Como Meus Agentes IA automatizam tudo na minha Empresa

Author: NoCode StartUp · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Como Meus Agentes IA automatizam tudo na minha Empresa · Channel: NoCode StartUp · 20:10 · Watch on YouTube

Topics: AI agents, Task and workflow automation, Data, databases and embeddings, AI content generation

Summary

The video demonstrates three AI agents used by the company: a personal assistant, a WhatsApp sales and support agent, and an operations agent that creates reports, insights, and content. It explains how the agents use tools such as Hermes, Discord, Notion, Composio, and Supabase, and emphasizes that useful agents need relevant company context and organized data. The creator then presents a five-layer methodology for building AI-enabled operations: context, data, intelligence, automation, and construction. The recommended approach is to build a foundation first, then add reporting and insights before moving on to automations and larger software projects.

Key points

The personal assistant connects to calendar, task management, email, meeting transcripts, and social media, and uses personal context to provide tailored help.
The WhatsApp agent handles customer service and sales, including audio, images, and files, and can pass leads to salespeople and update the CRM.
ISIS is a shared operations agent that supports employees with reports, insights, and content creation.
The company chose to develop one super-agent with multiple responsibilities rather than several specialist agents, to simplify testing and implementation.
Hermes provides the framework for agents; they also need an LLM as a “brain” and hosting to run.
The five layers of the proposed methodology are context, data, intelligence, automation, and construction.
Company context can include information about the business, products, audience, tone of voice, competitors, design, and internal processes.
Organized data is essential; the video suggests starting with a structured spreadsheet and moving toward a database as needed.
The suggested first use cases are read-focused reports and insights, followed by recurring automations and then complete applications or platforms.

Chapters

Start	Chapter	Summary
00:00	The company’s AI agents	The creator introduces the agents and explains that the video will show their roles, tools, and a methodology for building agents.
02:06	Personal assistant: tasks, tools, and context	The personal assistant uses an LLM, hosting, connected apps, and a maintained knowledge base to help with scheduling, meetings, and tasks.
05:43	Sales and operations agents	The video covers Sofia’s WhatsApp sales and support workflow, then introduces ISIS as a company-wide agent for operations and content.
08:16	ISIS in daily operations	ISIS supports content work and reporting, using connected tools and company context to share news, sales figures, and class evaluations.
12:21	The five-layer AI Business methodology	The creator lays out the five layers—context, data, intelligence, automation, and construction—and explains why the foundations matter.
13:22	Building company context and organizing data	The video recommends documenting company knowledge and structuring business data so agents can provide specific answers and access useful information.
17:03	From insights to automation and construction	With context and data in place, teams can add insights and reports, build recurring automations, and eventually create software or platforms.

Quotes

The idea of this video is to have a chat without the hype, okay? — 00:31

I created a knowledge base about myself, which I keep updated here for Lu to consume. — 05:11

Basically, for an agent to have a differential, it has to have context and access to data, okay? — 12:52

Providing context and providing data. — 17:34

The simplest automations to make are ones where it isn’t changing anything, it’s just grabbing data, seeing insights, and passing them on to us. — 18:05

Concepts and entities

Hermes, Lu, Sofia, ISIS, GPT-4o Mini, GPT-6, Hostinger, n8n, Pipedrive, Composio, Notion, Supabase, Obsidian, Discord

Full transcript (original language)

00:00 In this video, I’m going to open up and show all our AI agents actually running in our operation. I’ll show everything they do and how you can create your own AI agent to work today. To introduce you to each one, there’s Luc, my personal assistant; she organizes my daily tasks. Sofia is our sales representative on WhatsApp. And finally, we also have ISIS, who does a little bit of everything in the operation, from daily automated sales reports to content creation. These agents were created with the Hermes tool and are helping the company be much more efficient, automating several

00:31 processes in our day-to-day. The idea of this video is to have a chat without the hype, okay? No crazy promises, no talk about how it will run your company all by itself every morning. I’ll open the screen, show what each one does, what tools they use, how they access them, and the next automation steps I’m planning. And since I know the question that comes right after is: “Alright, Castelo? I already get it, but how do I do this in my company?” In the second part of this video, I’ll give you the step-by-step, a methodology of an AI structure for you to implement within your company and create your own AI agents from scratch.

