YouTube Instagram TikTok How it works FAQ

The fastest way to turn talking videos into clean, searchable text. Free, private and no account required.

About

Built by Rishav Raj with one goal: free transcripts for everyone.

rishav@yttotranscript.com

Transcribe

YouTube to text
YouTube with timestamps
Instagram Reels
TikTok videos

Learn

How it works
FAQ
YouTube transcripts guide

More

Chrome extension
Contact
Privacy policy
Terms of service

© 2026 YTtoTranscript. Made for creators & researchers.

Not affiliated with YouTube, Instagram or TikTok. Respect creators’ copyright.

In nur 12 Minuten KI-Agenten verstehen! (für Anfänger erklärt) — Transcript · YTtoTranscript

Home/Transcripts/YouTube

YouTube Transcript

In nur 12 Minuten KI-Agenten verstehen! (für Anfänger erklärt)

Everlast AI · 2,328 words · 12 min read · EN

Watch on YouTube

Share X LinkedIn WhatsApp

Below is the complete, readable transcript of In nur 12 Minuten KI-Agenten verstehen! (für Anfänger erklärt) by Everlast AI on YouTube. Read the full text, copy any part you need, or generate a transcript for any video with our free tool.

I've learned everything about AI agents for you , so you don't need to do it anymore. This video is your short version, to save you weeks of research and expensive courses. There is n't really one course that covers everything nicely. And that's why I myself have spent the last 12 months watching hundreds of hours of videos,

reading a ton of papers, and, together with my 30-person team, building hundreds, if not thousands, of AI agents. In this video, you will therefore learn not only theoretically, but also practically, why AI agents are the topic that virtually every billionaire in the world is talking about in Silicon Valley right now, and why they are also

one of the most important topics you need to deal with in your career . And since there's so much confusing hype surrounding the topic, let 's first clarify what an AI agent actually is. Then I'll give you a crash course on the different types of AI agents you need to know. And to illustrate all of this practically

, I'll show you how to complete your first AI agent workflow today without writing a single line of code. You'll be shocked at how powerful and yet easy to create these AI agent workflows are. And finally, for everyone who wants to gain a foothold in the AI agent field or install it in their own

company, I'll give you a brilliant eye-opener that, when I first heard it myself, made me think, wow, that's exactly how it's going to be. Everyone really needs to know this. So stay tuned, and let's first clarify what an AI agent actually is. And believe it or not, even for me, who has been

working intensively with AI for years and has already helped hundreds of companies implement AI, the most difficult thing for me at the beginning was understanding what this term actually means. And that's because it's such a new field, and even the leading companies are only just now desperately trying to figure out what AI

Transcribe another video

Paste any YouTube, Instagram or TikTok link to get a free transcript.

Free · No sign-up · Unlimited

Add to Chrome — transcribe in one click

agents are and what they can actually do. So if you 're still confused, it's definitely not your fault. The easiest way to explain what an AI agent is is to clarify what an AI agent is not. It 's definitely not an AI agent if you simply give ChatGBT a task. For example, if you tell ChatGBT, "

Please write a complete business plan for my company XY," that 's not an AI agent, it's simply called prompting. That's like telling an intern to quickly finish a presentation. And the result is, yes, you will get something delivered, but it will be superficial , bland, and definitely not what you actually need. An agent-based

workflow, on the other hand, works completely differently. Instead of a straight line, i.e., request and response, you have a self- repeating, iterative process. This means that the agent would first create an outline for the business plan , then consider whether they need to do research on your company , then write the first draft

, see which parts still need to be revised, do more research, revise the draft, and so on, until the perfect business plan is ready. It is important to understand that there are three stages involved. Level 1 workflows are non- agentic. You ask a question, you get an answer, and that's it. As I

said, this is normal prompting or a very classic automated workflow. Level 2 consists of agentic workflows. This means the AI goes through several thought loops, conducts its own research, checks its results and improves them again based on the guidelines you, as a human, have set for it. Level 3 are fully autonomous AI agents. In this process,

the AI decides completely independently which tools it needs and how best to proceed. The volumes are therefore primarily levels 2 and 3. The leading figures at Anthropic, one of the biggest ChatGBD competitors, differentiate these two categories as follows in their paper Building Effective Agents. Workflows are clearly defined processes. You specify exactly what should happen and when.

For example, a form is filled out, the data is checked, and then a language generator ( LM), such as ChatGBT, writes a text. and an email is sent, which is then perhaps checked once or twice more by the AI . Agentic workflows are therefore relatively stable and reliable, and they already work today

. Fully autonomous agents are a completely different caliber. Here, the language model, i.e., the AI, takes complete control. It thinks for itself, deciding which tools to use, in what order, and when to perform which steps. The agent plans, reflects, and changes his strategy, almost like a real employee. To better understand these two

concepts, workflows and agents , imagine the following situation. Let's say you want someone to hang a picture on the wall for you. Now you give a command. Here you have a picture, a nail and a hammer. First, you have to hammer the nail into the wall. To do this, take the nail and hold it

against the wall with one hand. Now you hit this nail with the hammer until it is firmly fixed in the wall . Once that's done, take the picture and hang it straight on this nail. Keep doing it until it works . This is a workflow, meaning that the human or the AI performs the task,

but as a human, you explain all the tools and give feedback at the end on whether the task was performed correctly . And the alternative is, you say, here is a picture and I want it to hang on the wall, and here are some more tools. Check the result yourself until the picture is hanging perfectly straight. This is a

fully autonomous agent, and as you can see, the difference is that you as a human no longer have any influence on the individual steps of the process, and the AI even controls itself until the goal is achieved. So we are currently at level 2, so to speak. Fully autonomous AI agents do not really exist in practice yet, or they

are very unreliable. But that's only a matter of time, for example through technologies like MCP. If you'd be interested in a video like this one about MCP , please let me know in the comments. But now it gets really interesting, because there are four recognized types of AI agents developed by Andrew Ing, one of the world's leading AI researchers

