20:08Helena Liu · Tutorial
How to Get ChatGPT to Recommend Your Business
A 20-minute GEO tutorial covering every on-site and off-site lever that makes AI chatbots cite your business instead of a competitor.
June 12thThe argument in one line.
Because leading AI models have converged to within 5% of each other in capability, the real edge left for a business owner is installing free, open-source multi-agent skill sets on top of a chatbot instead of prompting it one-off.
Read if. Skip if.
READ IF YOU ARE…
- You run Claude, ChatGPT, or another LLM as a single-prompt chatbot and haven't tried loading multi-agent skill packs on top of it.
- You make recurring strategic calls (pricing, growth channel, new offer) alone and want a structured second opinion before committing.
- You need fast, sourced customer or market sentiment (Reddit, X, YouTube, TikTok) and are currently relying on generic AI answers or manual searching.
- You want a working software prototype (landing page, small app) built end-to-end by an AI 'team' rather than hiring a designer and developer.
SKIP IF…
- You already run a structured multi-agent or agentic workflow in your own stack.
- You need production-grade, audited code — these are demo-speed prototypes, not reviewed for security or scale.
The full version, fast.
The video's thesis is that AI models have converged in quality, so the real differentiation left is tooling, not model choice — and free, open-source "teams" of AI agents already exist on GitHub for anyone to install. The core mechanism is Claude Code/Cowork's skill system: cloning a GitHub repo into the local skills folder registers new slash-commands and personas that run inside Claude with no extra cost. Three examples are demoed live: an LLM Council (5 debating personas modeled on Karpathy's multi-model method) that argues out a business decision and returns a verdict plus blind spots; a Last30Days skill that scrapes live sentiment from Reddit, X, YouTube, and TikTok instead of returning a generic answer; and Garry Tan's gstack, a full slash-command engineering team (CEO, designer, staff engineer, QA lead, etc.) that built an animated landing page from one prompt in about four minutes. The actionable conclusion: pick a well-starred, actively maintained GitHub repo, install it into Claude Code once, and reuse the resulting slash-commands instead of re-explaining context in every new prompt.
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 – 01:40
01 · Stanford AI report thesis
Models have converged in quality; most businesses still use AI like a chatbot, not agents. Sets up the video's premise.
01:40 – 02:15
02 · Channel intro + free masterclass plug
Helena introduces her channel focus (agentic AI, automations) and free Agentic AI masterclass.
02:15 – 07:00
03 · Skill 1: LLM Council
Explains Karpathy's multi-model debate method adapted into 5 Claude personas; shows the GitHub repo and star count; installs via git clone into Claude Code.
07:00 – 07:40
04 · LLM Council live demo
Runs a fictional bakery pricing/growth question through the council; shows the debate transcript, blind spots, and final verdict.
07:40 – 11:55
05 · Skill 2: Last30Days
Introduces Peter Steinberger's ~46k-star sentiment-scraper repo, installs it, explains what platforms it scans (Reddit, X, YouTube, TikTok, Instagram, Hacker News).
11:55 – 13:27
06 · Last30Days demo: Zapier + Claude IPO
Runs sentiment research on Zapier and on Anthropic's IPO speculation; reads back sourced complaints, praises, and links.
13:27 – 15:00
07 · Skill-augmented vs vanilla Claude comparison
Runs the same prompt without the skill loaded to show the generic, unsourced answer versus the sourced Last30Days output.
15:00 – 16:27
08 · Skill 3: gstack (Garry Tan's engineering team)
Introduces the 115k-star repo from Y Combinator's CEO; shows the roster of AI personas (CEO, designer, staff engineer, QA lead, etc.) and the install command.
16:27 – 17:14
09 · gstack live build + reveal
Prompts the skill to build a landing page for a fictional AI startup (NexSys); reveals the finished animated 3D site a few minutes later.
17:14 – 17:48
10 · Recap + CTA
Recaps the chatbot-to-agent-team framing, points to install instructions in the description, pitches the free masterclass, asks for like/subscribe.
Lines worth screenshotting.
- Stanford's AI report finds the performance gap between the best and worst large language models has narrowed to under 5%, meaning model choice matters less than the tooling built on top of it.
- Installing a multi-agent GitHub skill only requires a single git clone command run inside Claude Code — no coding knowledge is needed.
- A GitHub repo's star count is a practical trust signal for vetting free, open-source AI tooling before installing it.