01:03 This methodology has five layers, and the first steps are exactly what people skip, making their agents much more fragile. And in the end, not that good. So, if you have a company and want to move past just “using ChatGPT like Google for research” to actually having employees really working on your processes and projects, just stay with me until the end and you’ll understand the general overview and how to implement it in your day-to-day. Let’s go, alright? Let’s talk then about the AI agents we have here in the company running our operation. There’s Luc, who is my personal assistant, okay? She isn’t in the company, but I decided to bring her as

01:34 an example because she is quite useful for you to start creating your own agents. We also have Sofia, who handles all the customer service and sales here at our company. And we have ISIS, okay? Who is basically a super employee for the operation, automating several processes, alright? I’ll talk a little bit about each one so you can understand and get some ideas on how to implement them in your daily life, both personally and professionally, okay? First, we have Lui AI here. Lui is a personal assistant, okay? I use it here to help with my daily tasks. It’s

02:06 connected to my calendar and my task management apps. And, man, it’s really cool and ends up being quite useful, you know? In Lui’s case, I connected it to Discord, but you could connect it to WhatsApp or Telegram; it really depends on where you prefer to have your assistant chatting with you every day, okay? Of course, with Cloud, you could also build a personal assistant, but I really like Hermes because we can use it to connect our agent to any conversational tool, like WhatsApp, Telegram, and others. That’s why it’s very interesting to create

02:37 your agent using these tools. And for you who have already built with Hermes, we also need an LLM, which is basically the brain. We have Hermes, which is a framework, a harness with all the tools we need to deploy our agents. But we basically need a brain for this agent to work, okay? A brain or an engine, okay? Whatever you prefer to call it. Here I am using GPT-4o Mini, which ends up being very cost-effective. It handles simple activities, so it’s quite easy. And besides the brain, we also need hosting, right? So here, I’m using Hostinger for hosting

03:07 because it’s a very good value for the money. Everything is very easy there to set up your hosting and put your Hermes in place. So the setup here is simplified and the cost-benefit is also excellent. If you want hosting with a discount, we have a partnership with Hostinger. I’ll leave a link down below, alright? In summary, then, what does Lui do, okay? It already knows me, supports me daily, and queries tools for me. So, there are some tools I’m going to show you right now. It also summarizes my meetings. So, every time I join a meeting, a transcript runs, and it can summarize it and create

03:38 tasks in Notion if there are any. For me, she also creates my tasks, my to-dos in Notion, which is basically our task management app here. And she accesses some tools, right? For example , Google Calendar. So, she accesses my appointments. If I want to schedule a meeting with someone here, I can ask her to do it. My Gmail too, she can reply to emails. Notion, Fathom, which is basically this app to summarize meetings, and my social media, like Instagram, YouTube, everything here. She can have access so I can create content or get insights from some data,

04:08 okay? And basically what she knows, which is context, right, is super important. She knows my work, my projects, she knows personal things here, okay? About my family, she even knows my medical tests so she can help me be a healthier person and exercise more. So, she knows my life goals. Basically an assistant that knows a lot about me and helps me become a better person, okay? Not only professionally, but also personally, and helps me learn new skills. So, this here is Lu. And I created Lu with Hermes, okay? So I’m here in the Hermes app where we can

04:40 come here and create new bots, right? So you enter your bot’s name, what it does, right? What is its personality? So I did this process with Lu and, most importantly, I gave her context. So I talked a bit about my life and how things work, okay? And then afterwards, I connected her to Discord, okay? So on Discord, I can chat with her, right, directly through the mobile app or here , but I could also connect her to Slack , WhatsApp, or Telegram, okay? And basically here I can talk to her, okay? And exchange ideas about my tasks and my schedule. And she also summarizes my

