agents are worth running instead of local ones when a workflow needs to happen without a personal computer staying powered on.
A skill should be improved by testing it, telling the agent what it got wrong, and retesting in a fresh chat, not by manually editing the skill's instructions.
Low-risk creative workflows can be left almost entirely to an AI agent's judgment, while anything involving payments or sensitive data should get manual review.
Glossary
Terms worth knowing.
Codex
OpenAI's AI coding agent, similar to Claude Code, that can run locally on a user's computer with full control over files and apps, including a built-in browser.
Paper
An AI-native design tool positioned as an alternative to Figma, built so AI agents can directly generate and edit images and layouts on a shared visual board.
Remotion
A tool for programmatically generating motion graphics and animations, used here via a Codex plugin to create opening-frame graphics for videos.
BRENS framework
A five-part checklist, big, relatable, easy, new, safe, for building a video hook that makes a viewer want to keep watching.
SERP API
An API that returns Google Images search results, used by a custom skill to automatically pull relevant logos and B-roll images for a video.
SuperData
An API used to pull a full YouTube video transcript in about a second, instead of downloading the video first the way yt-dlp does.
WhisperFlow
A voice-dictation tool used to speak ideas, diagram outlines, and prompts aloud instead of typing them.
Chorus
The guest's own product for building AI agents that live in Slack, iMessage, or group chats, referenced as the platform behind his personal agent team.
Cloud Tag
A feature that lets a user tag Claude directly inside Slack so it can act as a team member with access to a set of skills.
Resources
Things they pointed at.
01:06productChorus
02:17toolYouTube Researcher skill
03:30toolRemotion
05:23toolSERP API
07:40toolSuperData
07:50channelAlex Hormozi (YouTube channel, used as hook-outline example)
17:40toolExcalidraw
18:37toolTypefully
19:17toolNotion
23:53toolWhisperFlow
25:44toolPaper
40:50productCloud Tag (Claude in Slack)
Quotables
Lines you could clip.
16:51
“I think the moat is quality over a long period of time versus batching. Batching will make it soulless over time.”
tight contrarian thesis statement that works as a standalone quote→ TikTok hook↗ Tweet quote
27:36
“I don't like to think about thumbnails for that long. I despise making thumbnails manually. So this is great for me.”
“I don't think about any of that. I think about that after. I feel it in my heart. Like my gut tells me what I should talk about.”
sincere, quotable line about creative process over data→ newsletter pull-quote↗ Tweet quote
32:08
“I've never looked at a skill file once. It's test-based. It's like ask AI to do a thing to use a skill. If it does properly, you tell the AI, hey, you didn't do a good job.”
counterintuitive workflow claim that reframes how to iterate with AI→ TikTok hook↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphor
I will just scrape a hundred thumbnails that performed really well and I'll put my face on it. I took this and then I converted it into one for me and here are my thumbnails. Turn this into a skill.
The more skills you use, the more you'll realize that you can mix and match. That's where everyone's brain goes at first is to like batch and create more. What I want is better because I think that moat is quality over a long period of time versus batching.
Batching will make it soulless over time.
Hey everyone, my guest today is the one and only Riley Brown. Riley makes AI education content for over 1 .5 million followers. So super excited to get him to show us how he uses Codex and other AI tools to run his entire creative business and more.
So welcome, sir. Thanks for having me. This is going to be a lot of fun.
Yeah. You know, I feel like you're pretty much on all the platforms, YouTube, X, Instagram, and so on and so forth. Like, can you talk about your business at a high level and how do you use AI tools at a high level to kind of like create all this content?
yeah so at the center of everything that i do personally i like to educate people on ai and i've been doing this for the past three years that's how i've gained almost uh one i'm almost at like 1 .7 million followers across all the different platforms and obviously like we've uh released a couple products um including vibe code and now chorus um and then we're also doing a lot of like consulting uh helping companies implement agents in their business and it's just been so fun over the past three months talking to people seeing how um people are using ai agents in their business and you know me i'm a content creator That's like my profession.
And I use it for almost every part of my content business, except for like all, I really, really do think a lot of the video editing and I hire a lot of video editors and video editing isn't there yet. But like for all of the ideation, planning, coming up with good hooks, YouTube thumbnails. um doing all of my research is all grounded in youtube so i have like a youtube researcher skill that my agent will use whenever i ask it to do research um and so like i have a list of a ton of skills that just make creating content at scale a lot easier using agents Got it.
So YouTube's kind of like the hub of your content. And then like all the other platforms, you kind of make clips and share social posts and so on and so forth. I make dedicated content on Instagram and TikTok.
That's actually how I got my start. I just prefer longer form content. And I think Twitter is a great place to post long form content as well.
