# How to have Hermes run your business

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=T4gibRVtt-I
Duration: 21 min
Language: English
Words: 3785
Transcript page: https://viewrankai.com/tools/youtube-transcript/T4gibRVtt-I

---

[0:00] I run multiple businesses with Hermes agent. Look, it's creating even video ads for me. Most people don't know the power that it has. It's doing customer support for me right here. Let me show you a few comments that everyone needs to know. Andrew, let's do a screen share. I will show everyone how to do this. Presented by Zapier, the AI automation company. What are we looking

[0:21] at here? What is this? This is my Stripe revenue dashboard from the businesses that I built with my AI agents. Okay. And so we're looking at one business that's got 56,000 in gross volume, the other 44, and then another 18, and it's AI agents that you're using. Is it mostly Hermes? It was Open Clo at the beginning, but we

[0:41] switched most of it to Hermes, yes. All right, let's take a look at how it's bringing business for you. How about the ads? So this is what? Telegram that you're using to interact. Did it create these video ads for you? Yes, it created all those ads for me. I just had to give a prompt that was two

[0:59] or three line. That's it. Can I see the way that you interact with it to get it to create video ads like this and how you know that the ads are going to work? So the way it works is my AI agent, Hermes, will give me daily reports on which are the best ads that

[1:17] are running. So it keeps monitoring what is going on. And for example, here you see that yesterday was 42 new ads and 33 disappeared. So we know that those ads are not that good. But what is this? way, this is ads This is ads for others or is it ads for your business? This is ads for others. This is people advertising Open Clo on Meta ads. I see, because one of your business is Hermes flat One of your business is Fly Hermes, which helps people install Hermes, so you're watching how other businesses that make it easier to install agents are doing. And so what's it using to

[1:52] figure out which ads are running and which ads are not? What access does it need to do that? Well, it access the Facebook Meta ad library, so it's going to go scrape it every day to see which one are ranking. Okay. All right. And so, it's check it's checking to see which one's ranking and it's learning from it and then it

[2:11] creates video ads for you? Exactly. So, because it checks all those ads, gather the text, can download the images, and even sometimes, if they have video, watch the video and start to understand the video. That's something that I'm implementing right now. Then it know what's working, what are the angles that other people are using, and we can reproduce the same ones with videos that

[2:36] we generate. What's the stack that you use or what access do you need to give it in order to have it study, rank, uh watch the videos even, and then create videos for you? You need something that's going to scrape Facebook ads, uh Meta library, or that you can have your machine directly doing

[2:53] it. And it's going to open the page. I use a personally directly with Apify to scrape. It's easier. I use Apify for all kinds of scraping, right? So, you basically are paying them for API calls, they're scraping, and they're charging you every time you use it. Great. That makes a lot more sense than doing it on your own computer because it's wasting tokens and then it's not that efficient. I mean, browsing with agent is good. It works most of the time, but it's not as reliable as API calls. Takes forever, takes a lot of tokens, not as reliable. Okay, so it's uh Apify for scraping. What else are you

[3:31] using to to study these videos and create your own? So, to study those videos, like first of all, like for the one the research that we're doing here, like the open clue ads, not many people are running videos. It's actually mostly static images, and I think it's the best way to start actually for ads uh on Meta. It's easy. You can throw like 12

[3:54] directly on the same day and test those. But otherwise, there is Meta actually that released a model called Tribe V2, if I'm not wrong. And this AI model can analyze a video and see what happened on the brain when people watch that video, which is really good. Otherwise, if you don't have access to that, which can be possible because I think it doesn't run on every machine. You can upload your video into Gemini and ask it to describe what's happening each scenes and to define a prompt that will help you to recreate this video if you were asking

[4:34] an AI. Okay. And we'll of course have list list of all these tools that you're referring to in the description. Okay. Um scroll down. Let's see the video that I created for you. And just walk me through what it's doing here. Okay. So, the three video we have, as you can see, AI is not always perfect.