05:11 meetings, okay? So here, look, she sent a meeting summary. So every time there’s a meeting, I can already get this summary and she creates the task in Notion about the meeting itself, okay? To keep everything organized, she even lists the participants. So it’s easy for us to do this integration. And a super important point here is that I created a knowledge base about myself, which I keep updated here for Lu to consume. So she understands a bit more about my life goals, she understands who I am, my health, my work, finances, so she has access to all this context which allows her to be a more personalized, more individualized

05:43 assistant. That’s why it’s really cool that we can create this whole structure here. She has access to the context and data to be able to provide me with more specific information. So she won’t respond in a generic way. And look, if you’re enjoying this video, comment below to help us by telling us about the agent you’re building or want to build, okay? I’ll read and reply to all the comments, giving you my tips. And speaking about our company, the first AI agent we have is Sofia. It’s an AI agent we’ve had for a long time, which basically handles customer service and sales via WhatsApp. So, at certain

06:13 times of the year, during promotional campaigns, a lot of people contact us, and she’s also running here 24 hours a day, 7 days a week, so if someone sends a message at dawn, she’ll be able to assist and make a sale, okay? This agent has been around for a while, she was created with, so there’s a whole structured flow for when she talks on WhatsApp, and she’s hosted here, in this case, n8n, over at HeyGen, okay? So, basically what does she do? She handles contacts on WhatsApp, understands audio, images, files, passes leads to the salesperson, and is

06:43 integrated with our CRM, our Pipedrive. Basically, she accesses several tools, right? WhatsApp, integrated with n8n, Pipedrive, Chatwoot, which is our conversational tool, and our entire database to register the leads and the sales, okay? And she also has access to all our company context, our products, and offers. This type of agent focused on sales is one of the most important, but later there will be a video here just on how it works and how you could make one, okay? I think my goal here is

07:14 more to talk about the AI employee agents in operations, which is ISIS, okay? I want to move on to ISIS now to talk a little bit about her. And now talking about ISIS, who is our super agent here that is in several areas of our company’s operation, okay? What does she do? She provides support to the team, creates reports, automatic insights, and she chats on Discord. So, the positive point here is that it’s not just me talking to her; any employee in the company can chat with her to get insights and create content, for example, okay? She was created using Hermes. The brain she is

07:45 currently using is GPT Sol 6, okay? Which is a very good LLM, a very good model, and it’s hosted on a VPS at Hostinger, okay? If I’m not mistaken, it’s the KVM2 or four, okay? This agent is one that I recommend you implement in your company, passing on processes and automations for it to create, okay? Basically, what does ISIS do? I’ve separated it into a few areas here; there are other automations and processes she’s doing, but she’s improving a little bit every day. So, regarding some processes that she already handles here, there are various

08:16 process areas. We are adding more and more automations and skills to her here . We decided to go down the path of building one super-agent with various automations and responsibilities rather than making several specialist agents, okay? I think it’s better for you to do the same for your company, as it’s easier to test quickly and simplifies the implementation and operation of your agent, okay? In content, she already does a daily news radar. So, she looks at the main news on X/Twitter and passes it on to us. She already

08:46 creates full YouTube scripts and tags our scripts there, creating them in Notion, which is our task management platform. She also makes a profile report for new students, so every student can fill out a form, and we can see their profile and who our persona is. So these are just some of the things she’s already doing on a day-to-day basis, okay? Tools she accesses. Most interesting part. So, there are many tools she accesses here: Meta Ads, Google Analytics, Instagram, YouTube, X, and Notion, okay. So, we can do all of this through Composio,

09:17 okay? So, Composio is a really good tool where you connect all your apps and social networks, and basically, my Hermes just needs to connect to Composio, okay? So, I recommend taking a look at Composio, as it turns out to be very interesting. And the context part here is also super important; she knows everything about the company. So it has context about our company, our persona, our product, our tone of voice , our way of creating content, and data , right? Sales data, social media data, man, it has access to all of that. So, when you give context and data to your

09:48 agents, they end up becoming really powerful. And folks, an important message: if you want to learn from scratch how to create these types of agents with Hermes and other tools, our training programs are open, with updated lessons for you to create your first AI employees and the whole structure to gain performance in your company. So, here is an example of the AI news radar. Basically, every day it looks at the top news. So I took the 24th here, it went through the main news here, and I can come here and ask it to create some content, right? So I

