47:51Riley Brown · Interview
OpenAI Merges ChatGPT and Codex
Riley Brown and Ras Mic dig into GPT-5.6, Codex's background computer-use, and why self-scoring agent loops are turning coding tools into a general operating system.
July 12thThe argument in one line.
AI cannot write good marketing hooks from a blank prompt; the system that collects, filters, and routes real buyer language is the only thing that makes AI output trustworthy.
Read if. Skip if.
READ IF YOU ARE…
- You run or market an ecommerce brand and your AI-generated ad copy consistently sounds generic or off-brand.
- You are building AI agent workflows for marketing operations and want a concrete pipeline architecture.
- You brief UGC creators or affiliates and need hooks grounded in actual buyer language, not model guesses.
- You have customer data -- reviews, support tickets, comments -- and no systematic way to turn it into copy.
SKIP IF…
- You are a service business or B2B brand with no direct consumer feedback at scale.
- You are looking for a single better prompt, not a multi-stage workflow to build or commission.
The full version, fast.
Most AI-written hooks fail because the model invents market language instead of reflecting it. The fix is a six-stage agent pipeline: mine raw customer noise from reviews, comments, and tickets; extract the exact buyer phrase and log its source; identify the emotion underneath the complaint (fear, doubt, confusion, desire, objection, use case); craft a content angle that matches the buyer's awareness level; route that angle to the specific asset type and platform before running the pipeline; and feed performance data back to the start so the system self-improves over time.
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.
Create a free account →Where the time goes.
00:00 – 00:42
01 · The blank canvas problem
AI cannot humanize content it was never given. Years of trying GPT and Claude from scratch, always falling short.
00:42 – 01:36
02 · Customer noise
Step 1: mine continuously from Amazon reviews, TikTok comments, Reddit, support tickets, post-purchase surveys. Goal is better raw information, not better content.
01:36 – 02:25
03 · Buyer phrase
Step 2: extract the real phrase a buyer used, preserve the source. Do not clean it up. A hook without a source just guesses.
02:25 – 03:45
04 · Buying pain
Step 3: go under the surface complaint to find the emotion -- fear, doubt, confusion, desire, objection, use case. Mushroom coffee 1-star review example.
03:45 – 05:08
05 · Content angle
Step 4: reframe the buyer's self-perception. Bad: 'try our supplement.' Better: 'you're not lazy, you're just stressed.' Angle should feel like the customer got caught talking out loud.
05:08 – 06:15
06 · Route
Step 5: define the destination asset before running the pipeline. Hook is not the output -- the route is. TikTok, Meta, email, UGC brief all need different framing.
06:15 – 08:18
07 · Feedback / readback
Step 6: map CTR, watch rate, hook rate back to the angle that generated them. Hermes OS agent scrapes TikTok and Meta ads automatically. The loop self-improves.
08:18 – 09:32
08 · Channel context and sign-off
Non-developer framing: built this OS using Claude Code, Codex, Hermes -- hundreds of markdown context files -- as an ecom operator, not a software engineer.
Lines worth screenshotting.
- AI cannot humanize content it has never seen -- the quality ceiling of any AI hook is the quality of the raw customer data you gave it.
- A hook without a traceable source is a line that guesses and sounds confident -- and confident guesses are the most dangerous kind of bad copy.
- The goal of mining customer reviews is not to find better content ideas -- it is to get better raw information from buyers, which then produces better content as a byproduct.
- Preserving the exact phrase a real customer used -- and the source it came from -- is more valuable than cleaning it up into polished language.
- Customers rarely hand you usable marketing language; they hand you surface complaints that hide the real buying emotion underneath.
- The angle that works is the one where the buyer feels caught talking out loud -- not the one that sounds like an ad.
- Defining the route (TikTok hook, email subject, landing page section) before running the pipeline changes what the AI optimizes for.
- The feedback loop is the hardest part of the system to build and the most ignored -- without it, the workflow never improves.
