From Strategy to Automation: AI Agents & n8n for Business Leaders - n8n Business Lab (November 2025)

All right, welcome everyone. My name is Amal. I'm the CTO and co-founder of Avani, uh, one of the three hosts today with you. And my presentation is entirely focused on the strategy from literally getting to know what is an AI agent to the level that you see an actual use case of this. I'm using this workflow that I'm going to show you guys next on a daily basis. I'm also an NATO ambassador. Um that means I am taking the Nathan brand to the world uh literally every single week uh in a workshop format. Uh yesterday I was in Lisbon, the week before that I was in Vienna. Next week I'll be in Madrid and it keeps going on. By the end of this year we will be in 21 cities. So today my mission is to give you guys a holistic view of what can an AI agent do for you especially in the light of an N8 workflow. So we'll start right away from a live demo. I hope it's not going to break. Uh but the demo that I'm going to show you is like this and it's something that I am using on a daily basis myself. So literally when I do these travels, our accounting department really gets mad if I don't give them these receipts, the scan copy of these receipts. So having AI agents work for you uh and having like a real use case is something that I'm going to show you today. So what I do is I go to um these cities every week, a new city. I get a lot of these ones. The moment I get them, I take a snap of it. You'll see it in a minute. And then it goes through this workflow which we'll build it together um at the kind of like 70% of the progress of this uh today's talk. And then you'll see that it's no more a kind of uh future technology. It's already there and people are using it or nerds like me are using it. I'm a technologist. You'll know a little bit more about me. But let's give the demo a try. So I have my phone attached through this cable to my screen. You can see it here. It's um simple Google Drive app. You can see that um right now I'm running Google Drive and Google Drive has this little scan feature and I have a folder called invoices. So what I'm going to do is go to invoices. you'll see all of the invoices that I have been using lately. And this is a question that a lot of people ask me uh when I go and do these workshops that what is the front end for an A10. The simplest answer is that the front end can be anything. And right now you're seeing the front end a Google Drive mobile application that takes a snap of this uh this receipt and then it turns out into something that's more usable. on the right side, you'll see that uh this entire receipt will be scanned and then you'll see all the data in here. So, let's give it a try. I'm going to scan this receipt right now. Here, there's this little um scan feature. I'm going to scan this and see how the scan goes. Work. This is not an NA10 feature. This is an app that is built by um by Google. They did a good job. So you can do like cropping only if the screen do not go out. I touched the cable. There you go. So now I'm doing these cropping things in here just to give you guys an idea of what's going on. Fingers crossed demos can fail. This is a real demo. So you can see that now it's uh uploading. The upload is right now completed. Um now if everything worked fine, I should be able to run this workflow. And what it's doing is using Mestral AI's OCR. Mal AI is a company based out of uh France. And it just worked. Nice. My stress is no less. So, it put all the information from this in here that I need with a short description of what is it about and also a link that my accounting department can go and find the invoice and get all the information that they need. And this is an actual use case that is being used on a weekly basis. And this is something uh I used to do lit

literally by hand a little bit of OCR here and there but now that we have things going on in this way and also kind of demoing it it's insane like this technology was not possible to this extent um let's say a year ago. Now the main topic of my talk today is that I show you what is an AI agent and what can you do with an AI agent. This is one example of this and there will be thousands of more emerging in your daily work life. uh at Avani we are exactly challenging this that we are reimagining work and sometimes uh if I take workflow as workflow orchestration NA10 it is actually a work building tool so NA10 is a work building it's basically reimagining how we work so a few words about myself I speak five languages human languages around 20 computer languages And I have been building companies for the past 17 years. I built six tech companies. Uh one of which is so dear to my heart. Uh after consulting all those big uh clients. This is the one that I really like the most. This voice AI agent that I built in 2023 and 2024. By show fans, who knows this picture? 40 years old. Oh, 40 years old. So this is a picture from 1980s uh TV series. and David Aselhov talks to this car or for those of you who are young, you might know this picture. It's basically the very early uh voice AI agents showing up as a TV series or as a movie and it was not possible until just recently and I was building this company in 2023 and 2024 and things that are possible today was not possible two years ago, one and a half years ago because of technological shift. What I learned after all of this startup hype and liquidation uh is this amazing technology that literally took the entire orchestration of my company N10. So this in this technology was in a very good time uh coming to me kind of like going down from one side from this startup but also going up from other side as an educator. I am now educating the world uh teaching them NAT through my YouTube channel and as well through my engagement as a co-founder and CTO of um Avanoi partnering with NATO and taking Nathan education to the world. But my educational content is not kind of