10:19 come down here, I asked it to create some content. I said, “Hey, create some content about GPT-5.5.” It went there, it worked on it, it created a new thread, and it created this content. If I click here, it will open Notion, okay ? This content of mine, all structured as a script. And not only that, it can also create Reel scripts here, okay? So here, look, it already created a more compact Reel script so I can just, sometimes, put it on a teleprompter or manage to review it, have the bullet points here to make this content for Instagram, and it can already include

10:50 editing tips, besides creating the Notion link for it. And here, it also generates a sales report for us every day, and it sends this report, an image for the team to track how sales are doing, how things are going, and we can stay in close contact without needing to keep opening the dashboard. All of this is possible because we have a well-structured database and we also have our business rules well-structured , which we invested time to be able to do. It also sends us weekly reports on class evaluations here. So we can see here which classes we need to update or

11:20 improve. So we look at how people are rating them, there are the scores, and whenever it falls below eight, we have to take action to improve and update those classes, right? So we can have ISIS tracking this for us every day. Not to mention that anyone can talk to her here, ask for content creation, ask for news, ask for sales data, data about our leads. So every company that is inside Discord can talk to her to extract this data and really be this employee. So, it’s still in the early

11:50 stages, there are many processes here that we can automate to keep it running and really help not only to gain insights, but also to handle this whole part of automations. Guys, a super quick message. From October 19th to the 22nd, we are going to do an agent immersion. Basically, it will be four live and free classes for you to create your first agents in your company. We will make agents from the most basic to more complete ones. So, if you want to create them in practice, just sign up for this class, okay? I will leave the registration link here on YouTube, in the comments, there in the video

12:21 description. Just sign up to be able to participate. Let’s go. And now that you understand a little more about agents, let’s talk about how you can create your first agents within your company, okay? For this, I have separated our AI Business methodology here for you to make your company agentic. Basically, there are five very important layers there, okay? And the initial layers will be the base and then we can create our agents, our applications and platforms, okay? They are, first, context, then there will be the data

12:52 part, then intelligence, automation, and construction, okay? So this will be the foundation that it is interesting for you not to skip, okay? If you start creating your agents, they will be a little more fragile and a bit without that power of the agents. Basically, for an agent to have a differential, it has to have context and access to data, okay? Let’s go one by one for you to understand. First here, context. So, basically when you are going to create agent projects, you have to provide context, okay? For that agent of yours, because it will be able to have a much

13:22 more specific answer and an answer that won’t be generic. So here, what will you have to put? You have to put information about your company, you have to put your product, your persona, your tone of voice. So, the more context you structure here about your company, the better it will be. And for you to organize this, you can organize it, for example, in Drive, you can put contents there, documents in a simpler way. Here is a very simple way, which is Notion. So, basically, here we have everything in Notion organized with text about our company, which is simple and easy for us to consume. You can

13:53 also organize it in Obsidian. So you go into Obsidian and you can create your company’s second brain. Basically, you keep providing more context so that later your own agent can access this Obsidian, okay? Another way you can do this is by basically creating the whole folder structure and leaving it on GitHub. For those who don’t know what GitHub is, it’s a place where we have a cloud library, and then everyone in our company and our agents could access this context, okay? I recommend you start in a simpler way. Go with Notion

14:24 itself, create a document there about who you are, what the company does, and the tone of voice so you can structure it. So here at our startup, for example , we have a whole structure in Notion that is super simple, okay? The ideal is really for you to eventually reach an Obsidian or GitHub level. And here we have context about the company and the product. Then we have context about who our persona is, who our B2B persona is, and our tone of voice. Tone of voice here based on the way I speak. Here we have the analysis of the most

14:54 interesting players in the market, who are our competitors. We also have the design system here on how to create images, documents, and what the visual identity is. We have here context on our internal infrastructure, how we organize ourselves. So, this way, when the AI can access this, it already knows everything about the company and can give much more specific answers tailored to your reality. Got it? So, the second part here, the second layer is data, which is basically having a well-organized data structure so your AI can consume it, okay? So this is a

