⚡AI-Agents: Der nächste große Schritt der KI-Automatisierung! 🚀

This is a topic that shouldn't be lost on any of us, AI agents. Why are they relevant when we think about the AI applications we work with today? How is an AI agent different from what I do in ChatGPT or Copilot?

And what does that have to do with my work? That's what we want to look at in this video. You will have understood what exactly an agent is, how it works, how it differs from classic automation and what I do with ChatGPT.

So stay tuned, it remains very exciting. And even if you haven't written a line of code in your life so far. Okay, where can you start, how can you understand the whole thing?

if you've only ever used ChatGPT or Copilot? So imagine it like this. Level 1 would be that you enter a prompt in ChatGPT and get your result.

And the result is good or bad, depending on how good your prompt is. With a really good prompt, you can solve almost anything, to be honest. So in most cases, I look at it and realize relatively quickly that it's not prompting well.

And many people think they can already do it, but it's worth practicing. That's the easy part. That is, you take an email, say to ChatGPT here, this is my problem and so on, put it in, get your result manually, make yourselves personally very fast, cool.

Level 2, you get an email. This email is automatically sent to ChatGPT or to GPT in the background. This GPT passes this email on to the next person, to the next person, to the next person.

And they go through various steps, as far as I'm concerned. Well, that's how it works in some of the experiments we carry out. An email comes in, the topic is analyzed, the question goes to CRM, the CRM plays something back, an answer is formulated, it is counter-researched again, result comes.

That's already several steps in a row. That's AI automation or what you recognize from Zapier or Make. com or N8N.

There are lots of platforms that make this possible. And now it's getting exciting. Now comes the third step.

You'll be hearing the buzzword AI agent again and again in the press over the next few weeks. That's the hype that's building up right now. I'm a big fan of categorizing this properly.

Is it really hype or not? I would say it's a bigger thing than we think. AI agents now do the following.

An email comes in or an issue or a problem or whatever. And you've got a bunch of different little agents here that can do things. And in the background with large language models, they just understand human input.

I got an email in here, I have no idea what to do with it. And now these agents can decide for themselves whether they want another agent to do something. And they can give it back and pass it on to the next one.

And maybe to someone who re-reads the text and someone who checks it again and so on. And these agents can pass things back and forth autonomously. That's called a swarm, an agent swarm.

Large-language models get better when you break tasks down into small parts. Now imagine you don't have to do that, but the model does it on its own. And each of these agents, if you take a look, has certain instructions, they're called instructions.

That's actually nothing more than a prompt behind it. Along the lines of, you're an agent who specializes in recognizing the tone in sales inquiries. Is someone stressed, whatever.

And by the way, so that you can do this well, you have here, and this is the second thing they have, a tool, a knowledge database in the background that you can access with our old cases that we're training on. It's like an agent that you've trained to do exactly that. It's working extremely well, already.

Before I forget, if you've enjoyed the video up to this point, because this is all generated without AI, it's so real, it's so deep, it's human in here, this is where it comes out, it's totally sacred and important to me, then I would be incredibly happy, and so would the whole team, if you subscribe to the channel, switch on the bell so that you are also notified of new videos, because that brings us to the forefront and makes us very happy to continue. So, now it's on to the agents. That would be a bit like one of them saying, hey, now it's really good, if the research bot would take another look.

Or maybe there's even a manager agent who says, watch out, your result is okay, but I'd like the research bot to take another look at it now. It's so amazing what's possible with this, because just like when you break a prompt down into lots of individual steps and the result is much better, that's what we're currently seeing with the O1 Preview model at OpenAI, then you could imagine if you take a more complex system, which is a company for example or a team, into many agents that work on certain topics, and the more of them there are, because you can have thousands of them, then you are like the CEO of yourself with thousands of agents and you suddenly have a lot of these individual agents who can then do things for you autonomously. The term agent is not new.