- Every platform has a different hook grammar; Meta, TikTok, and Instagram are not interchangeable destinations for the same angle.
- An AI operating system built on hundreds of markdown context files -- even by a non-developer -- can automate the parts of a marketing workflow that would otherwise require a full analyst.
The input determines what AI can output.
WHAT TO LEARN
AI will never write a hook better than the buyer data you feed it -- the workflow that collects and refines that data is the real leverage point.
- Collecting raw, unfiltered customer language from reviews, comments, and support tickets gives AI something real to work with -- polishing the data before the model sees it removes the signal.
- The source of a phrase matters as much as the phrase itself; the same complaint on Amazon versus Reddit often signals a different buyer segment and calls for a different angle.
- Surface complaints rarely contain the emotion you need -- you have to extract the fear, doubt, or desire underneath the literal grievance before the angle becomes useful.
- An angle that reframes the buyer's self-perception lands because it triggers recognition, not because it promotes the product.
- Defining where a hook will be used before writing it changes the entire structure -- platform context is a constraint, not an afterthought.
- Closing the feedback loop by mapping performance metrics back to the content angle that generated them is the step most people skip and the one that makes the whole system compound over time.
- An AI operating system built on hundreds of context files -- even by a non-developer -- can automate the parts of this workflow that would otherwise require a full marketing analyst.
Terms worth knowing.
- Customer noise
- The unfiltered volume of raw customer language across reviews, comments, support tickets, and surveys -- messy, context-free, and the highest-signal input available for any AI-driven copy workflow.
- Buyer phrase
- The exact words a real customer used to describe their experience, preserved with its source attribution, before any AI cleanup or paraphrasing.
- Buying pain
- The emotional layer underneath a surface complaint -- the fear, doubt, confusion, desire, or objection that actually drove the customer behavior, not the literal grievance they wrote down.
- Content angle
- The framing perspective for a hook or ad that makes a buyer feel immediately seen -- typically built by reframing their self-perception rather than promoting the product.
- Route
- The destination asset type and platform (TikTok hook, email subject line, UGC brief, landing page section) that a content angle is written for -- determines format constraints before writing begins.
- Readback / feedback loop
- The stage that maps real performance metrics (CTR, hook rate, watch rate) back to the content angle that generated them, so the pipeline learns which buyer profiles and framings are working.
- Hermes OS
- The presenter's custom AI operating system -- a harness of hundreds of markdown context files that powers agent workflows for his ecommerce business, including a performance-scraping readback agent.
- Solution-aware / product-aware buyer
- Awareness levels from direct response marketing used to segment how much a prospect understands about their problem and the available solutions -- affects which angle and hook structure will land.
Things they pointed at.
Lines you could clip.
01:52
“A hook without a source is a line that just guesses and sounds confident. We cannot have that.”
04:24
“The angle should in a way feel like the customer got caught talking out loud.”
05:40
“The hook is not the final output here. The route is really the output.”
00:08
“AI should not be inventing your content from a blank canvas.”
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.
metaphoranalogystory
00:00AI should not be inventing your content from a blank canvas. If you're asking AI to generate hundreds of hooks for you, you are wrong. You need to give it the proper context.
00:10And in this workflow, I can show you how to start from scratch and end up with a polished script, hook, whatever you're looking for that is based on actual evidence for your target audience.
00:21And trust me, the past several years, I have been stubborn just thinking that GPT, Claude, is gonna just give me an output of hooks for my TikTok videos, for my landing pages, whatever, and it never can do it correctly. It cannot humanize the content.
00:35It just is bad at doing it unless you give it the right tools. So let's dive in. The first and one of the more critical points to this workflow is the starting point, which is going to be the customer noise.
00:48Now the customer noise can be from anywhere, and that's why you need to develop an agent workflow that's going to mine it continuously, and it's gonna obviously get it from the right direction. So this could be from Amazon reviews, TikTok comments, Reddit posts, comments, competitor reviews, support tickets, post purchase survey analysis, and even chat boards.