My long form content does as well on Twitter as it does on YouTube. And so, yeah, I'm like making content native to all the different platforms. Clipping is actually one thing I'm just now getting into, especially with AI, but we can talk about that a little bit later.
Okay, so can you actually show us how you do this end -to -end? You don't have to make the content, but I'd love to see the skills and workflows that you have. Absolutely.
And I guess what I'll start with here is kind of how I think about... Codex in general and platforms like Codex. So right here, we have Codex open.
Normally, when you come to Codex, it looks like this, right? It looks like ChatGPT. But when you go to ChatGPT, everything that you upload to it, if you were to upload an image, many people have done that, uploaded images or PDFs, etc.
All of that's stored in the cloud. When you use Codex, all of it's stored. locally.
And so Codex, just like Cloud Code, can fully control your computer. And in fact, it has a computer use skill built into the platform. It can actually control your computer for you, which is really cool.
And one really useful thing that I do for my videos, if you watch my videos, you'll see at the beginning, there has like really high quality animations. And the way I do all of my planning for this is I use this Remotion plugin. And so this Remotion plugin, I actually have it open right here.
This was one of my best performing videos on Codex. This was the opening frame. It was me saying like, today we're talking about Codex and all of the different things that it can do.
This graphic right here. Can you see it? Yeah.
Yeah. So this graphic right here is a graphic that I just generated. And you can very easily export this.
graphic and then i just placed it on top of my video when editing it and all you need to do to do this is you can this is just this isn't my skill this is a built -in plug -in into codec and whenever i want to use it i can just use my re -motion um i also have a re -motion best practices skill which is uh it has like my creator brand kind of infused into it so whatever it generates will be in my own like style if that makes sense Yeah, colors and stuff.
The colors and stuff. And so that's basically how I use it for creating these little graphics. So to make this graphic, you just update like a slash remotion, make this flowchart codex logo.
And then do you have to provide the assets or it just goes on and finds the stuff? So this is another thing that I've set up. So I have a skill.
I actually, I think it's called image pull.
Internet image puller. There it is. So this internet image puller is one where it'll go off.
It'll actually use something called the SERP API. And so it uses Google images within the SERP API, and it'll actually find the relevant logos related to whatever video it is. So if I say, hey, I need some B -roll for this, it'll automatically look at my transcript of the video that it's creating B -roll for, and it will find the logos, and it will just pull them.
And you can use multiple skills for a prompt. So I would probably say something like, please pull the relevant logos and then make a graphic for this. And then I would just use the Remotion skill, which I call Remotion Best Practices, and it will just generate it, and it will pull the relevant logos.
But you can manually just drop them in as well. Okay, so you're basically chaining multiple skills together, right? Yes.
I guess you could say that. This episode is brought to you by Google AI Studio, the easiest and fastest way to build and share your idea with real users. For example, I use AI Studio to create three different landing pages for this bakery using images generated by Google's built -in nano banana model.
One highlights fresh bread, another highlights awesome cakes, and the last highlights the bakery story. I can publish the website easily and get a custom URL, in this case, corgi -confections .ai .studio, to share all three variations with potential customers for feedback in one tap.
When I was a product manager, Google AI Studio was my go -to tool for building prototypes like this. And it really is an easy and free way to share your ideas with your team, customers, and everyone else. Check it out at aistudio .google .com.
Now back to our episode. Let's go to the beginning, actually. We're actually still brainstorming ideas for the video that you can make next.
Do you like a competitive analysis or some sort of what people are talking about skill? Yeah. So one thing that I like to do, so I have my favorite skill is this YouTube researcher skill.
If you want to set this up, and I also have a link to a lot of the skills that I've created on my website. If you guys want to check that out, it's free. I don't charge for that.
Great. but it uses this thing called super data which is one of my favorite apis it allows you to like pull the full transcript from your videos from any youtube video like in one second so i can scrape an entire channel if i wanted to i could just say like youtube researcher create a peter yang bot um and or skill and it can scrape all of your videos and it can act like you or something like that um and so this is my youtube researcher skill and oftentimes when i use For me, when I create content, the intro is the most important part.
And then the rest of the video is kind of free flowing. I'm naturally good at kind of explaining how to do things once the video, once I'm into the video, but the intro is really important. So another prompt that I have is called a hook outline.
So hook outline will create a very specific type of document for me. And so I'm going to say like, please look at the latest Alex Hormozy. video and write a hook outline for it.
And like whenever I find a good YouTube video, you can condense it down to a hook outline. What is the intro? And then just outline the rest of the video.
And what it'll do is it'll allow you to like, you can take that exact format and apply it to a different video. And so that's what I do a lot is I'll create a hook outline. based on someone else's really well -performing video.