- The Karpathy-style council method assigns distinct reasoning personas (contrarian, first-principles thinker, expansionist, outsider, executor) to debate the same question instead of asking one model once.
- A skill-augmented prompt for market sentiment can return exact sourced quotes and links from Reddit, X, YouTube, and TikTok, while the same prompt without the skill returns a generic, unsourced summary.
- A full landing page with animated 3D graphics was generated end-to-end from a single prompt in roughly three to four minutes using a multi-persona engineering skill set.
- Loading a skill once creates a reusable slash-command, so future requests skip re-establishing context that would otherwise need to be re-typed in every new prompt.
Free multi-agent skill packs beat one-off prompting
WHAT TO LEARN
Because model quality has converged, the real leverage left in using AI is installing pre-built, open-source multi-agent skill sets instead of re-prompting a single chatbot from scratch every time.
- Model choice matters less than it used to — leading LLMs are within about 5% of each other in capability, so the differentiator has shifted to the tooling and workflows built on top of the model.
- A well-starred, actively maintained open-source GitHub repo is a practical, low-effort way to vet whether a piece of free AI tooling is worth installing versus wasting time on it.
- Structuring a single question as a debate among multiple distinct reasoning personas (contrarian, first-principles, outsider, etc.) surfaces blind spots that a single-pass answer misses.
- A tool that pulls live, sourced, quote-level data (real Reddit/X/YouTube comments and links) is categorically more useful for research than a generic AI summary with no citations.
- Installing a skill once creates a reusable command, which removes the need to re-establish context or re-explain a task in every new prompt going forward.
- A working software prototype — landing page, small app — can now be produced by an AI 'team' end-to-end in minutes, compressing what used to require separately hired design and development work.
Terms worth knowing.
- Claude Skill
- A packaged set of instructions, personas, or slash-commands that gets cloned into Claude's local skills folder, extending what the assistant can do beyond a single prompt-response exchange.
- Claude Cowork / Claude Code
- The desktop version of Claude that supports installing external skills and running local commands, as opposed to the web-based chat interface which cannot install these skill packs.
- LLM Council method
- A technique, attributed to Andrej Karpathy, of asking multiple AI models or personas the same question so they can debate and cross-check each other before producing a final answer.
- Slash command
- A short typed command (e.g. /last30days) that triggers a pre-loaded skill or persona inside Claude instead of requiring the user to re-describe the task from scratch.
Things they pointed at.
07:40toolLast30Days (Peter Steinberger)
15:00toolgstack (Garry Tan)
10:50productZapier
Lines you could clip.
00:12
“At this point, artificial intelligence has been commoditized. It doesn't really matter which model you use anymore.”
00:26
“AI can now act as agents where they can start taking actions.”
17:14
“This is how you can go from just typing in a prompt into ChatGPT and getting an output to actually getting much more meaningful results from your AI and elevate yourself from a basic AI user to the top 1%.”
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
00:00Stanford has just released their annual report on AI. It's over 400 pages, but there are two findings that I think are worth noting, so I'm going to mention it here. The first is that, basically, all of the large language models are converging.
00:13Now a couple years ago, there's a big difference between the best models versus the worst models. Today, that difference is less than 5%. So, basically, it doesn't matter if you're using a paid model, free model, or an open source model.
00:25They are all working pretty well at the moment, and that that gap is only going to further decrease and decrease and decrease. And what this means is that, basically, at this point, . It doesn't really matter which model you use anymore.
00:42And the second key finding in this report is that most businesses have not caught up with the technology. The technology is basically sprinting forward faster than businesses can keep up. Most business owners are still using AI like a chatbot.
00:58So you go into ChatGPT or Cloud, you type in a prompt, you get a response. But, however, AI can now act as agents where they can start taking actions. And that's exactly what I'm gonna show you how to get started in this video.
01:11In fact, there are prepackaged agents that I will give to you and show you how to find more on your own as well. This will speed up your learning curve and make it easier for you to adopt AgenTeq AI.
01:23Now if you're new to my channel, hi. My name is Helena. And on this channel, I love to talk about everything AgenTeq AI, automations, tech.
01:31So if you want to be in the know on the latest technology advancements, please make sure you like and subscribe, and this will help out this channel so much as well. I've also got a free course on AgenTek AI, and I will show you in-depth how you can start building your first AI agent. So make sure you sign up for this free training below as well.
