I will admit something right now. I still stare at a blank page and just freeze up sometimes. It happens to all of us. Oh, absolutely. Yeah, it really does. I just sit there looking at the cursor and the cursor just blinks back at you. It expects absolute brilliance on demand. That blank page is deeply intimidating. It is. But imagine if like 95 % of your marketing work happened in just one tool. I mean, research, writing, designing, publishing. Right. Not just some basic
chat bot you argue with. We are talking about a full automated team. That is exactly what we are building today. Welcome to today's deep dive. We are exploring a really fascinating architectural system. We are going to build an AI marketing team. And we are doing this inside a workspace called Codex. Yeah, and whether you are a creator, a founder, or just someone trying to scale your ideas, we want to show you a powerful workflow. You are going to move from that terrifying blank
page to a massive automated content system. We are not just going to list a bunch of software either. No, definitely not. We are going to explore the actual underlying mechanics, like how do we ground the AI properly? How do we build specialized departments for research, design, and video? And how do we finally tame your chaotic inbox? Let's start with the foundation. What exactly is Codex? Because it feels fundamentally different from standard chatbots. Oh, it is completely
different. Codex is not a chatbot where you just type a prompt and get text back. Codex is a super app workspace. It actually executes your tasks natively. You can preview apps right in the interface. You can edit spreadsheets dynamically. You basically call upon specific skills and plugins. Wait, let me pause right there. So skills are basically reusable instruction files, right? They tell your AI exactly what to do. Exactly. And plugins
are bundles of specific skills. They are grouped together for your AI agent to use autonomously. You have a left panel for your chats. You have a middle panel for the agent's actual work. And you have a right panel for live visual preview. So Codex actually does the work instead of just talking about it. Exactly. It is a workspace where your AI actually executes the tasks. Okay. That makes sense. But to build this team, we first have to fix a glaring problem with AI.
Generative models often give us incredibly generic outputs. Oh, completely. You ask for a marketing hook and it sounds completely robotic. It feels totally devoid of any real human soul. Yeah, and the fix for that is a concept called grounding. Grounding means giving your AI specific, high -quality examples to learn from first. Think about how a large language model actually works under the hood. Right. It predicts the next most likely word based on its training data. Exactly.
So if you ask a standard AI for a viral intro, it just guesses based on everything. It pulls from incredibly broad Internet data. It mixes the styles of a fitness creator and a B2B sauce founder. And then it throws in a finance educator for good measure. Right. So you get one bland, completely useless output. It is kind of like handing a chef a very specific family recipe rather than just saying, make me some dinner. If you do not ground the AI, it is cooking with
every single ingredient on the Internet. That is a perfect analogy. That is why you're. output tastes like a strange mix of sauce jargon and fitness tropes you absolutely have to limit the ingredients instead of letting it guess You feed the AI actual YouTube transcripts. You give it your personal swipe files. You share your specific, highly detailed brand guidelines. You give it top performing hooks that you admire. Once the AI has those specific references, it narrows
its mathematical vector space. The probability of it generating your specific tone just skyrockets. So a normal workflow is just a prompt leading to a generic output. A grounded workflow finds specific structural patterns in your curated references. Yeah. That extra context changes the output quality immensely. Does grounding basically give the AI a creative compass? Right. Specific references prevent generic robotic sounding outputs completely. So since grounding requires
great references, where do we get them? We build a research department. We start with the YouTube Researcher skill. It pulls video transcripts to study hooks, pacing, and specific word choice. It looks at how a successful creator transitions seamlessly between ideas. Let's look at the actual mechanism. To set this up, you use an API tool like SuperData, right? An API is basically a software bridge. Yeah, exactly. It lets two applications trade data quietly in the background. It sends
a request to YouTube's server. It grabs the raw text file of the captions. It strips out all the messy timestamps. Then it feeds a clean text block right into the AI's context window. You got it. So you could study someone like Cleo Abram, and you feed her transcripts in, and the AI generates short -form hooks matching her exact rapid -fire pacing. Or let's say you pull Andres Karpathy's LLM video. Karpathy is brilliant at taking incredibly dense nodes of information
and breaking them down. Oh, he really is. You learn to explain complex topics in his beginner -friendly teaching style. Because marketing is not just about writing faster. It is about understanding and transferring ideas clearly. Exactly. And once you have that external data, you need your internal data. That is where you add the ReadWise CLI skill. CLI stands for Command Line Interface. Meaning your AI can use direct text commands to control ReadWise. It skips clunky visual menus
entirely. Right. And Readwise is essentially a digital second brain. It stores your saved notes, book highlights, and podcast transcripts. It holds your saved tweets and those random midnight thoughts. Because most good content does not actually start from a blank page. It starts from a fragment of something you already saved. I have to push back here, though. My Readwise is an absolute mess. I mean, whose isn't? Right.