like here's the ABC of it and go figure it out yourself. All of the content that I come up with is an actual use case as you saw. So, as a quick icebreaker, I would like to ask randomly some of you one of those questions and then I'll go on. So, is it okay if I ask you a question? Okay. So, what's the coolest AI feature that you have seen recently? voice. — Tell me chat GPT voice AI feature. — I like voice — to be honest. Not only that uh I mean like the response uh how they responds. — Yeah. Every time you ask them even if you give them more pressure to give me the straight answer. Don't go uh sideways or something else. Um, this is what I like really because it's it makes it more human. This is what I uh for me it's uh pretty um — Yeah. makes sense. Anyone else wants to participate? Answer one of those three questions. — Come on. — Yeah. — Um let's say if AI could automate one task in your life, what would it be? — Handling my inbox and managing my calendar. — Absolutely. Low hanging fruits. Easy to do it. and also we have some slides coming up about EU AI act regulation wise this is the low hanging fruit. So if I put what is an AI uh agent in one sentence I would choose this sentence and the keyword I would like you guys to focus on is this digital coworker. I think that moment is not so far away that our bosses will introduce Lisa and you'll say, "Hey everyone, please welcome Lisa in a daily standup call. " But the only difference is that Lisa is not a human intelligent. She has a face. She can talk to you. She can learn. She can reason. But she's a digital coworker. The scary part of this is if you're not a part of this whole movement, the scary

part is that she has a PhD plus level IQ. She doesn't sleep. And her salary is a€,000 per month. Good luck competing with her. So, but on the other positive side, if you're a part of it, which you are, therefore you're in this room today, we're building these digital co-workers so that we can save time and really have that 10 hours working week and spend the rest of those 30 hours with our loved ones or do whatever we like to do. And in order to demonstrate this, uh, I'd like you guys to a volunteer to come and talk to my AI agent. It sounds like me. It's my voice clone. One volunteer, please come here. Anyone? It would sound really strange if I talk to myself. Anyone wants to come here and talk to my voice agent? Just a few sentences. Please come here. — I'm just doing it because I can't bear the site. — Thank you. We need some more rebels like yourself. — Uh, should work. if the voice. — Hello, dear. How are you doing? — I'm doing fine. How are you? — I'm doing great. Thanks for asking. How's everything on your end? — No, it's fine. Had a long drive today. Um, let me ask you a question that I was asked recently. What's the coolest AI feature you've seen recently? — Uh, a long drive sounds exhausting, but hopefully peaceful. As for the coolest AI feature, I'd say it's the increasing ability of AI to generate highly realistic virtual environments in real time. With advancements in AIdriven graphics, you can now create immersive virtual worlds for games or simulations that look almost indistinguishable from real life. It's incredible how AI can model and render environments with such detail and precision, opening up a new realm of possibilities for entertainment, training, and beyond. What do you find cool in the world of AI? — Yeah, also that. — Also that. Thank you so much. Thanks a lot. Thank you. — Thank you. — So that Thank you for actually breaking that ice which I couldn't not do it with the icebreaker I had in coming here and talking. So this digital version of ourselves is inevitable. As Mark Zuckerberg said, everybody will have an AI agent and I think this is the early sign of that. Almost everyone will have one of these things. And instead of sending your Calendarly link to people, you will send your AI agent to talk to them and then you get the gist of it. Or if the AI agent realizes that it's the time to call you, it will call you on your phone. Then you jump over and take over the line. And that digital coworker is the key term that I want you guys to remember. It's coming. Some examples of this is lowhanging fruits. Invoice reconciliation. There's a lot of information flowing in an invoice and invoice reconcilation is very easy to build and AI agents ace at this right now. I don't know why that is not doing it in a massive scale although I know they are doing it. They invited me in that headquarters. I give a talk about a project I did about nan and dative and AI agents. Um, another one is cash flow prediction. Like big businesses really can suffer from cash flow, but this is an simple one like all the data moving in different spreadsheets and different accounting systems. This can easily be automated. Compliance uh is another big chunk that we can do because a lot of texts move around and I don't know how many audited reports have audit errors because there was a human in the loop like a human auditor did it but an AI agent capable of doing uh tax audits will do almost zero mistakes. Um, I worked for or I was a partner to a $300 million scaleup company, an e-commerce uh company. And there I learned that you optimize little bit of your uh your payments, especially for your suppliers. There is like 200 pages of each supplier contract and you find one small little clause that you can pay them 5 days in advance. You end up saving thousands or tens of thousands of euros every day. That's not possible with human brain but an AI agent can do that for you. My uh beloved feature of AI is this voice AI. Imagine this uh kind of a real state company implementing a call system, a phone call system that all the tenants who rent get a phone call on the first day of the month. Your rent is due. You didn't pay. Can you please pay? and have all the integration with the bank APIs and so on to make sure that they paid or not and call only those who didn't pay.