[4:54] No, yeah. He has the new MacBook maybe. He's too in advance. Yeah, the the screen is on the wrong side of the lid. Uh-huh. [laughter] Yeah. We have some sound before, but basically the angle that you wanted to play on is that everyone is struggling to install AI agent or they're struggling with too much work. And then once they have Fly Hermes, wow, life is good and then they change

[5:18] the thing. Most of the AI videos that we see online are actually heavily edited. So, what you would want to have is like multiple scenes and then you plug those scenes together cuz video models, like I think C 10 2, the max output is around 15 to 30 seconds max. And still you're not going to have like the greatest thing. Like you want multiple shots and then combining them together. But once you get few reactions, you could actually split test different videos. So, it's not creating the full video, it's creating snippets. You're personally editing it, and then you're running the

[5:54] ads. Well, those ones I could try to run them like that, but I'm not sure if they would work. Like 8 seconds is okay, but you I think you would rather want 15 seconds, and I have like different scenes. So, I would ask him probably like someone who is really struggling, and then another one that would be like now he's not struggling anymore. And then to edit the video, I will use AI. I actually have an example of a video that I edited with AI on my Twitter. So, this one for example OpenClaw just shipped its biggest update in months. Two things matter. It now joins your Google Meet

[6:31] calls like are you using when when uh Hermes puts together all those reaction videos, what are you using to edit them together? Directly, Hermes can attach them together, and he can also add text on top of it. So, you can add text on to do it, and it'll piece it together, and and add Can you Can you tell it to do it now? I can try to do it now. Let's

[6:51] say, okay.

[7:00] Okay. And normally I would give also an example if I want to put some text to have like the kind of a social media style where you have a rectangle that is white at the top with some text. Right. You give it a little more guidance. What does it need in order to do these this editing? Are you using any any

[7:20] third-party tools? No, this going to be uh probably FFmpeg, you see. It's calling it here, which is uh the easiest tool for uh managing video formats, and it's using a Python. And then in now it's trying to plug them together. As you can see, it's used the seed ends to generate the UGC here.

[7:42] Okay. All right. All right. It's done. [laughter] Uh obviously, you would have told it not to put that first clip in there. Um and you give it a little bit more guidance. All right, I see how it's working. I see how it's helping you with the ads. Um let's go on to some of the slash commands. You were actually going to show someone today about the power of slash commands, how people aren't using

[8:05] it enough. And so, you created a slide. Can I see how you created the slide? Sure. Okay, let me see. Um it's uh you said to your agent, "I want to create a slide presentation to show how we use Hermes agent together. What are the best options? Give me three uh because I want you to create the presentation." Got it. All right. And so, it started talking with you. You then got the slides at the end. Can I

[8:31] see what it looks like? Yeah. Okay. All right. So, why don't we go through some of them cuz these are the things that are important for people to know. What is new? New is when you on the same uh chat window, but you want to go on a totally new topic or totally new thing. So, for example, if you have the subtopics, like I do on Telegram, you're not going to have to use so many times new unless like you have a channel that has like tons of

[8:58] different uh topics in around it. Mhm. Compress it's if your chat is too long, but Hermes does it itself most of the time, so it's not that useful. Q is super important because when you start to work on a task and sometime it may takes 15 minutes, and that's quite frustrating because then you have another idea, you can just do {slash} Q and then say the next task that you

[9:24] wanted to do. Mhm. Steer is also really important cuz as you can see, I'm pretty sure sometime I do a mistake or there's a word that when I did the voice to text, it didn't get it. Steer is going to be useful. I do {slash} tier and it's going to adjust the direction. But it could be also adding information because instead of stopping the current task, like if you say to Hermes like, "Hey, what's the weather like today?" It's going to start to search and then if you send another

[9:53] message, it's going to stop the task. So, you don't want that. But tier is going to be okay. And the next tool that he uses going to be, "Let me get that information and add it to what I'm working on." And if you see it, for example, it starts to not use a skill or not use a tool, which can still happen