So how does the AI know which of my random half -baked saved notes are actually worth turning into content? A lot of those notes... is complete garbage right and that is exactly where skill stacking comes in skill stacking means combining multiple ai abilities into one automated workflow you combine the readwise skill with the youtube researcher skill Codex Cross references them. It finds the exact mathematical overlap. It merges your private, messy idea bank with proven market
data. Oh, I see. Yeah. You ask it to review your saved items from the past week. Then you ask it to study the transcripts of your last 10 successful videos. The AI finds the semantic overlap. Then it generates 20 new video ideas that are both deeply personal and statistically proven. It merges my personal taste with proven market data. Exactly. Your personal ideas combined with proof. Okay, so now we have researched solid data -backed ideas, but we need to communicate them. Plain
text is not always enough for an audience. We need to build a design department. For that, we use the ExcalDraw diagram skill. ExcalDraw creates these simple, unpolished visual structures. It uses boxes, simple arrows, and minimal text. Under the hood, ExcalDraw is really just generating a JSON file, isn't it? Yeah, exactly. The AI writes code dictating coordinate points. Codex then renders those coordinates as a visual diagram on your screen. It is phenomenal for explaining
abstract concepts very quickly. For example, you can visually explain how your skills and plugins connect. Right. You just need the clear, bare -bones shape of the idea. So Excalibur is like the quick whiteboard sketch you do with a colleague over coffee. It's about clarity. But sometimes you need real polish. We all know the danger of overdesigning a simple concept. I have definitely spent three hours tweaking a graphic that should have been a napkin sketch.
We have all been there. But when you do need that final polish, that brings us to the paper MCP skill. MCP stands for Model Context Protocol. It is basically a standardized protocol letting AI safely operate external software tools. Right. And PaperMCP is for polished UI and high -end layouts. Think animated explainers and gorgeous website mock -ups. It is perfect for YouTube thumbnail concepts and slick Instagram carousels. Excalidraw is for quick, structural thinking.
Paper is for when the visual needs to be shared publicly with clients. And you can use subagents to speed this entire process up, right? Subagents are smaller, specialized. AI helpers handling parallel tasks. Exactly. One subagent is researching the color palette while another is building the layout structure. And with paper, you get live steering. Live steering. Yeah. You can literally watch the AI design the graphic in the right
panel. You get feedback mid -process. You just type, fix the overlap on the top left, make the label slightly shorter, the AI adjusts the code, and the design updates instantly. You guide it exactly like a creative director looking over a designer's shoulder. So we use Excalibur for clarity and paper for polish. Yeah. Sketch the structure. Static visuals are incredibly useful, but modern marketing constantly demands motion. We need quick iteration and dynamic content.
We need to build the video and media lab. This is where things get really advanced. We use tools called Remotion and Hyperframes. Remotion provides clean, professional UI and overlays. Think of a slick seven -section YouTube intro graphic. And Hyperframes handles advanced motion and complex, realistic physics. Right. What is fascinating is that Remotion is entirely code -based. It uses React. The AI is not dragging and dropping
clips on a time machine. It is writing React components that mathematically define where a visual element lives at any given millisecond. Whoa. Beat. Imagine generating complex video keyframes. Like an animated phone flying in, showing a scrolling group chat, and zooming out into a logo. Just by typing a single descriptive sentence. Beat. It honestly feels a little bit like magic. It changes everything about media
production. Because it is code, you can tell the AI to change the background gradient color exactly at the 10 -second mark. You can add a smooth ease -in spin right before the exit transition. Right. You do not have to build those keyframes manually ever again. You can also reuse old templates effortlessly. You just ask Codex to update the text and swap the brand colors. Then you export the final render straight into Premiere Pro or CapCut for the final polish. Exactly. Now let's
talk about the Gen Media Mini app. A mini app in Codex is a shared visual workspace for you and the AI agent. This specific app uses the FAL API for its media generation. I want to explain the mechanism here. The FAL API is essentially a lightning -fast cloud engine. It handles the massive computational weight of rendering generative media almost instantly. Yeah, so your local machine doesn't crash. And how does that specific media
workflow actually look in practice? The AI generates four different thumbnail options based on specific reference images. For example, it mimics the bold, contrasting visual style of Matt Wolfe. The options populate immediately in a visual grid within the mini app. It stores your image -to -video outputs and upscaled videos all in one place. The visual grid is a game changer. The human looks at the grid and picks the best one. Then you ask the AI to refine that specific