So where do you see someone a little bit audience engagement? Where do you see that AI agents making the most impact? There's one very visible area. — Yes. — Pictures. This gentleman says pictures. Anyone else? every everything where you have a lot of very small details which is hard for a human to write — small details text or whatever content it is okay picture content text — making appointments in German uh doctors do with German doctors very true what else — there's a very visible one nobody's saying that right now in this crowd it disrupted this industry it made a huge impact coding. I have been coding for 23 years non-stop. The last lines of code I wrote was like literally this afternoon when I was doing a podcast. So coding was immersively disrupted. Right now we have this concept of vibe coding and this is happening because AI agents now can run in the so-called YOLO mode in 3 four hours build the entire work for you autonomously and that's amazing. So there are a lot of these platforms like the most awesome one NA10 uh relevance AI you can I like their LinkedIn scraper uh Lindy an end toend integration vapia voice AI agent 11 Labs uh an EU based polish company that allows you to do voice cloning voice technology stuff crew AI a Python library that you can build AI agents orchestrate multiple AI agents together. Mastra for typescript engineers, AI SDK agnostic model uh selector and open AI and also Google are getting into this software development kit game bringing their own technologies. But there's another corner of AI agents that uses computer. This is very interesting. Like if you have an API that's fine you can automate almost everything with NA10 but if you don't have an API nowadays AI agents can also use computer they can m move the mouse take a screenshot understand the picture on the screenshot and then click somewhere type something execute your wish for example you say can you please buy me some um uh some bananas from I don't know flink or or whatever grocery it goes on that check list and then buys you this stuff and then does a full checkout. Uh one from Antropic is called computer use uh the first one and then operator from open AI and from China Manus AI. These are projects that are doing this. And I'm this dude uh so deep into Linux and kernels and Docker containers and so on and virtualization and I say that I want to poke things around and think and see how do they do these things. It's not so complicated to put a virtual machine. And then I end up building this one. Uh this is datative accounting software on the right side running inside a web top docker container and a nexjs application here literally taking my commands here I upload multiple invoices it goes there and then uploads it one by one and runs some OCR fills all the fields basically um creates like a full-fledged um reconciliation of an invoice system. Uh this is a technical diagram but just to mention the project's called uh it's an open source project I did. It's called blind clicks. Uh it basically blindly clicks until it can't find the click and then the moment it doesn't find the click it basically wakes up the dragon which is an LLM with a project from Microsoft called omniparsser that takes a screenshot understands in what xy coordinate what icon is there. Can I click it? Can I mouse over it? This is a drop down and so on and so forth. and then the LLM understands it and then it create creates a new sequence and that sequence is executed inside this docker container that's running the entire UI very simple fun project I did imagine if big companies like that if they start building something like this it changes the way uh accounting is done and when I did this post went viral that dudes found it and then they invited me they were so nice and that of HQ and I give this presentation uh to that of uh 200ish engineers in the room. Very pleasing experience. Uh in EU, we need to kind of like it's a little bit controversial. What I'm going to say is kind of like think about these guys like North American guys like Satya and end of uh 24 said like SAS is dead.