01:09I don't know. Anywhere your customers are and they're posting about it or they're writing about it, your agent needs to mine that information because it's very valuable. The goal is not to create better content.
01:22The goal is to get better raw information from the buyers themselves, and then that will equal better content.
01:30So once you have all this information in this customer noise, what we're going to do now is figure out the buyer phrase. With this information, you should be able to use AI to help you build out phrases that work for these buyers.
01:47Save the real phrase that somebody used. Save save where it came from, and you'll see a correlation between these because these types of audiences typically say the same things on these sources, these social media platforms. You don't wanna clean it up too much.
02:01You wanna leave it and make sure that in the back end of your OS, it is showing the distinction between the comments and the source. Because ideally, . We cannot have that.
02:15Once you find the real phrase, the agent finds the pain underneath it. So that will lead us to our next point, which is to get into the buying pain. Customers rarely hand you the perfect marketing language.
02:28They just don't they're gonna ask things like, does this actually work? I hate the taste. I don't wanna feel weird taking this.
02:35I bought something similar to this, and it did nothing. Just a lot of fluff in my opinion. There's zero context to any of that.
02:43If you didn't like the taste, what did you not like about it? What flavor did you have? Was it sour and you just don't like sour?
02:48Then why did you even buy a sour flavored product? I had a mushroom coffee company that you grounded like regular coffee. So you put it in you put the grounds in the coffee pot, let it brew, etcetera.
02:58But because traditional mushroom coffee was instant, people weren't reading the directions on the bag and they were just throwing in a cup of water, stirring it, and drinking it.
03:07Obviously, it didn't taste well, but we got this bogus feedback in a lot of one star reviews on Amazon around this. And it was so frustrating because the customer, in our eyes, was incorrect. But that kind of information that we're getting was not gonna be helpful here in this workflow.
03:22What the agent needs to find in this buying pain is the emotion behind it. The fear, the doubt, the confusion, the desire, the objection, and the use case. But it's very important that that's what we're pulling.
03:34We're extracting this from that buyer pain even if we have no context on it. That's why this is important because when we're crafting an angle, it's not gonna be random. It's really gonna come from emotion from other customers.
03:46And that leads us to the next point, which is the angle. Let me give you some examples of angles from this D2C supplement brand that we have called No Stranger. A bad angle, obviously, is try our supplement.
03:57A better angle and kind of what we attack is, look, you're not lazy, you are just a stressed out person, and you don't even realize you're stressed. So to reduce your stress levels, you need to take something that's easy once a day and it's all natural. See, it's changing the perspective of how you word the angle because they feel the relevancy right away.
04:17They they hear like, oh, I I know I'm not lazy, but maybe I am stressed. You gotta understand the phrases and also what kind of buyer are we talking to?
04:26A solution aware person, a product aware person, all of these things matter. The angle should in a way .
04:36Right? Like, I'm not gonna go and tell people this product tastes great. K.
04:40Let's just pretend it has flavor to it. The better angle is the hardest part of taking supplements is not the discipline. It's not hating the taste of it, which some people have problems with that.
04:51Because then if you have these agents working up into this point, it's gonna take the angle you had based on the previous three stages and it's gonna route it. And that's the next step is routing. Routing is important because we have customer noise, we have the phrases, we have the pain points, and we know the content angle that we crafted based on those.
05:12How do we route this? Well, this is gonna vary for every business. Where are we using these?
05:17Because hooks are used everywhere for ecom. Could It be an ad hook of a static image ad of a UGC video. It could be an email subject line, a landing page section, an FAQ section, an entire TikTok script, a YouTube script, customer support information, or just you're testing your offers, which is one of the more important things to do.
05:36And that's why it's like the hook is not the final output here. The route is really the output. Yes.