And then I'll tell AI to take all of my ideas for a new video and convert it into that structure, if that makes sense. Are you following? Yeah, I'm following, yeah.
So I've been using YT DLP, but you're saying that this other thing can just pull the transcript in one second? Yeah, I mean, it's just like a quick API call. Like YouTube DLP, if I'm not mistaken, it'll download the full video and then pull the transcript.
that is super data intensive and it's yeah super data will just it'll re you can say spin up sub agents that's one thing you can do uh with codecs and they have a really cool ui you can just ask it to use sub agents and it'll just it can like it'll like split off into like six agents and it will scrape it all in like 30 seconds you can scrape an entire channel in like 30 seconds Got it.
I'm not promoting that you do this, but it is really good for research. Yeah, I mean, it is a fact that the first three seconds, if it was boring, people would check out, right? So basically, you don't do like a full voiceover guide when you make a video.
You just have the intro scripted out, and then you have like an outline for the rest. Is that how it works? Yes, exactly.
Yeah, I mean, why do people... I think with AI, the fully scripted videos are going to go down in value. I think what will go up in value over time with AI is you have to hook the viewer.
You have to play the game to some degree, but you also just want to get good at talking about the thing that you're super passionate about. And so that's what I like to do. I like to show the human side and then having guests on is a really good way to do it.
Like your podcast, you have a structured intro, and then you just talk to someone for the rest of the time. That's why I think people are here. And so you can see here, this is a very specific format.
And one thing that's really annoying about Codex, and one of the reasons why I've been moving off of it, is because the GPT models won't pull the transcript. You can see here that it summarized his intro. The reason it did that is because of copyright infringement.
It didn't want to because it doesn't. Yeah, it doesn't. But Claude's models will do it without thinking.
So that's why I've been using Claude for this use case. Interesting. That's funny.
That Claude is actually more open than GPT. It's the other way around. It's task dependent.
The open models don't care either. Like GLM 5 .2 will do it any time. Let's talk about the hook a little bit.
I actually watched a couple of the recorders' videos. And I actually like how they're like more educational. Even the hook is like, here's what I'm going to cover in this video.
As opposed to, you know, Codex is like the most freaking amazing, like super, super hype stuff. So is that like a choice or? Yeah.
I kind of feel like you have more, you have more substance. Yeah. I mean, I, I try when I make videos, I try to make it about the topic rather than any specific tool.
right i think that i always like to make it clear that even if someone uses cloud code they're going to get value out of the video and so the my best video that i did on codex which got 300 000 views which i think is the one you're referring to um that was like the nine like nine ways to use codex but i made it clear in the beginning by the way you can do all of this in uh cloud desktop as well most of the things you can do in cloud co -work as well and i like to focus on like the specific outcomes that they'll get from the video which I think is really useful to establish in an intro.
Okay, so if you have some tips for newbie YouTubers like me, talk about the outcomes first and then get into the content, right? Yeah, talk about the outcome. In fact, I don't know if you would...
Did we film the intro already or do you do a separate intro? Are you going to film a separate intro? I usually cut a separate intro from the best parts of our conversation.
Yeah. Okay. That's what I do.
Okay, that makes sense. And then do you do, like, another little intro where you say, like, today you're going to learn this, this, and this? Or do you just kind of, like, go straight into the video?
When I was introducing you, like, that was kind of my intro. Like, talk about how you're going to use it to run your creative business and so on. Yeah, I mean, for me, if I were you, I would be doing testing.
Like, obviously, you want to get really good guests that are really knowledgeable on certain things. And that's something that you can and probably already are trying to improve. But I would just test different...
intro types until you find the winning format. You will find that a good, you know, there's a, there's a, there's actually a really good acronym. It's called B -R -E -N -S, which is big, relatable, easy.
Let me see if I can zoom in here. Easy, safe, new, and safe. The more of these you can include in your intro, the better, right?
Big, like there's a brand new model that just came out. It's the best in the world. Make it big.
Relatable is you find some theme. So like you say, like there's this new model. It's absolutely insane.
It just came out and it's going to affect everyone or it's going to go into every business, right? That's relatable. Easy is like, and by the way, I have a six step method that you can follow that will guarantee that you will be successful, right?
Like when you introduce some sort of like, methodology i'll let here are these nine tips that you can follow that will guarantee easy uh new obviously people like things that are new so like you could you know this brand new model also is big and new and then safe is the one that that is really important for you um safe is good is important this is only really important for videos like over 30 minutes it's like You want to make sure the user feels safe watching, that they're not going to waste their time watching the video.
And the way that you do that is you say, like, by the way, this guy's an expert. He's going to talk about these things. And like, I promise you, by the end of the video, you will get value out of it.
And you want to make sure that they, like, their time is being valued. Yes. Got it.