Microsoft is a SAS company and Satia is saying like SAS is dead because they're seeing this new wave coming in. Mark Zuckerberg says there will be billions of AI agents and that's we need to kind of get to that ambitious level and true almost every single one of us will have an AI agent and that's the most convenient way to kind of augment ourselves put it out there in the world scroll feed give us a summary when we are taking a shower or compose a LinkedIn uh post when we we're driving these things that I am currently doing I'm a nerd I'm super computer geek I want to build things but this is going to come to the masses and it's inevitable. It's coming. Our Nvidia CEO gives this beautiful diagram that we are right now here before hitting that physical world. We are going to build these agentic AI and this arena is just starting and Salesforce CEO says this is bigger than the internet. It is because it's a tectonic shift. It's the way muscle was uh kind of like productized or commoditized with machineries. Now intelligence is commoditized. So a Lisa that takes a salary of a€,000 per month a PhD plus IQ never sleeps is going to take our jobs. And that's how big it is. It's bigger than the internet. So a quick question from you guys. Do you believe 2026 will be this the breakthrough or the changing year for AI agents especially in Germany or in Europe considering all the regulations GDPR EU AI act and everything else. We have some EO act slides coming in anyone — please. — I believe it might well be the case. Um, I have hit upon madness earlier this year and I was having this like wow moment and um I thought — So you're optimistic it's going to be there hopefully. Yeah. — Yep. there. — Yeah. Because ROI is explosively positive. — It's the same as an Uber killed the taxi company. — Exactly. It's coming. Anyone else wants to please? — Technically definitely but I see problems with itself ad yes advisor and I see many management — mentality mindset — fear and I think the biggest problem especially in Germany is that people are not ready — yes — I've been developing with GPS and agents since years — and I know what is possible but the problem adoption, change management. Exactly. — Is the biggest issue, — right? — And if you try to transport it, you always see a war. — Exactly. This is this — we can and I think there will be some companies that will adapt the technology and use it and companies that don't will get — right. Absolutely. Absolut I am 100% with you. If we do not shift our mindset, therefore the previous slide was about these North American leaders. Now, if we do not shift our mindset to kind of like let's take the risk, there is a 7. 5 million penalty that we could end up paying, but the ROI is probably 20 million. So, let's take the risk. Let's go and implement this despite the fact that this might not be 100% compliant. And who knows that these big players like Mr. AI and Alfala and those dudes are also not compliant. And so I am the CEO of this 500 employee company. I know that I'm going to either go bankrupt in two years or I'm going to take this risk and go build something at Gentech and automate something that will give me 20 million euros every year and I'd risk 7. 5 million of that. So I think that mentality needs to be developed more. I want to talk about EUA act. I'm not a lawyer or a liar or neither of them. I want to talk about it because it's important. I think everyone needs to have at least one or two slides about this act. The first question I ask myself is that why is it there? I have two kids, one and a half years old and six old. They're going to grow up here. Uh I want a safe place for them. I wouldn't go in a country not having traffic lights. And if I ask myself, what does EU act for? It's just like traffic lights. It's a regulation that prevents harm to human beings and it protects our

framework called my EU50 which is a very simple framework that um that is built I think when EU was built that protects one of the 50 elements that are in these six pillars human dignity, freedom, democracy, equality, rule of law and human rights and any of those are threatened by any AI application then you need to go through one of these four layers that are unacceptable in EU that's kind of like literally no go or it has high risk therefore you need to go through a very high uh degree of regulations or it has limited risk therefore the number of regulations you need to fulfill can be limited or it's also called transparency risk and there is minimal risk this low hanging fruit like literally things that you can do with email filters this is minimal risk things that you can do with deep fake this is limited risk for example an AI recommendation system doing the wrong recommendations for example for a child Iris can be anything to do with healthcare or credit scoring system if there is based in data and certain people do not get a good credit scoring for their mortgage loan and so And prohibited is anything social scoring or real time biometric surveillance or predictive policing. For example, if John Doe is someone who has been committing the same traffic uh violation throughout multiple years and we can write a model that can predict with 96% um accuracy that John Doe will do another like go through another uh red light next Tuesday at around 3 p. m. which AI can do if it gets enough data. And why don't we send uh a pace like a ticket a week in advance to John there and say like you can do it or please pay this in advance. This is productive policing. It's something that's prohibited