05:41The hook is what matters, but you need to know where it's being routed. And that should be actually delivered from the beginning of this process understanding this workflow can be used in a variety of areas.
05:55But today, I wanna just focus on my UGC creators, my affiliates. I wanna help them find good hooks. So you know this going in.
06:03We're gonna route this as TikTok hooks, and that context is gonna matter because the hook matters on which platform or which area we are utilizing it. And then the last step of this workflow and a thing that gets under mind all the time, people just ignore it, is the read back.
06:21Now it's very important you do this. In almost all my agent workflows, I use Hermes agent as my harness for my, um, entire OS of my business of the code base I built. But the readback has been the more difficult things to build at times, but also the most important.
06:39What we need to do is train this OS on what's working and what's not working. I mean, that's name in the game of ecom.
06:46Right? We all know that. So if we're using all of these different alternatives in contextualized phrases and making sure we're going this phrase for this buyer, this phrase for this type of target audience, etcetera, we need to give it some sort of performance review.
07:01Now I have an agent that is literally designed in my Hermes OS that generally scrapes my Tik Tok, my meta ads, all of the points of where my channels or revenue are coming from, and it points revenue or just performance, CTR, watch rate, hook rate, etcetera, back to the content angle that we ended up crafting or the hook that we crafted.
07:24This is a little bit more complex and probably needing of another video, but you have to close the loop here. You've got to give the performance to what was done because this workflow will improve over time if that's the case.
07:38If we know that people that are taking this are hybrid athletes that also work nine to fives, that are just so stressed out, they're very motivated individuals, and we know the hooks are working for that buyer point, now we can tell the beginning of this workflow to find more like this. Or it could be the opposite.
07:55If it's not working, we tell it, look. It's not really working. We need to expand our horizon of our search and find different niches, different pain points, whatever the case may be.
08:06This is gonna help you. I see too many ecom brands that are just shooting in the dark from what they think from common sense wise is going to work for their brand, and every platform is different.
08:18Meta ads is way different than TikTok. TikTok is way different than Instagram in my opinion. So because of that, you need to make sure you're really mining the correct information to give you the better output.
08:31Now if you like these workflows and you want more of it, please like, subscribe to this video because I wanna be dropping more of what's working for our brand and how we are doing e comm because I don't see a world where a lot of people are going after these agent harnesses and utilizing them in the e comm space. I know people are doing it, but I don't see much content around it and that's kind of why I opened up this channel.
08:53Every time I see people using OpenClaw, Hermes or just Cloud Code, Codex, etcetera, it always is coming from a place of software development and I'm not a software developer. I have used these tools to develop software, but I have used these tools to create me a giant operating system for my brand, which is basically hundreds of markdown files for context.
09:16So I can build automations that are based on that context and give me better outputs for my business and for my brand. It has been helping a ton.
09:25So if you want more like this, like I said, like, subscribe, let me know. I'm always open. My DMs are always available.
The Hook
The bait, then the rug-pull.
The problem with AI-written hooks is not the model -- it is the starting point. When you prompt an AI to generate a hundred hook variations from nothing, you are asking it to invent an audience it has never heard from. The result is confident-sounding language that fits nobody in particular.
Named ideas worth stealing.
00:42model
The Hook Workflow
- Customer noise
- Buyer phrase
- Buying pain
- Content angle
- Route
- Feedback
A 6-stage agent pipeline that turns raw customer reviews and comments into platform-specific hooks, with a performance readback loop that improves the system over time.
Steal forAny brand building AI-assisted copy workflows -- the architecture applies to email, ads, landing pages, and UGC briefs equally
How they asked for the click.
VERBAL ASK
08:18subscribe
“If you like these workflows and you want more of it, please like, subscribe to this video”
Soft and earned -- delivered after the full framework is explained. Positioned around the channel mission (ecom operator workflows), not the specific video.
MENTIONED ON CAMERA
Visual structure at a glance.