There's a presentation by Andrew Ning, this year, last year, I don't know, he's been doing it for a while. There's Chat-Dev, there was Hugging-GPT, I think, on the Hugging-Face platform. So there have been many attempts to organize agents.

But now there's another very dominant player in the whole AI space, which is OpenAI. And I was sitting with friends of mine a week ago on a Friday evening and suddenly saw a release from OpenAI on GitHub, the developer platform, where a significant contributor with a team from OpenAI had published code there for an application of swarms, i. e.

agent swarms. These agents have orders, just like I just described, have instructions, can access tools, can access the OpenAI platform and can be built in with a few lines of code and, be careful, have these handoffs to the other agent built into the code. If you guys say, but I can't code, I don't know how to do it.

Ey, guys, there are cursors now, there are lots of things, you can generate code with a simple prompt and I tried it at the weekend and it actually works. That is, this principle that I explained before with, my task comes in and I now somehow need ten agents to solve the whole thing, please think about what they are, what knowledge they need and so on, give them to the AI, get a result and have a finished program in a file, in a directory, a ready-made program that solves this for you. OpenAI has deliberately said that this is not yet intended for production, it's just for testing, just for learning, just for playing.

But the very fact that they have published it and that you can work with it clearly shows me that this is the direction we're heading in. Microsoft has just announced that they are also relying on agents in Cooperated Studio. Google already said a lot about this at the beginning of the year, but the way it would be integrated into code now, for me it means the next step in app development is going in this direction.

That means we will probably see in the next few years how we become the CEO of our own agent swarm, that does things for us. I think that's pretty rad and I'll explain to you now, why I think this will even be built into the systems in the near future. And if we now take another look into the magic ball, why isn't it integrated into ChatGPT yet?

And I think it's totally worth taking another step back. There were already ways to interact with large language models before ChatGPT. We already had that in various videos, when there were AI scientists here who explained, hey, we've been using them for a long time, it's nothing new.

But the way of interacting with the chat was new. We are now in the early days of figuring out how we can best interact with AI. I find it so exciting to think, what could this mean for my company?

Can I learn something? And to play around with what can be automated in the company. Two specific things you can do now.

Most of you probably know, maybe if you don't. You can go into a normal chat from ChatGPT with Add to one of your GPTs that you have saved. So if you've built your own little GPT, you shouldn't know what that is.

There's a video about it, we made it. Then you can use this GPT with Add, I don't know, workshop feedback, chat to it. And then it's like you pass the question on to a specialized agent.

The second thing is, you make a test account on Zapier or Make. com or one of the platforms where you can build something with almost no-code experience. And play around with what happens when I send a message in Slack automatically to an assistant, that's what OpenAI calls it, Playground, etc.

It's not that difficult, there are great channels that show this. Those are two specific things. And if you say I want to go in full-blown, then download Cursor, look at the post on GitHub.

In Cursor, it's a coding tool, you can ask, hey, how does that work? And Cursor will explain it to you step by step. And suddenly you're writing code yourself, that's a pretty cool experience.

It doesn't always work straight away, but you can have it explained to you step by step and test how little effort it takes to build a Swarm Agent network. It's absurd what's possible. That's my view of the future.

I strongly assume that OpenAI will incorporate this at some point and we won't have to make proactive decisions anymore. And then you will fulfill a new product. I don't know what it looks like yet.

It's a chat app or whatever that we interact with. Because it probably won't stop there, that we type things in or just say things, but the AI will see where we are. And we'll say, hey, how do I get this solved?

And the agents will recognize independently in the background, where am I right now, what do I need right now? Pretty crazy. What else is my job that I have here?

I don't know. Never mind. So as long as you subscribe here and stay tuned and comment, then I definitely still have something to do.

Because the way we interact here is absolutely not AI, but genuinely human, interpersonally important. So subscribe to the channel, turn on the notifications, comment down diligently and I'll go and think about it now, what my job will actually be in the future.