And how do you think about, so do you start with the hook or do you, how do you think about the title and the thumbnail, right? I don't think about any of that. I think about that after.
So like, I think about the thing that I want to talk about in the video, which is like codex tips, right? This is something that I would want, like, I don't do a lot of research on what I should talk about. I feel it in my heart.
Like my gut tells me what I should talk about. For 90 % of the video, it's just me. I'll outline it, but it's just me getting into the weeds and talking about how I actually use this tool on a day -to -day basis.
But then the last 10 % is like, okay, how do we get people to watch it? And that's where the science goes in. And so I start with everything that I'm going to talk about in the video, and I just outline that.
And then I film the whole video, and then oftentimes I'll film the hook after.
completely separate video because sometimes i'll go off on a tangent in one of my videos like some of my best videos i end up on like a 30 minute tangent and then but i want to make sure that my intro talks about that thing that i do so i that's why i just do my intros at the end and i use ai to analyze everything i talk about and then that informs my my transcript and then i'll have the youtube researcher say like here's the transcript for my video.
I need a good intro. Scrape YouTube until you find a good format and then just like copy it and make it about what I'm talking about, if that makes sense. Oh, that's like super interesting, dude.
Because like I know a bunch of other YouTubers who make the thumbnail first and the thumbnail is not good and they will not even make the video. But I guess, yeah. Well, I mean, maybe, you know, like for me, the most sustainable strategy for content is just to like keep doing things that I like.
And if I let... packaging and and how many views it'll get dictate what content i make then like it just becomes a soulless game to me like i would much rather make what i want and then package it after the fact um then to just like be you know only like about viral ideas then i would end up you know it's like why not then i'll end up making dance videos on tick tock which is something i don't want to do you know You got to think about what gives you energy and follow your own curiosity.
If you can make content from your own curiosity, that's the best.
To continue talking about Codex, I have a lot of skills in here. One thing that's really fun to do is I use Excalidraw in my videos a lot. A lot of people watch my videos.
I diagram things out. A lot of the times, so there's a skill called Excalibur diagrams that you can use. And you can literally just tell Codex, hey, I need an Excalibur diagram skill or make an Excalibur diagram.
And it'll just make it. And then the link, you can just like paste it into Excalibur. It'll actually open it in the browser.
And it makes really good diagrams of different ideas. That's one that's really useful. I also have a YouTube thumbnail transcript.
This one will get me into trouble a little bit because... literally my youtube ideation strategy when i have to come up with a youtube thumbnail is i will just scrape a hundred thumbnails that performed really well and i'll put my face on it and then it'll i'll just ask ai to just like replace those seven with my face and then i'll have ai edit the color um which is which is some people would get mad at me for that but like that's it's been a really useful skill um And yeah, I just have a ton of things that I've ended up just using over and over again.
Also, there's a typefully skill. So if I get an idea for content, this can control my typefully, which is connected to my Twitter. So it can really quickly draft up my tweet ideas.
Like if I'm working on something or come across something, I'll be like, hey, draft a tweet on this. And then every Monday I check my typefully and then I fire off those tweets. Oh, interesting.
Yeah, and so... What I'm trying to get at is like all of these skills did not come from me sitting down and like thinking like, okay, what skills do I need to create? I just use AI to do stuff.
And then everything that I. that's useful that i do over and over again i just tell codex turn this into a skill and then over time i get in the habit of using certain skills and sometimes i'll stop using skills but that's kind of the thing and the why the reason why i love codex so much is they have so many built -in plugins and their in -app browser is really cool um in fact here this is just right before we were filming this video i'm working on a short form script and the way that i write all my short form scripts is i just start a new notion page in codex and so in my so i have this so i have this video database this is where i like keep track of everything right this is my long form videos here are my short form videos that i keep track of because they're in -app browsers becoming a full in -app browser i can use it directly inside codex and so what i want you to what i want to show you really quickly right i could say hey i want to make a short form video
on the latest meta Ray -Bans. Please add a new short form in progress to Notion and give me link, right? I'll often do this.
I'll usually say more stuff, like more ideas. I'm just trying to illustrate the point that like right now, the agent is going to go into my Notion. It's going to add it into the database correctly.
And then it's going to give me a link. And then I'll be able to open that up directly in the app. I won't have to leave Codex.
It'll just open directly inside Codex, which is such a fun workflow. Like I love not having to leave the app. I get lost when I switch from tab to tab.
You see here, it gave me a Notion link. I can just click open in browser. And look at that.
It added this. So it added the notes in here, which is fine, but it did create the Notion database. And I can say like, please make the outline.
in the body and please make it with a good hook. I would normally use say that this will be a huge opportunity if they take off for creating apps on these glasses. I don't know.