though it's really black mirror-ish idea but uh it's prohibited in Germany. So now it's the time to build. I didn't want it just to be a talk talk. Although I'm not very good in talk. I am this engineer who likes dark basement and dark mode to code. So I'm going to build something in front of you and I will show you things that normal engineering time will take forever. Just by show of hands who is an engineer in this software engineer in this room. Okay. So therefore I will keep the numbers the the vocabulary to non-technical uh caliber. Uh I want to show you guys that this technology NA10 has revolutionized this world of software development. I've been building software as a full stack in total 23 years. Before that I was a desktop engineer. For those of you who knows what Notepad++ is or what Visual Basics 6 means, I am someone from that time and I've been continuously building. Now, what I'm going to show you is probably I can't explain how hard it is to build this thing that you're going to see in the old school way of building things. But when I keep building things, I will try to explain it at least with words that how many hours would it take for a normal software engineer to build such a thing. All right. So, I'm going to go to NA10 and explain a few things uh as I go. So anything is this um very simple orchestration layer where you can build things and those things are called workflows and any workflow can have certain nodes as a tiny node can be doing certain specific uh thing for example here I'm going to add a so-called chat node and this chat node does one thing it allows you to do chat when you do chat it triggers the workflow So you can see here there's a trigger. If I say hi, it just puts a check mark there. That means like this node was executed. It's done. It's executed. Now I want to add an AI agent to this. How do I do this? Very simple. I put an AI agent node. And there we go. An AI agent is here. Now let's take a step back and do a very simple definition of what is an AI agent. Anyone wants to participate before I go on and describe it? What's an AI agent if you want to describe it? — Anyone please? — Agent has a goal for example. — A goal.

— Yeah. — Goal actions. Uh those are the terms I'm hearing. Goal actions — um autonom — autonomous aut um agency. That's the term. Okay. Anyone else? What is an AI agent? Give me some terms and then I'll put the definition there. Yes. — Maybe humanike. — Humanike. I like it. Somebody else uh used the term non-human intelligence. I really like that term as well. Somebody else defined it as uh alien intelligence or something like that. Yeah. Anyone else? What's an AI agent? — You mean just like something large? — An LLM could be a part of this. An AI agent. Nobody talked about tools or tasks. So an AI agent can also do have access to our tools or execute tasks. The simplest definition of an AI agent that I can come up with is I s I show it to you guys earlier but a little bit more technical could be something like this. An AI agent is a software that have access to your tools, understands your natural language, and it can perform tasks on your behalf just like your digital coworker. I really want to coin this digital coworker thing. So if you and see this definition contains certain keywords that I really want you guys to kind of like remember it when you go home especially this digital coworker is is inevitable. It's coming. This the work is going to be disrupted first before even the robots come and then disrupt the muscle part. The intelligence part will be disrupted first and the most important part is performing tasks. So this performing tasks these digital beings or these digital co-workers will perform tasks on what? On the actual tools that you give access to them on your tools. So like these little tools that you give access to them be it your Gmail be it your Jera or uh Google Sheets or Microsoft Teams or whatever it's going to disrupt that and the beauty of it is that it is going to go with this natural language. One of the 90 languages that these LLMs today speak this can change. It's amazing like it will increase the number of languages. All right. So now I have this AI agent in here. It needs an LLM. I'm going to give it an open AI LLM access. That's nice. Now if I type here something, you'll see what happens. So I'm going to type hi. This time it's going to respond something to me. You see here it says how can I assist you today. This is basically a chat GPT integration. I just click click and then everything is done. The term that I really liked about anything is Lego. An is like a Lego is like smaller pieces. You put them together and then it works. So now there is a concept in um in programming you might have heard is called cache caching. Look how easy it is to implement a cache layer here. Why do I need this cache layer? I'll explain it here. So I'm going to use this workflow to do this. Hi, my name is Amal. See what happens. This workflow says this thing. Hi Amal, nice to meet you. How can I assist you today? The next question I'm going to ask is what is my name? If I ask this, look what happens. Previously it says I amal. Now it says I don't have access to your name. What is the problem here? There is a big technological problem here that every single time that we send this traffic this request to Chad GPT we basically send plain first fresh information now somehow we need to cache it somewhere we need to save it somewhere now I'm running on NA10 cloud and N10 cloud is super scalable you can imagine what $200 million can buy you