choice. You tell it to dim the background lighting. You add bold white text to the foreground. You make the main subject pop with higher contrast. You do not have to guess what the prompt will do. You iterate visually. Exactly. Visual grid stops us from endless scrolling and chat. Yes. Side -by -side comparison speeds up your creative decisions. We'll be right back after this short break. And we are back. The content is researched, designed, and fully rendered. Now we have to
manage the actual business side of things. We have to distribute the work and handle the inbound traffic. It is time for the operations department. We use the email manager or the brand deal manager skill. It actively searches your inbox for valuable sponsorships. We all know your inbox gets incredibly messy very fast. It is a chaotic stream of unstructured text. Missing the right email can mean missing a massive brand deal. Oh, absolutely. I want to focus on the logic here. It doesn't just read
words. It filters out the noise by running logical operations. It removes duplicate threads. It checks the brand's audience fit against your core demographics. Yes. It actively filters for actual paid opportunities versus people just asking for free exposure. It parses all that unstructured data into a highly organized priority table. The table explicitly shows the brand, the offer amount, the audience fit, the priority
level, and the exact next step. For example, it flags a high -paying sauce tool as an immediate high priority. It flags a random, vague agency email as a very low priority. It even integrates directly with your account? API to find open slots for intro calls. Yeah. It's incredibly organized. But wait, if I let an AI filter my inbox, isn't there a massive risk it archives a $10 ,000 brand deal just because the email
was formatted weirdly? That is a totally valid concern, which is why the AI does not delete anything. It simply labels and categorizes. It builds the priority table for your review. You still see everything, but the high value signals are pushed to the top of the pile. That makes sense. What about publishing the actual content we created? use the Buffer publisher skill. It moves the generated approved drafts out of the Codex chat environment. It places them straight
into a structured scheduling queue. Good ideas completely disappear if they stay buried in old chat threads. Buffer acts as a dedicated holding space. Right. You review your recent research. You choose the five absolute strongest ideas for LinkedIn. The AI drafts the posts and adds them securely to Buffer. I love this because it forces a separation between the creation brain and the publishing brain. Trying to do both at the exact same time is a guaranteed recipe for
creative burnout. But there is a vital non -negotiable rule here. You must never automate the final send. Never. The AI drafts the sponsor reply. The AI queues the social post. But the human always clicks approve. Your money, your reputation, and your relationships are on the line. Exactly. So it acts as a highly organized gatekeeper for my inbox. Great. It sorts out the noise. You make the final call. Hearing all these different skills and departments can feel completely overwhelming.
How do we actually build this system without breaking our current workflow? Well, the core building principle is patience. Do not try to build 20 skills on day one. That creates a very messy, fragile system. The progression should be simple and deliberate. Start with a single manual task. Improve the prompt carefully over a few days. Once the output is excellent, save it as a reusable skill. You just tell Codex to
save that exact workflow for future use. And you only automate the process once the output is reliably good. If the output is messy, automation just creates messy work much faster. Yeah. You can automate a daily read -wise summary to hit your inbox at 8 in the morning. But you should only do that when you actually love reading the daily output. Exactly. And that brings us to the... Ultimate big idea of this deep dive, skill
stacking. One single skill is definitely helpful, but chaining the YouTube researcher plus ReadWise plus Excalidraw is absolute magic. It equals a fully automated, unstoppable assembly line. We really must highlight the final 10 % rule. The AI handles the heavy lifting, the unstructured data sorting, and the busy work. It creates the rough first drafts and the various visual options. But you are the creative director. Taste. judgment, and high -level strategy strictly belong to the
human. The AI simply gives you more cognitive space to think clearly. It does. So my role shifts from being the intern to the creative director. Exactly. The AI grinds through the work. You direct the vision. Let us bring this deep dive to a close. I want you to think about this for a second. If an AI system can perfectly mimic our pacing patterns, parse our private notes, and flawlessly design our graphics, does our true enduring value as creators actually come
from our imperfections? Maybe the completely unexpected, unpatterned connections that only human intuition can make are the only things AI will never be able to replicate. Two sec silence. That is a profound way to look at it. The imperfections are the signature. Pick just one repetitive task today. Take something that drains your energy. Run it through an AI, refine the prompt, and see what happens. Just start building the foundation.
Thanks for joining us. We will catch you on the next deep dive.