And yeah, just control the Notion, which is really fun. And this is how I use Google Docs and Notion and my calendar as well. I'll just say, give me a link to the calendar event and I'll open it in the in -app browser so I can just see it right there.
And it can also write all of my emails, which I have it do all the time. Yeah, I have it write my emails too. So basically then one day you can just wake up and you have a bunch of scripts already and you can just batch process.
record them in batches right or something yeah yeah i think i think that's where everyone's brain goes at first is to like batch and create more what i want is better i like whenever i create content i i'm always thinking about how to create stuff that's better and what i want is like i want multiple tabs to open up you know in the background of like relevant things that might be useful like codex would be like hey like By the way, I opened up this link.
You may want to, in this one specific line, you should do this. And so I will actually use it to just get AI to fact check everything that I'm talking about in this video. Or I'll just send it off.
And I'll say, hey, can you go and check this creator's videos? He has really good pacing in his videos. Can you please figure out a way that we can implement that into this video?
It'll go off for 10 minutes and come back. And that's kind of how I see this. You can do research so much faster.
So it allows you to put more total research into one video. Because I think the moat is quality over a long period of time versus batching. Batching will make it soulless over time.
So that's kind of how I think about it. And I would encourage others to do so. Got it.
At this point in your creative journey, how many short forms are you publishing every week? One or two that are high quality? Yeah, I mean...
I would like to create a video once a day, honestly, like five days a week. I would love to create a high -quality short -form video. As of late, I'm trying to get my long -form to like four videos a week first.
And I think four long -form, five short -form is my goal, for sure. I think that would be good. That's great, dude.
I make one or two videos a week, and I'm already exhausted. So I need to use code as more.
Yeah, I don't know. I like it. It's fun to make content.
I don't know. Yeah. Dude, I have to ask you, okay, so let me ask you a few more questions about the skills that you kind of showed.
So the Excalibur skill, do you use that after you have the script or before you have the script? Like you make the diagrams? Yeah, I mean, I think I will just blabber.
So 99%, I know I'm typing now because this is a podcast and I don't want to like use the Whisperflow. I use Whisperflow for everything. And so what I will do sometimes, And I got the AirPod Maxes because they have a better mic.
I'll just turn on WhisperFlow, and I'll just walk around for 10 minutes just speaking all of my ideas. Like, oh, I'm going to make a video on Codex. The things that I've been thinking about this week are this, this, and this, and this, and this.
I want you to create an Excalibur diagram of this, this, and this. And then I'll literally tell it. I want nine slides.
My Excalibur diagram skill does it in a very specific style, which like has just like sections to them. And yeah, I'll just blabber for 10 minutes and then I'll be done and I'll just like run it. And then I'll come back and it will have like 80 % of the diagrams that I'll actually use in my videos.
And I'll edit them. I'll take 20, 30 minutes to edit them. But it's like a really good starting point.
It just organizes my ideas. And yeah, huge fan of Whisperflow for that reason. Okay, so the slides actually kind of frame the rest of the video, right?
Because you want to cover like eight slides and like do some demos and stuff. Yes. Got it.
So then I have to ask you about the thumbnail skill because I have a lot of trouble using AI to make thumbnails for some reason. Sure. So what I found is that it's actually really bad at changing your face or maybe I don't know how to prompt it.
If I give it like a picture of me smiling and I tell you like change your frown or like a surprise, it totally distorts my face, dude. So like, do you mostly apply your face afterwards? So one of my...
most popular tools that I talk about that I use with Codex is a tool called Paper. Paper is an AI native Figma. And when I say AI native, I mean that it is an application built for AI first.
They basically took a lot of the good stuff about Figma and they just made it really easy to connect to any tool. And so when I'm using my thumbnail skill, I will say, hey, I need you to scrape. This is a short -form video that I'm releasing in a couple days where I ask AI to scrape Alex Ramosi and Dan Martell's YouTube thumbnails and place them on the paperboard.
Codex will go off for 10 minutes, finding his most popular videos for each one, and then it will place them on this board. Now all I have to do is I can come here, and since I have images of me, right? This is a thumbnail of me.
I can duplicate it. And then I will just come here and do a lot of it manually. So, like, I'll grab this one of Alex Ramosi.
I'll drag this over here. And then paper has AI image generation built in. And so watch this.
I can just go like this, select this. Please replace the guy with a beard with the guy that doesn't have a beard. Just replace the guy except make his shirt white and make him wearing a chain.
You want him to wear a chain?
Just because this podcast is important, it's important that I'm wearing a chain in this thumbnail. It's quite important. But it looks like it's generating multiple thumbnails?
Oh yeah, I just spam pressed enter. So it's creating seven versions of it. And then I'll just kind of select the right one.