. The first type is called Reflection. Self-reflection in German. This means you are not only asking the AI to perform a task , but also to critically review its own results. Example: "Write me a code for my website." The AI writes the code. Then you say: "Now check your own code for errors, efficiency and

style, and improve it." The AI looks at its own code, finds errors in line 5, improves them and optimizes the whole thing. This is significantly better than simple prompting or classic workflows. The second type is Tool Use . Here you give the AI access to various tools that it can use independently. For example,

you ask, what is the best YouTube channel on the topic of AI in a business context? A normal AI would tell you something based on its training data. However, an AI with Tudius can search the internet, find current ratings and statistics, summarize them, and then of course return Everlast AI as an answer.

Or it can execute code to perform complex calculations, e.g., game-related channel statistics and user feedback. The third type of construction is called Planning and Reasoning, i.e., planning and logical thinking. Here, the AI is given a complex task and plans for itself which steps are necessary and which tools it needs for this. This

example from Andrew Ing illustrates this very well. He says he will show me a picture of a girl reading a book in the same pose as the boy in this example picture and describe the result to me using your voice. The AI then decides, "I first need to analyze an image, recognize the pose,

then generate a new image, convert it into text, and then into speech." Everything is automatic, without you having to take any steps beforehand. Exactly, that 's planning and logic. And the fourth building is Multient Systems. These are several specialized AI agents working together. Just like in a company, where you have different specialists. Instead of

giving everything to an AI, you have, for example, one agent for research, one for text creation, and one for quality control. They then work together like a real team. Studies already show that the result is significantly better than if an AI has to do everything alone. In practice, as mentioned, these systems are

not yet mature enough to be used in a scalable business environment . Here is another example that illustrates very well how such an agentic workflow works. This is a Vision AI Agent where you can upload video sequences to have them analyzed by the AI to identify objects. I've uploaded the 2014 World Cup final here and I'm telling

the agent to find the sequence where the goal was scored and then tell me the player's name. The agent now begins writing his own code. It then uses its vision capabilities to extract individual sequences from the video, analyze the frames, and verify what a player is, what a ball looks like , what a goal looks like, and in which

minute—that is, in which sequence— the goal was scored. The agent responds accordingly: "I can confirm that Mario Götze's historic goal was scored in the 113th minute ." The code successfully extracted the sequence and analyzed the scene. It then delivers the JPEG files with the individual sequences and marks the corresponding objects, i.e., the

players. And it's truly more impressive than you might think that all of this happened without human intervention. This demo, by the way, is also from Andrew Ing, and you can try it out for free to get a first impression of AI agents or to have your old videos analyzed. I've provided the full link

below this video . In practice, there are already many other examples of how AI agents are used. For example, research agents that can conduct in-depth research in internet sources, but also in your own data. AI text agents, who then elaborate on and visualize these topics. Coding agents who create software. Personal agents who have access to

your emails, calendars, and notes, as well as countless sales and customer service agents, such as AI telephone agents. And to illustrate what a real AI agent workflow can look like in practice, I created an N8N myself. This is a sales agent that helps me prepare for calls. There's a Telegram bot called Javis, and

I can send it messages and ask about a customer. This agent then accesses a customer database, like this simple Google Sheet here, which currently only contains a website . First, we have the Telegram trigger note. This checks if I've sent a message . This switch note then checks whether it's a text or

voice message. If I send a voice message to ask questions , it's transcribed by OpenAI Whisper. Then it checks if I'm actually an employee who's allowed to access this bot in this spreadsheet, and if so, it confirms that I am. have access. And then there's this AI agent, and this AI agent now has

access to its brain, the OpenAI model, a GBT chat model, and a memory. It has access to another sub-agent, a research agent, which now goes into my spreadsheet and opens the LinkedIn profile and website to retrieve information. Then it also checks my simple product database to see which product would be particularly suitable for this customer and

then sends me back a Telegram message , as in this case, with a detailed summary of the customer, including product recommendations . Furthermore, the agent helps me by allowing me to send it voice messages and ask questions about my prospect to prepare for conversations . So, let's finally look at where this is all headed and how

companies will use AI. I myself have watched numerous videos about AI agents and came across this statement by Nvidia CEO Jenson Huan , which I found quite insightful . In many ways, the IT department of every company will be the HR department for AI agents in the future. Today, they manage And you'll be dealing with a lot of software from the

IT industry. In the future, you'll be managing, developing, training, and improving a whole range of digital agents and making them available to companies . So every company will be using AI. Just think what an incredible opportunity that is, whether you're an entrepreneur or want to build your own business in this field. AI agents are a

multi-billion dollar market that will forever improve the way we all work by making AI agents an integral part of teams in companies. And what Jenson Wange says is exactly what I experience every day in my own work. We have several IT teams involved in our consulting work on AI agents and receive dozens of inquiries every day from companies that

want to fundamentally integrate AI agents and increase their profits as a result. So, if you've read this far, then congratulations and well done , because you've just fundamentally grasped one of the most important AI concepts for the coming years . And if you're now interested in how exactly you can build AI agents yourself in practice, then

check out this video here. Go in. I'm looking forward to seeing you there again. Until then. Take care, Leo.