01:49Now let's get right into it. A great hack is to find prebuilt AI agents for free on GitHub.
01:56So there are a lot of developers all around the world who have spent hundreds of hours building these AI agents, and then they just decided to open source it. There are a couple of projects that are really good, and there are three in particular that stood out to me. And I'm gonna show you what these three projects are.
02:12The first one is LLM console skill. So this is based on Karpathy's method. So Karpathy, he is one of the cofounders of OpenAI, and, basically, he invented this method where there are different LLMs that you can ask the same question to, like ChatGPT, Gemini, Claude, Perplexity, and they will debate each other in order to come up with the best answers for you.
02:37So this skill and this set of agents is a variation of that. Instead of using multiple LLMs, it will all use claudid, though, but it still works quite well.
02:47So let me show you what's inside of this skill right here. There are five advisers that will be computing your answers in and parallel. You have the contrarian, the first principles thinker, the expansionist, the outsider, and finally, the executor.
03:01Okay? So after you learned what this project is about just by scrolling down on the GitHub page now another thing you want to know is, like, how many stars and how many forks it does in this particular project have. When it has a lot of stars, it's an indication that this particular project is very popular.
03:18That's another way to help you vet good free projects versus the one that are just not worth your time. Now the next question becomes, how do we load this GitHub project into our own Cloud account so that we can start using these AI agents? Right?
03:33So in order to do that, you want to download Cloud CoWork. So in order to make any of these skill sets that I'm about to share with you work, you need to download the desktop version of Claude.
03:45They will not work on the web based version, but it is free to download this as well. Now just go to claude.com/download, and then choose your operating browser.
03:56For me, it's the Mac OS. Click here, and then log in to your Claude account once you have this downloaded. This is what it will look like once you have Claude downloaded onto your desktop.
04:07You will have three tabs here across the top, and you want to go over to the third one that says code. Now in order to get the LM Console GitHub project into our own Claude, this is the line of code that we want to put in. Basically, we're telling Claude to clone the GitHub folder right here, and we want it to clone it into our own Claude folders that is sitting in our desktop.
04:35Okay? And so once we run the skill successfully, this is what it will look like.
04:40Now we can actually get our quad account to debate each other. We can ask it business questions like, should I raise my prices? Should I launch this product?
04:48Does it make sense to do a or b? I'll give you an example of how I use this. So I'm going to use my fictional bakery company as an example.
04:56So you just type in something like console this to activate the LLM console skill right here. So I put in console this.
05:04Helena's bakery has a monthly cupcake subscription. Subscribers are loyal, but growth has flattened. Should we focus more on getting new subscribers or raising prices and adding perks to the existing ones?
05:17We also want to do this in the next quarter. Should we either launch a new referral program, or should we open a second location? Uh, debated.
05:26So as you can see, the LM console skill has ran, and we can see where the console clashed, and you can see all five of the console members debating each other.
05:38We can see the whole transcript here. Here's what the contrarian thinks. Here is what the first principle thinks about the two questions.
05:48Here's what the expansionist think. Here's what the outsider think. Here's what the executor thinks.
05:53And then we can see how they reviewed each other. And finally, here is the verdict.
05:59They all agree that we should monetize the loyal base first, not focus on acquisition.
06:06Here's the blind spots caught by the console, and finally, the recommendation. So if you're stuck with strategic business questions in your business, you need to install this skill set right now and help you get unstuck.
06:19I use this almost every single day. It's basically like having your own team of board of advisers for absolutely free. Now for the sake of time, I'm not going to read you out the entire MD file, but I will include it in the resource section below if you do want to read this entire MD file in more details.
06:38Okay. Moving on to the second set of skills that I want to introduce you to, and this is called last thirty days. And as you can see here, it has over 46,000 stars over on GitHub.
06:51So, of course, I just see what the hype is all about. And as I scroll down, you can see it's the number one repository of the day on GitHub. And, basically, what it does is that it will scan through Reddit, x, YouTube, TikTok, Instagram, Hacker News, and all of these sites that you see here for the latest sentiment about any product, any service, any topic that you're searching for.
07:15So for example, if I wanted to make a YouTube video on the latest Claude IPO and its update, I can just now put slash last thirty days Claude IPO, and it will go out and do the research for me. Now I'm getting ahead of myself a little bit. Let's first install this skill first.
07:32So, of course, we're gonna go up here and click on new session, and this is the bit of line that we want to put in. And, of course, you don't need to know how to code. I will give you this line, and it will be included in the installation documents in the description below.
07:46Okay. Once I have this skill installed into my GitHub account, I can see that I get a confirmation here from Claude. This means that this skill is now ready to use.
07:57So now all I need to do is type in the backslash line, and you can see that this skill is now loaded into my account. If you do do not see the last thirty day skill when you type and this turning blue, that means you have not loaded the skill set properly.
08:14If it's loaded properly, you will see this new skill set in your Claude account now. Now I can ask it any question that I want. For example, if I wanna do some research on Zapier, who is one of the sponsors of this channel, I can say something like this.
08:29What are people saying about Zapier in the last thirty days? Give me the honest sentiment, the biggest praises, the biggest complaints, and whether you'd recommend it right now. So I can just type in this this particular prompt and then wait a couple minutes, and this team of AI agents will go out to x, YouTube, uh, Reddit, etcetera, to compile all of the, uh, sentiments that users have been saying about this brand right here.
08:54You can use it for product research. If you're thinking about starting a new business, maybe you're using it to look at what people are saying about the pain points they have in the in the industry that you're trying to work in. I mean, there's so many use cases for this.
09:08It is absolutely a game changer for any type of research that you want to do. Alright. So now this took about five minutes or so, and now you can see we have a comprehensive report on what people like, don't like about Zapier, and where the points of improvements could be.
09:24As you can see from the last thirty day skill, it is pulling the exact URL where people are talking about your industry, your product, or the pain points that they have on the Internet, and it's giving you the exact links to those conversations and is summarizing everything for you, telling you exactly how you can differentiate yourself from your competitors or improve upon your own products by seeing what the pain points are and what people are complaining about.
09:55Some here, um, people are looking for a human approval layer for AI agents and Zapier workflows. Alright.
10:04Score of 41. So that is the highest point of improvement for Zapier here.
10:11Another and then the second one is another cash grab from Zapier. This is why I'm moving to make.com.
10:17Okay. Interesting. Third and then the third one here is anyone generating PDS from Zapier without it being a nightmare.
10:24This is another great idea for a demo that Zapier can produce is showing people how they can create PDFs and invoices, and this is exactly what people are asking for. I went ahead and I ran this skill again by doing the back slash last thirty days, so this is what calls on in this skill, and I put in Claude IPO. Okay?
10:43And I went over and it did all the research about Entropic's upcoming IPO for me, and it summarized all of the user sentiments here into this document. So what people are saying about it, what the speculations are, what's happening in the poly market, etcetera.
11:00Right? So it's giving me all these URLs. So if I wanted to create an upcoming YouTube video on the upcoming, uh, IPO of Quad, like, all the research is already summarized here for me.
11:11And if I want more details, I can click on all of these links here. Now just to show you the difference between using last thirty days and using, like, Claude normally, here is the result when I put in the exact same prompt into Claude without using the last thirty days skill.
11:30So you can see here is the Zapier sentiment. Okay? The overall vibe is still the default, but the love is more conditional than it used to be.
11:39And it shows us some statistics here from g two and Capterra, and it summarizes the biggest praises here, and then it summarizes the biggest complaints here, and then it gives us an overall recommendation.
11:54So you can see how this result right here is very generic com compared to the last thirty days document that we see here with all of the exact link and exactly what people are saying with their exact wording. So I did the same for the Claude IPO, and you can see when we're just using it without these skill sets attached to it, the result is very bland, very generic, and this is how you can elevate your own AI and not get it to sound like everyone else.
12:25Right? When we use these agents, when we use these skill sets, we get ourselves to the top AI users instead of getting the same responses as everyone else. Okay?
12:35Alright. Now I wanna move on to the third set of AI agents that I want to give to you. And this is actually created by Gary Tan, and he is the CEOs of Y Combinator, which is basically the world's biggest incubator for startups in general.
12:51Right? And he released his entire team of AI agents that you he uses to code and test software. And this is really a must have.
13:02So it has over 115,000 stars on GitHub. So that means this is super, super popular.
13:09Okay? And when we scroll down to the descriptions, it tells us the list of AI agents that this entire project will give you access to. So it will give you access to all of these AI agents here.
13:23It gives you examples of how they work, and you will get a CEO, founder, engineering manager, a senior designer, design partner, staff engineer, tester, design engineer, QA lead, etcetera, etcetera.