And you can see kind of my natural progression, right? You can see that this says Game Changer. I took this and then I converted it into one for me.
And here are my thumbnails. it's kind of fun to kind of see the progression of the thumbnails that I've created over the, like, you know, I created this one right here, which was based on this one. This is Andre Jick.
He has great thumbnails. And so that's kind of how my thumbnail process. I don't like to think about thumbnails for that long.
I despise making thumbnails manually. So this is great for me. Oh, dude, this is amazing.
So, okay. So it basically just replaced it. It's kept his hand, it looks like.
Yeah, it kept its hand. I could have specified to change it, but then at any time, you know, like what I like to do is I'll zoom in on this one, right? You know, this is not bad.
I can come in, oops, I can come in again and make it small. I like to make it smaller and then I'll just make more variations. Like, hey, please smooth out the face without changing like what the character looks like.
And then also add like an outline glow that is like kind of like whitish orange around him. to separate him from the background and make the green line brighter and then i'll just kind of iterate on it um and like this gives you this kind of like starting point but the the useful part about codex is codex has full control over paper um and so it can just like add useful thumbnails to it because again you know going back to what you were saying about uh what did you call it chain prompting or something Yeah, it can go off and do a thing and then it can use a different skill and place it on the board because it has that access.
And yeah. And do you have some sort of a, okay, so right now you're just prompting in paper, right? With WhisperFlow?
Yeah, yeah, yeah. Do you have some sort of like a style guide or some shit? Like, you know, your text is always white or something.
You don't need that. So like I have, I paid, I actually have a YouTube thumbnail designer too and he uses AI all the time and he helps me with this and I'll actually send him these. I could use these and it would do pretty well.
I love having someone who is constantly doing the A -B testing and everything. I do a lot of A -B testing because why not? It's worth paying for because if your video does twice as good, that's a big deal.
See, it kind of made the outline here. But in terms of style guide, if I want it to look like something, I'll just use it as a reference. So here I have somewhere I have all my thumbnails.
I don't know where I put it, but let's say I wanted it to look like this. I would just copy this thumbnail, bring it up here, and then I would say, and you can very easily copy multiple. So I can literally select this one, and then I can select this one, and I could say, please make the guy look more like this and make the text match the one that says Mythos and Fable.
And that's how you do style references. You don't need to do any prompting. A lot of these image models are called elements.
And so you can just think of prompting. You can prompt in visuals. You can say, make it look like this.
Make the text look like this. And so that's why I like the board. I can just be like, oh yeah, I want it to look like this.
And then I'll just use AI to make it look more like that, if that makes sense. Dude, this is super educational for me. Okay, I think I'm going to become a paper customer now too.
Yeah, you should 100 % do this. Yeah, you should 100 % do this and get all of your top 10 performing thumbnails and toss them on the board. You don't need to use Codex.
If you had my YouTube thumbnail skill, you could just say, Codex, do that, and it would be done. All right, dude. Well, you definitely blew my mind with this thumbnail thing.
This helps a lot. I don't even know what other question to ask you. That's something that's so mind -blowing.
Yeah. And then I think what it all comes down to, if I were to summarize, is just find useful workflows. and just like turn them into skills.
And the more skills you use, the more you'll realize that you can mix and match. Like this here is a mix between my YouTube researcher skill. I just had an idea one day.
I'm like, hey, can you pull all these images? And then it gave it to me in this list. And then I had to like go into my downloads and see it and I couldn't visualize.
I was like, oh wait, you can add it to paper. And like, because I'd been using the paper skill to create designs, right? This is, paper's not made for images really.
It's made. to like you can ask it to create full designs like and it'll do that. And then I was like, wait, can you add this to paper?
And then it unlocked this whole new workflow for me. And you'll you the biggest unlocks you'll find when using AI will come from this like experimentation process. That's just like really fun.
Do you look at the skill files manually? No, you just look at the output, right? No, I've never looked at a skill file once.
It's test -based. It's like ask AI to do a thing to use a skill. If it does properly, you tell the AI, hey, you didn't do a good job.
And then please change the skill so that you don't do that again. Then you want to go to a new chat so the context is cleared so you can kind of test it again. Test the agent in a new chat.
See if it does it successfully. If it does it successfully, great. right and once it does it incorrectly you update the skill right that is the way better way to do it than to like manually edit the skill and any sort of like oh there's a lot of people saying like oh you need to write your skill like this or else or else it's not going to perform well it's like first of all that's only true with that model and it's only true with that model today right as soon as there's model updates so i think i think you just want to be outcome based right hold your ai to a certain standard And then you can say, turn it into a skill and you can improve it by measuring their outcomes.
That's basically how I do it for most creative workflows. Yeah. And I feel like a lot of these skills are like just super personalized for you or like for me.
That's kind of the power, right? I feel like that's why skills is kind of taking over a lot of the SaaS stuff. Because like, you know, you can't make a SaaS product as personalized as this.
Yes. And actually, I will caveat that with, this is for my creative workflows, right? They're very low risk.
You know, if... If we were doing some mission -critical thing, which we actually do when we're dealing with clients or when it has to do with anything that has to do with payments or sensitive documents, yes, we'll go in and analyze the skills. And there's kind of like a process.
We'll get someone else to analyze the skill if it's super high impact. But for my creative workflows, just realize that there's no risk. Just let your AI cook and tell it to do a better job.
And then the next thing for me is figuring out how to make this uh for my team and that's what i think codex and claude are anthropic and open ai they're trying to figure out it's like how do you uh enable this process with teams and that's going to be a really interesting situation and it feels like anthropics going after slack um and so that's kind of the next wave maybe i'm not fully sold on slack as the interface but like it's there's something really interesting going on there that um Yeah, I find it to be really interesting.
Yeah, I think you're right. I think as good as the Codex app is, it's like a single -player experience, right? Other than putting PRs up and stuff.
It's pretty much a single -player experience. It is single -player, and it's really hard to create these creative workflows in a multiplayer way, which is something I think about a lot. Do you have a bunch of automation setups that do work for you?
Yes, I have a lot of automation setup. I don't use Codex for my automation simply because I don't want my computer to have to be on. uh to run those um and so i have cloud agents yeah and so like i have a lot of cloud agents in fact i created a peter yang bot right before this episode i didn't create it and i was going to show you but like it's not i i created it in like five minutes so like i have like my chief my chief marketing officer agent that i have in slack like he sends me um this guy he sends me um And then I have another guy named Content Man who analyzes a bunch of other people's content.
And it'll just like give me video ideas for short form content. And so I have this little channel called My Agents. And so I have three agents in here.
One of them is Peter Yangbot, which is trained on all your YouTube videos. But I didn't finish it in time to like make it good. No problem.
So these agents are based on your own product chorus or is it based on like us? Yeah. No.
We've been testing it with a bunch of different products. These three agents are created on Chorus. But there's a lot of products out there that you can create agents.
And I just recommend trying it because no one's figured this out yet. No one's figured out how to create a team of agents running in Slack that your whole team can use. It's a really tough problem to solve.
The way Anthropic's trying to do it is they're trying to build a little mini... for lack of a better term, God that knows everything. Where other companies, including us, where we think of it like each one will be team members.
So like you'll tag your marketing agent and you'll also tag your developer agent, company developer agent. And each, you'll kind of have a relationship with the agent. Like each person, it'll react differently to each person based on what it knows about the person.
There's like a whole bunch of problems that need to get solved, but you can very easily export your Codex skills into these cloud agents, which are really fun. And so, yeah, it's just been a lot of experimentation. But cloud agents, tell me wrong, if I want to tweak something where I'm signing to like some sort of app, cloud agents can't do that, right?
Or like they need your credentials? Yeah, I mean, with browser use, I predict by the end of the year, you'll be able to be signed in to all of your apps and your agent will be able to use some browser and you won't even be aware of it.
It'll just be able to use a browser almost perfectly well and it'll be able to be signed in. And I think this is something that the big labs companies want to do as well. Being able to control a browser in the cloud is much better for automations and cron jobs as well.
I don't know if you have any automation set up in Codex, but it is kind of annoying. I have none set up right now. I have Codex running on my little Mac Mini.
I have some automations there. That's kind of how I get around it. Yeah.
That makes sense. But then if it's a separate computer, it doesn't have access to your core files, which might be important. It's a pain in the ass, yeah.
Pain in the ass in so many different ways. Every agent type has its trade -offs, which... which we're seeing right now.
All the companies are trying to figure out the right way to do agents. And maybe it'll take another six months to figure out the best way to do it. Cool.
All right, dude. Well, let me ask you one last question on the overall business. So I feel like the content, from one content creator to another, I feel like just growing the audience and getting more sponsorships, that's kind of a surefire path because there's a lot of AI companies who want distribution and attention.
And then there's also a lot of demand. Like you said, there's a lot of demand from companies that want to build agents, train their employees. But in terms of actually building a product, that's a lot more difficult than building a SaaS product.
It is a lot more difficult to do. And I do brand sponsorships and I run mid -roll ads and it's a great business. I mean, it's an insane business.
I could very easily just kind of ride off into the sunset and make a very good living making content. I think my content would run dry if I did so because I learn being in the trenches. You know, like I enjoy, the only reason I know a ton about Codex is I spend time with developers every day who like teach me really cool things.