[10:09] because AI agents are not like perfect. Mhm. It's going to adapt and you're going to have a way better output. Okay, so you're not interrupting the action, you're just kind of guiding it to do whatever it's doing a little bit better. Background, what's this one? This is a little bit like sub agent, but for like small tasks. So, if it's working on a task like, let's say those video they're going to take a 20 minutes to generate, for example, I can do {slash} background and say like, "Give me how much we spend in ads in the last 10 days." And then he's going to do that little task on the side and that's

[10:44] pretty helpful, too. I saw you using the goal command earlier when you were scrolling through your chats. What do you use it for? Yeah. Basically, goal is like, "Hey, this is really important for me. I allow you to spend a lot more time on it." And by that, it means you can spend a lot more

[11:03] tokens and you can have multiple loops. So, the budget of loops go from like 20 to 150, I think, on Hermes most of the time. And that means it's going to allow itself to go through like prompting, reprompting, analyzing, and going again, again, again, again. Mhm. It's just going to be a lot more token intensive, but also the result

[11:25] will be way better. Okay. Next. Stop is super important. Stop is just chill the task that you're doing right now. If you see that it's doing something dangerous or whatever, you just slash stop. Okay? And with Fly Hermes, do people switch models or does it come with just one model? It comes with multiple models. There's also smart routing if you want, so like this it will direct it to the best model you want, but you can switch model easily here. You have all of them and you can switch on Telegram as well. And this is Fly Hermes and in addition to using Telegram people can use the chat

[12:05] within that comes within it and seems like a lot of people prefer to use the browser chat. I mean, the browser chat has the advantage of having better displaying than the Telegram. For example, if you have a some kind of dashboard or tables or things like that, it's not always so well displayed on Telegram, but here

[12:24] it's going to be a way easier to read. Okay, let's see how you're using Hermes to build. Are you going into Hermes to ask it to build sites for you instead of Cloud Code? It depends the the size, I mean, of the website. Like if it's anything like content-wise or whatsoever, I can do it directly in Hermes, it's fine. But if we're going to build something that is let's say that require a lot more work, then I'm probably going to do it

[12:54] directly in a Cloud Code or Codex. Okay. So, here we have wait. I saw that. Was that a hilarious Tinder something? Yes, you're going to see. So, like this one, this one it did a full research about me. Okay. there was like Antoine from Handsome Finance. Let me read that. Let me read that

[13:15] prompt for a second. Uh go up a little. You said, "Deep research. Find everything you can about about me, YouTube channel, who I am, etc. And then search for and install any research skills you need. You have two prompts left after this one. Tell me about why you why you're telling me you have two prompts left after this one. Uh it was because I was doing a challenge at that time where I was like comparing okay like how far can I go with just few prompts? Let's say for building a website or something like

[13:44] that. Okay. Oh, here was a deep research. Like they had the three two prompt max for the deep research. I mean if it was wrong after, it was off. So that I told Okay. I told it like okay, we're going to do this this game. Okay. And right. And I see that it got a whole lot of data on you. It actually got data that I wish I had before you and I started. But this

[14:03] is great. Okay. What's next on this? Yeah. So then once I had that, I say okay like uh it will build me a a Tinder website that shows like who I am and everything around it. And it did. In one prompt. So let me show you. I wanted to like have a like unfortunately the text is over. Like we could arrange that by correcting it. But otherwise it was pretty good. Like you

[14:33] had some animation. face. Yeah. But it had like you have the Twitter links. You have the YouTube. Like it kind of like I it I did a pretty first good version of the website I think. What was this for? This is like a replacement for your personal website? I mean Yeah. No, it's not going to be. But like yeah, I wanted to see that okay like a simple um yeah, display website but still with an original thing. Like I wanted to have a something different. And then it went fast to build it like an easy. It's on GitHub and Vercel. And that's it. I didn't have to do much with it. It

[15:13] created that and then after like I bounced back and forth a little bit to modify some stuff. And that was it. I get how this works. How did it know that you wanted to see it on Vercel and that you wanted it to be on GitHub, too? I told it to to do it and I give it a GitHub key and Vercel key. But, that's not here. That's in one of the skills