13:37So, basically, when you download the set of AI agents, you have an entire development team that can do all of these things for you for absolutely free. Again, here is how we can now download this entire team of AI agents into our cloud cohort. Now in order to install Gary Tan's entire team of development AI agents into your own Cloud account, just put in this piece of code right here.
14:04And, again, I will give you this exact code in the descriptions below so that you don't have to figure out what to put into your own cloud account here. And then once you copy and paste this, so press allow once it asks you if you can install it onto your local desktop. So then you just have to wait a few minutes for this entire repository to be cloned into your own cloud account.
14:28You may have to click on allow a couple of times for this entire process to finish. And once the g stack skill finished loading, and this is the prompt that I put in just to test it out.
14:40So, basically, I told it to build a landing page for a fictional AI startup called NexSys, which is an AI agent platform that runs task for businesses on autopilot. Then I gave it some additional instructions here. Another thing that you will see once you have this skill properly loaded is when you put the slash command, you will see additional AI agents loaded into your account.
15:05So you can see I have the diagram agent now, which matches. So another thing to note is that once you have all of these skills successfully loaded into your Cloud account, when you type in the slash command, you will now see these slash commands here.
15:22So if I do office hour, you can see now I have the office hour skill here, which does this right here. Okay?
15:31So I can put in plan, uh, plan CEO review.
15:35You can see now I have the skill, and I know it does in this. It will rethink the problem, find the 10 star product hiding inside the request, and the four notes, expansion, selective expansion, hold scope, and reduction.
15:49Okay? So I don't have to re prompt everything here. I just type in this skill right here after I load it in that entire GitHub repository, and now I can ask it any product CEO related questions.
16:03The same thing goes for design review, for development, for debugging, for testing, etcetera. By using these three repositories, you can see how now you can elevate yourself from just using AI like a chatbot to getting meaningful and more detailed responses from AI that no one else has.
16:23Okay. So it's been about three to four minutes, and our website here is done. You can see it has a nice fancy three d graphic here on the on the back of the website, and it moves as I'm moving my mouse as well.
16:37I mean, this used to be something that you pay a couple thousand dollars for an UX designer to design out for you and another couple thousand dollars that you pay a developer to develop for you, and it probably take a month or two to get something like this up, but now it can be done within minutes with one prompt. Isn't that mind blowing, guys?
16:57I mean, my mind was blown. Anyway, so I hope you can see now.
17:02This is how you can go from just typing in a prompt into ChatGPT and getting an output to actually getting much more meaningful results from your AI and elevate yourself from a basic AI user to the top 1%. Again, I will include all of the instructions on how to install all three of these repositories in the descriptions below.
17:23Thank you so much for joining me for this tutorial, and make sure that you sign up for my free AgenTeq AI master class if you want to dive deeper into how to build your own AI employees that will work inside of your businesses for your use cases. And if you got any value out of this tutorial, I would really appreciate it if you can like and subscribe, and I'll see you soon in my next tutorial.
17:47Bye for now.
The Hook
The bait, then the rug-pull.
A 400-page Stanford AI report boils down to two findings for Helena Liu: model quality has basically commoditized, and most businesses are still stuck using AI like a chatbot instead of a team. Her fix is three free, open-source GitHub skill sets that turn Claude into a debating council, a market researcher, and a full engineering staff.
Named ideas worth stealing.
03:20model
LLM Council (5 advisor personas)
- Contrarian
- First-principles thinker
- Expansionist
- Outsider
- Executor
Five distinct reasoning personas debate the same business question in parallel, then cross-review each other before producing a joint verdict and a list of blind spots.
Steal forAny solo founder decision that would benefit from simulated pushback before committing (pricing, channel choice, new offer)
15:00list
gstack engineering-team roster
- CEO
- Engineering manager
- Senior designer
- Design partner
- Staff engineer
- Tester
- Design engineer
- QA lead
A slash-command-accessible roster of specialized personas that collectively plan, design, build, and review a piece of software from one prompt.
Steal forPrototyping a landing page, small app, or feature without hiring a designer/developer first
How they asked for the click.
VERBAL ASK
17:20link
“make sure that you sign up for my free AgenTeq AI master class”
Soft CTA delivered verbally in the closing 30 seconds, reinforced with an on-screen URL lower-third (productcamps.com/free) and a like/subscribe ask immediately after.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