You know, like every day I'm learning like three or four things and we're actually like, you know, and the content that I make is I actually want to help people implement agents into their business. You know, I want to be a very useful creator and a motivating creator. I don't think I could do that if I was just making content.
And then and so, yeah, I think doing like especially like consulting stuff, I think it really does help, especially if you build like a really talented team. You know, my goal is to only be making content. I can still contribute to the business through top of funnel.
But ideally, like all I'm doing is making content and you just have to surround yourself with a really good team. And there's a lot of talented people who would love to partner with you and do all of the work in creating a product. And I think you would learn a lot from it or anyone in your position.
It is just, there's a lot of upfront work that goes along. Yeah. I don't want to just be a same, same here.
I don't want to just only make content or like be like some sort of news aggregator. You know, I want to actually learn how to build the stuff and talk about the building process. So a hundred percent.
I mean, yeah, like, like, you know, Peter levels on Twitter. Yeah. You know, like he could never be a content creator.
Like he's only interesting because he talks about the things he's built and the experiences he goes through. And so that's kind of how I think about it. It's like, you know, that's the creator I want to be is someone actually trying to build and use this technology.
Yeah. Yeah, he's one of the few folks who actually have a very successful, like a bunch of successful SaaS companies because his distribution is so high. And no, but a lot of failed companies as well.
He's failed. 40 times. He just takes a ton of swings at bat, which is also something that makes him really interesting.
Cool. All right, Ryan. Well, I think people know where to find you online.
Do you want to do like a pitch for your new product course or should we talk about it next time? I mean, yeah, I mean, we're building, you know, Cloud Tag just came out. Cloud is a product that allows you to tag Cloud in Slack.
And then you have all your skills built into Cloud. We're building a product that allows you to use any model. You can message it in iMessage.
You can message it in Slack. You can add it to group messages, which is what our users like the most. And we're constantly iterating, trying to help businesses.
Right now, I'm not even really selling the product as much as selling our services because companies want to work with us. And then we're using Chorus to implement agents for their business. And that's just like an insane driver for us.
Cool. All right, well, you're an inspiration, man. Thanks for being so open and sharing your whole process.
And I personally learned a ton. So thanks so much. Amazing.
Yeah, thanks for having me.
The Hook
The bait, then the rug-pull.
Riley Brown opens with a needle-drop of his own best lines, then spends 40 minutes showing exactly how a 1.7-million-follower content operation runs on named AI skills instead of a production team.
Frameworks
Named ideas worth stealing.
13:55acronym
BRENS
Big
Relatable
Easy
New
Safe
A checklist for building a strong video hook. Big means claiming scale ('the best in the world'); relatable ties the idea to something the viewer already cares about; easy promises a simple method ('nine tips that guarantee success'); new leans on novelty; safe reassures long-video viewers that their time won't be wasted.
Steal forany video intro or headline that needs to earn attention in the first few seconds
07:28concept
Hook Outline method
Feed a well-performing YouTube video's transcript to an AI research skill and have it condense the intro into a reusable outline structure, then apply that same structure to a new video idea.
Steal forreverse-engineering competitors' intros without copying their words
CTA Breakdown
How they asked for the click.
VERBAL ASK
40:50product
“we're building Chorus... you have all your skills built into Claude... Right now, I'm not even really selling the product as much as selling our services because companies want to work with us”
soft, low-pressure plug woven into the closing conversation rather than a hard sell; earlier in the video he also points to a free public library of his skills on his own website
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Anthropic's Thariq Shihipar walks Peter Yang through /goal, sub-agent workflows, and a planning habit built entirely around finding out what you don't know before you build.
Replit's head of product engineering walks through three non-coder founders who built real six-figure businesses, then live-demos the four steps that turn a raw prototype into one.
A creator puts GPT-5.6 and Claude Fable 5 through six real build tasks -- travel sites, games, browser agents, and a nutrition tab -- to settle which one earns the daily-driver seat.
00:32
01:18
01:49
02:17
02:50
03:11
04:26
04:57
05:29
06:00
06:28
07:09
08:05
08:37
09:08
09:39
10:11
10:42
11:14
11:45
12:16
12:48
13:19
14:22
14:53
15:24
15:56
16:27
16:58
18:01
18:32
19:04
19:35
20:06
20:38
20:58
22:43
23:15
23:46
24:17
24:49
25:20
26:09
26:54
27:39
27:57
28:28
28:48
29:31
30:33
31:05
31:36
32:08
32:39
33:10
33:42
34:13
34:44
35:16
35:47
36:50
37:21
37:52
38:24
38:55
39:26
39:58
40:29
40:58
41:17
Video of the Day44:39
36:06
28:52
28:29
Video of the Day21:17