15:26 part that, for me, is the hardest for companies: having this data organized. We spent quite some time here organizing our entire database. In our case, we use Supabase, which has all our data modeling, everything communicates through there, but you can also have access to other data, right, from CRMs, social media, forms, calendars, and other tools. Let me show you how we do this here. So here we use Supabase, which is a backend framework, database, and it’s within the modeling and authentication part, and there’s also the media storage section. So we

15:58 have our entire database structure within Supabase, okay? So all the data connects there. This is the most important thing for you to have for your company, right? If you are in the initial stages, make a well-structured spreadsheet, something organized, but later it’s good to have a structured database. And as I had already mentioned here, we have Composio, which can connect with many, many tools. So, we use Composio here to connect with tools, such as Notion, which is where we organize our entire task management process. There’s my Twitter there to

16:31 fetch the latest news, there’s data about our YouTube, there’s Meta Ads here about our paid traffic campaigns, Google Analytics, and there’s Instagram here, personal profile and company profile. There’s Metabase, which is basically a structure for organizing charts, okay? And data, dashboards. And there’s Typeform, which is basically for forms, okay? So I can get everything in one place, I connect it here, and my AI agent can connect to this place and have access to everything about the company, right? So basically, this is all read-only, okay? Most of it is reading, but the AI can

17:03 create data, though it’s mostly reading , but it could perform automation operations, creating Instagram posts, or managing our Meta Ads. So, man, you can really do a lot of things. But the most important thing here is that I’m providing context and data. From there, our agent will already be able to have access to the company’s context and data. So, it will be much more powerful than what you can build. This way, we can move to the next layer, which is the layer of actual execution, the creation of the agent, right? We have our turning point here, see, it’s this connection, right? Because with context

17:34 and organized data, now you can really create whatever agent you want, whatever project you want. That’s why this part here is very important and many people end up skipping it, okay? Providing context and providing data. And then we start with intelligence. Basically, it’s when the agent or the automations can bring insights, intelligence, reports, which is what’s easiest to do. So, instead of starting by creating automations that, I don’t know, create a carousel for you, focus on insights and reports, okay? Okay? So take what you already have. That’s it. If you have sales reports, if you have

18:05 lead reports, you can already get that from the data. Ask your agent to bring that to your WhatsApp, your Discord, or your Telegram every single day. So here , for example, every day ISIS sends a sales report and an analysis of the classes, plus student reports. So, every day we get these insights. So, the simplest automations to make are ones where it isn’t changing anything, it’s just grabbing data, seeing insights, and passing them on to us. The coolest thing here about Hermes, right, is that we can exchange ideas. So, whenever I want a specific analysis of a specific piece of data, I can ask

18:36 it there and it will bring me information about those data points, okay? So, the first thing to start building into your agents after you create them is to add intelligence. And that’s where automations come in, okay? So here we can create automations to run every day or automations to create something for us, right? In our case here, we have report automations, yes, so it runs every day, every week, but I also have automations where it creates, for example, YouTube scripts following our standard. Every time we ask, it already creates the YouTube script. I’m

19:07 finishing up here so it can also later upload this video, create thumbnails, create titles, doing the complete process of creating YouTube content here. So that is the automation part. And finally, we have the construction stage. So here is where your agent or your team is already able to create software, applications, platforms, interfaces. Here in this case we are using Discord a lot for conversation, but you can already create platforms and automation systems like this. In this way, with the five layers, right,

19:39 we have data in context, we can create intelligence for our operation, then automation, and finally the construction of complete projects here, whatever they may be. So, those would be the five layers of the OS. Comment down below which part of this layer, of this process, you are in. Guys, if you liked it, comment down below as it helps us a lot, every comment helps us understand if you are enjoying it or not so we can bring more content like this, alright? We have lots of free videos here teaching you how to create your first agents on the channel, okay? So just leave a comment here and then

20:10 take a look at our other videos. See you in the next one.