the recent funding of NA10. Now if I click on simple memory, click on back, those were just three clicks I did. It literally added a cache layer. Now for those of you who are a programmer and you know what cache invalidation means, it's one of the complicated things in computer science. And this just implemented a cache layer for me. Now if I go back and do the following, I say, "Hi, my name is Amal. " It goes and stores it here in this tiny database on the servers of NAN. And if I say, "What is my name? " Now, it goes and pulls it back from that. And then it now have access to my name. And this is just a tiny example of how cool this technology is and how cool the 1. 1 million lines of code that was stacked upon each other through human brain. At the time there was no vibe coding. There was no AI agent code generation. It has happened throughout six years almost 15,000 so-called pull requests. A pull request is a bunch of code that is get together in very close collaboration of multiple developers and then it ends up being building such products. I will demonstrate one small thing here just to make the point on how cool this product is. And I I'm in Germany. I can't really brag about this but when I go to other countries I brag about this German perfection or made in Germany thing. It's uh it's like this. For example, here I'm adding a so-called trigger. And when you add a trigger, look what happens. If I want to add a Google Drive trigger for the workflow that I just demonstrated earlier, I need to have a Google Drive trigger like this, right? If I add a Google Drive trigger, do you see a tiny little feature here that must have gone through a feature request, a product owner worked on it, a programmer implemented it is this thing. This little dude is collapsed by default because it's a trigger time. You don't you can't add an action here. So it's a very tiny little uh example of this. But there are others like this. So for example, if I add this particular workflow that I have created here that enabled me to do this thing, you'll see what happens. I add this invoices folder and I look for files to be created. And when I execute this, it goes and pulls out all the metadata of that newly updated invoice. Not the most complicated thing. The thing that really sparked this idea of German perfection I want to keep talking about is coming in two steps. If I now want to download this file, I download it. That's fine. Um, and I want to download it from this particular ID that's sitting somewhere in here. I will need to find it. Here you go. It's just coming in a few seconds. That perfection, that Lego blocks kind of like clicking together. And that could have not been possible without going through all that evolution of that product across six years. So here I'm adding a mistral AI integration because now I have this particular file with me on this canvas which is literally a file that can be downloaded. It's a PDF file with 556 KBs and it's available. You can see it just downloaded and if I open it it's exactly that PDF file that I scanned earlier. Now if I integrate this with Mr. AI. Mr. AI is a French company that's building LLMs and also their document extractor is very nice. Now, see, I opened it. I don't do anything. It just clicks. Why? Because the programmer sitting in Berlin said, "We're going to call our field data. " And the programmer sitting in Paris said, "We're going to call it data. " So, that German perfection like literally everything clips together. I don't change anything. And then I execute this. It just goes on and then it basically uh when you put the credentials it basically executes and then it just works. I was giving this advanced example advanced course earlier um and showing them if something fails because you provided an incorrect

account how can you trace it? Therefore, that error that you saw is from there. Now, it literally three steps and I'm send downloading some data from Google Drive, uploading it somewhere else and Mr. AI is giving me back all this information. This level of product development with without having such a tool would take me forever. Leave alone the fact that I could literally demo it here. This would have been almost impossible almost. I want to make sure that I have enough time left for the Q& As's. But if I want to conclude this presentation, I would like to conclude it with this that from this basics agents basics uh to summarizing emails and everything uh where we are going next is amazing. It's going to be this agentic workflows, enterprise focused wrap-ups. And the bottom line is that AI agents are no longer experimental. It's there to stay and we are going to see in 2026 major shift towards adopting AI and NAN is going to be this technology that people adopt and then automate things with AI and this the the areas where AI agents fail are very important to also know. the ambiguity in your inputs. This geigo garbage in garbage out, the missing data or context, the elucination. Nathan recently announced something called guard rails to kind of prevent AI agents to elucinate and to prevent doing things that they're not supposed to do, evaluations a couple of months ago. These [clears throat] little tiny builds and whistles around agentic workflows are going to change the way it's going to do. But the lesson to learn here is that success of an AI agent is not just an AI automation. You put it in the while it's going to work. The human in the loop is very important and also the clear constraints and instructions. Thank you so much. Any questions?