[15:32] that you gave it. I know, it's actually it's in the prompt just above, but I didn't show it. Okay, I didn't see that. All right, great. Let's look at customer support. So, for customer support Hermes, how many disputes do we have here? And it said 18 active disputes total. Six need response. 12 are under review, meaning like Stripe is checking them out. One is

[15:55] one, and 22 lost historically. So, the urgent number is number six. Yeah. So, actually it's since before we were doing the disputes manually, and now we say to Hermes to I mean, this one is run by OpenClaw to run the dispute skill to to be able to answer them. And then, since then it started to win one. Let's see if

[16:19] it's going to win the other ones. So, wait. Is this an OpenClaw chat that you're having right here? Yeah. Okay. Can I see how our Hermes is doing customer support? Is there a difference? Uh no, it's going to be pretty much the same. Like, it's because how they handle the mailbox and everything is going to be pretty much

[16:39] the same. There's no no big difference. Okay. What are they disputing? Oh, it's just like we had like on on this one on the OpenClaw we get like 500 people and more. So, it's like it can happen that the people like they say, "Oh, I disubscribe." And then, they try to get a refund from it. What's that website? What's that

[17:02] business? Uh this is Heroku. We do OpenClaw hosting. Okay. So, you're running both OpenClaw and Hermes. Which do you prefer and why? What's the difference in your experience? Well, me I don't I don't really want to to pick a side, right? But I have to be honest and then like recently, the one I use the most is

[17:26] Hermes. And I don't know, it's just overall, I just got better output. And I came from different things like Open Claw often broke after updates. A lot of people complained about that. Happened to me once or twice as well. And that's a bit annoying. And then after I found the Hermes way better at using skills and tools on repeat. Um which is why like uh with Open Claw I was often like repeating now, you should use that. You have to use this skill to be sure to use that and things like

[17:59] that. The way like it's self-evolving for Hermes I just find it more efficient. I have heard a lot about how it knows what to do better. It doesn't break as often and it learns from its experience. Yeah, it's I mean already when you chat with it, when you see like how it's working and how it's building the skills automatically, it's just a night and day. Like the thing is, for example, Open Claw at the beginning like now maybe with some updates they're changing that, but it didn't automatically build the skills unless you say like build a skill around

[18:35] this now. I see. And then it would not always use it. That's why people had actually at the right beginning of Open Claw and I had those where you had the prompt job that will analyze all the chat and build some things around to make your AI agent self-evolving. It was not built to be self-evolving at the beginning. That's what people need to

[18:58] understand. Can I see some of the skills that it might have built for itself in the skills tab? What I like about Fly Hermes is that you have all the skills easily accessible on the lower left. So, here you see, boom. For example, same improvement review skill Fly AI video generator created. So, here it created, for example, um a skill cuz I'm running different tests on some agent, and this one I was like, "Okay, I want to make

[19:23] some UGC ads for matcha brand." Okay. And it did those ones. Which some of them are like really good. Like this one is It is a few movements that are wrong, but like otherwise it's pretty impressive. And we have a those one. It's like a It is a few own improved the skill or created the

[19:43] skill? Can I see that? Yes. So, here like after it was done with creating the ads 3 AI generated, it says self-improvement review, skill file over file video generated created. So, it's using file AI here to be able to generate those. Thanks, Antoine. And for people who want to, they can just go to fly.hermes.ai to sign up and try it themselves. I

[20:09] think you've got a 7-day free trial. Exactly. Cool. All right. And we'll have a list of the resources below to anyone who wants them. Thanks, everyone. You noticed Antoine uses both OpenClaw and Hermes. Many people will run many instances of each of these agents, and it becomes hard to manage. The next video you should watch is how one woman is managing multiple agents using an open-source tool called Paperclip. You'll watch me install it, see how easy it is to set it up, and you'll see how easy it is to both use on your own with multiple agents. And if you want to bring your team into it, how

[20:41] to make it easy for them to interact with them. So, we got a link for you to watch that video next. See you in that next one.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt