I Tested OpenAI's AgentKit Against n8n: What You Need to Know

OpenAI just killed Naden. Or did they really? Because they just dropped their agent kit. We may be seeing that narrative right now online. But today, I am here to actually compare Agent Kit and Naden and talk about the future of both of these agent builders. As always, we're not just going to be talking about it. I'm going to be comparing both of these platforms across different categories so we can highlight each of the platform strengths and weaknesses. So, to start off with a quick overview of each of these tools, we've got Agent Kit, which just dropped October 6th of 2025. It's OpenAI's agent builder that looks like this. It's very very simple to use and it's very drag and drop low code visual interface. Here's a quote from Sam Alman saying, "Agent Builder is like Canva for building agents. It's a fast visual way to design the logic, steps, ideas." I'm assuming you guys already know what Naden is, but if you don't, it was released October 8th, 2019, so about 6 years ago. But in the last year, it really started to take off. As you can see by this Google Trends chart right here, Naden's in the blue. And then we have other AI automation platforms like Make and Zapier down below. And you can see it just hockey stickked right past them. You can plug AI into your own data and there's over 500 integrations which makes it really easy to connect to your tools and this is what NN looks like. So like I said today we're going to be comparing agent kit with NN across different categories. To answer the question is Agent Kit going to kill NAND? And in short my answer is no. When we started hearing rumors about this dropping everyone immediately ran to it's going to kill NAND? I don't think it's going to at all. And I also want to be really clear that I don't think OpenAI's goal was to compete with NAND. I think that Agent Kit is built for teams and consumers that are really integrated into OpenAI's environment and they're not looking to build super super custom robust AI automations and autonomous systems. They're looking to build some really quick and easy workflows that's going to save themselves time and maybe even some conversational agents internally for their team. Now, NN of course was built to be completely everything. It's super powerful. It was honestly built for developers. It's a little bit more technical, but because of that, you have the ability to truly do anything. Like when we saw Google drop their Opal, which was like very similar visual interface. People said, "Oh, is this going to kill any?" No. Real quick, why should you trust my opinion on this? I run an AI automation community where I'm teaching over 150,000 people how to build AI solutions to implement into businesses. And all of that is being done with NAN. Our AI company, True Horizon, consults and implements custom AI solutions. And pretty much all of that is NADN. And we've been working with enterprise companies as well. So the categories that we're going to look at today are ease of use, triggers, agent tools, model support, UI chat components, and deployment and control. So starting off with the first category, we have ease of use. So if I'm like a complete beginner and I want to test out agent kit, which is OpenAI agent builder, I would come here and I would search OpenAI agent builder. I could look at the release notes or I could go here to the OpenAI platform, click on agent builder, and then I could open up my agent builder right here. This is pretty much available to everyone. You just have to have an OpenAI API account. You can see here we have the ability to create a workflow. And what's key here in my mind is it says to build a chat agent. Now a chat agent or a conversational agent in my mind is very different to like a fully autonomous tools agent because the chat agent is going to be looking up some information for you, helping you get your questions answered, but not as much like take action on your behalf autonomously. So anyways, we have some templates down here that we could look at, but I'm just going to go ahead and create a new flow so we can get familiarized with this environment. What you'll notice here is that we have a start and I actually cannot delete this start because every workflow needs a trigger. You can see that once you have this start, you have to drag the circle to say where does this go next just like end where you'd connect your nodes together. And we have an AI agent right here which we're able to name. We're able to give it instructions. We can include chat history. We can choose the model, the reasoning, we can give it tools. And we have an output format section where we can do text, JSON or widget, which we'll talk a little bit about later. But what I'm immediately seeing off the bat is this whole interface is a lot less intimidating than nitn. I can easily choose my model, give it instructions, and I can also on the lefthand side see everything I have available to me, which is an agent. I can end it. I can do a file search. I can do guard rails, MCP servers. I've only got five logic and data operations to choose from. If, else, while, human approval, transform, or set state. And this is a lot less intimidating to me as a complete beginner with no coding background to get into compared to when I jump into something like niten create a new workflow and then when I open up this first step I have tons of different things in here. Let's say I want to pull in an AI agent and I can click on that agent right here. I have this input where I'm looking at this variable. I have all these options down here to choose from and I might not know exactly what they mean. You can also see I have an error already because I didn't connect to chat model. I also have the option to add memory and all these different tools. And this just initially is going to be a lot more intimidating than hopping into the agent builder by OpenAI. One of the first things that I wanted to do when I built my first agent ever is I wanted to connect it to the internet so it could look up things for me. And with OpenAI's agent builder, all I have to do is click on this plus for tool and I can just do web search. And that's basically going to search the way that you can do web search when you're in chatbt. You can also specify to only look through certain websites and you can also do location-based searches. So just like that, I already have an AI agent that can search the web and I can also connect a chat model right here without ever having to go and get any API key. But in any if we wanted to search the web, we'd have to give it a tool. We'd have to choose from all of these hundreds of different tools. We may even need to set up our own custom HTTP request and we would also have to go get an API key for both the web search and also the chat model, which is just a lot more intimidating and definitely raises the barrier to entry a little bit compared to the agent builder. So if I was to think about the time that it would take for a complete beginner to build one automation system in OpenAI agent builder or the exact same system in Naden, they would do it much quicker and feel a lot less overwhelmed in agent builder hands down. Which is why for this first category, agent kit gets an 8 out of 10. Naden gets a 6 out of 10 and that is our new total for the scores right now. Anyways, let's move on to the next category which is triggers. So remember earlier when we were looking at this workflow section and it says to build a chat agent. Now when we look at the triggers, you can kind of see why I am drawn to that or not drawn to that fixated on that because when you have your AI agent and you have just this start option, there's no other triggers available as you can see which means that if we want to communicate with this agent, the only way we can actually trigger it is by talking to it as you can see over here in this preview. So if I want to test my workflow, I can chat with it right here and I can say search the web for NDN. We can see visually that the agent right now is using its web search function and it's using its reasoning to search the web for Naden. And that's really cool because we spun up a web agent in matter of seconds. But the issue here is I don't see a great way to have web- based events or scheduled triggers. Even if we go up here to publish our AI agent, let's just say we're going to call this test web agent or text, I guess. And when we publish this, it said right there that it could be used in API calls. I don't know if you guys caught that, but what we're able to do is either embed it as chatkit, which I will talk about a little bit later, or we could trigger this agent from an agent's SDK, and it gives us the TypeScript or the Python. So, what this tells me is you technically could be able to trigger this agent from uh an HTTP request or from like an action in an app, but it's not super intuitive or easy. It would be really nice if OpenAI had scheduled triggers or app events, stuff like that. For instance, if you wanted to make an open AAI agent to respond to your emails, it would not be as easy as an NAN because there's not like a Gmail trigger on new message received trigger. Whereas an NAN and you want to look at your triggers, we could go to Gmail right here and then we could see our trigger is on new message received. So we could easily have this agent go off whenever we get a new Gmail. We also have eight triggers in Slack. So on any event, on bot mentioned, on file shared, on new user, on a reaction added, there's so many triggers that are already natively baked into Nitn. We we could of course have a web hook. So our CRM could send data, we could have Google Drive, we could have Twilio, we can have pretty much anything. And in my mind, the systems that are most valuable when it comes to automation or AI automation are the ones that the human doesn't have to manually trigger. So real quick example, a lot of businesses reach out to us and they want to integrate this personal assistant that I had built. However, when we're trying to actually communicate what would be the best quick win to show them ROI really fast, I never would recommend a personal assistant. What I would recommend is something that nurtures leads or responds to leads quicker and stuff like that because those type of systems help grow the business and as the business grows, those systems will get used more and more. So, the throughput of the system increases as the business grows, which is just a scalable compoundable ROI of the system. And it's not the exact same with something like a personal assistant because as the business grows, it doesn't necessarily mean that you're going to be using the system more. So, in a nutshell, the most powerful automations in my mind are ones that run in the background when you even know they're running. And because in OpenAI's agent builder, we're kind of limited on our triggers and being able to schedule or have things go off in the background. That's why I think any definitely takes the cake here, coming in at a 10 out of 10 and agent kit comes at a 5 out of 10, which makes our new score agent kit 13 and 16. All right, moving on to agent tools, which is really, really important because this is what gives our agents the ability to actually do things on our behalf. So, we're back in our OpenAI agent builder flow where we already gave our agent the tool of web search, which was super easy. No API key, really quick. I love it. But now, when we go to add more tools, you can see we only have a few options. We have a client tool, which is the ability to send data back to our chat kit widget, which I will say that is very cool. We can do MCP servers. So, this is nice because you can connect really easily to Gmail, Google Calendar, Drive, Outlook. You can see these other integrations right here, but there aren't a ton here and they all have to be MCP servers. You also could connect to a different MCP server, which is cool because there's, you know, tons in Zapier. You can even connect to your own custom end workflow and there's also lots of other MCP servers that you could connect to. And it's not as powerful as how many APIs there are out there and how many you could connect to with a standard HTTP request. And in Nitn when you come in here to connect tools you can first of all also call on an NIDN workflow which is super powerful and then of course you have all of these native integrations which is just there's literally hundreds and hundreds I think it's over 500 native integrations that they have. So almost any tool that you could think of you can connect to with your AI agents here. And if Naden doesn't have a native integration, you can use this HTTP request tool which is the most dynamic and powerful tool in the world because you can literally talk to any service that has API documentation. And like I mentioned with being able to call on subworkflows, you can see here with the ultimate personal assistant, for example, it has an email agent. And if I open up this email agent workflow, what we see is a completely separate AI agent that I built in Naden. And this agent in itself has tons of other tools that it can use in here. So you can build some really really powerful orchestration agent systems in NN. And all of these are super modular and reusable because now if I ever need an email agent, I can just hook it up to this one. Even if this agent is hooked up to like 100, it doesn't matter. And yes, I did say that you can hook up an OpenAI agent to an NN agent, which does work, but you would need to use this, which is the MCP server trigger right here. And you would be able to hook up this MCP server to a bunch of different tools. So, you do have the ability to talk to some custom agents and stuff in these workflows down here. It's just not as powerful as the sub workflow as a tool option. So, for agent tools, agent kit's going to come in once again at a five. N's going to come in at a 10. And our new total is agent kit 18. N 26. So, let's move into the next category, which is model support. So, this one's pretty simple and it's pretty easy because in OpenAI, of course, you only have access to the OpenAI models. And that's really not a huge problem. As you can see, we have tons and tons of models. You've got heavy ones, fast ones, reasoning ones. It's really not that big of a deal. And you also in here can change the reasoning effort really quickly. You can change the verbosity, the summary, the tool choice, all of this kind of stuff. And once again, because it's all baked into your OpenAI environment, you already have everything right here. You're also able to easily toggle on or off if you want chat history to be included or not. But in NN, what you can do is choose your own chat model. So, Anthropic, Azure, Bedrock, Coher, any of these. Or you could also just go to open router. And now you can choose pretty much any of the hundreds of models that are on Open Router. So you have way more freedom here. And honestly, I think that is very important because I don't always only just use OpenAI's models. I do a lot of the times, but for certain use cases, I really want to use a Google or I really want to use anthropic. And when you choose any of these models, you still have the option to change things like your top P, your frequency, your sampling temperature, all that kind of stuff as well in NAND, which is why NIAND comes in at a 10 out of 10 because you can also use local models because you can host NAND locally. And agent kit's going to come in at a 6 out of 10 here, pushing our new total to agent kit 24 and N36. Now, this one's really interesting. We are moving on to UI chat components. So earlier in this video, I alluded a few times to something called Chatkit, which is like Agent Kit's built-in tool to help you create like website widgets and embed your OpenAI agents really easily and really cleanly onto your own website. Users can add slick branded chat interfaces to their apps without extra coding. Everything looks polished and works with OpenAI agents out of the box. And that is actually a really big value ad for OpenAI's agent kit. And it end on the other hand focuses way more on the backend workflows. So, it has no fancy or easy way to embed chat interfaces by default. You would typically need to embed your agent into some sort of front end if you want it to actually look nice. Here's a quick example of all of the different widgets you could create and you could have your OpenAI agents power these different widgets. And that's where this output thing comes into play. If I change this to widget, we could basically set up the type of widget we want and what little things that our agents could change based on its output in that widget. And so when I click on create widget here, you can see that we could either describe and start to mock up what we want it to look like. You can go to the gallery and see these are types of widgets and this is what you could have your agent actually interface with and change. And so that's why I think this is really cool for their front end. But with Naden, you have the option with the chat message trigger to open up a chat interface and embed it, but it just looks like this. It's really hard to customize and it just doesn't look great. So here's a cool quote from HubSpot. Chatkit saved us weeks of custom front-end work, making it easy for us to prototype enhancements to the UI of HubSpot's breeze assistant and agents. With the custom response widget, our agent can deliver interactive guided solutions instead of static replies. So, it is very cool, which is why agent kit is going to get a 9 out of 10 and Naden's going to get a 5 out of 10, moving the total to agent kit 33 and end 41. Hopefully, I've done all this math right. Let me know if I have not. And now to talk about our final section, we have deployment and control, which is obviously a big deal, especially if you want to do this for your own business or help other businesses implement AI automations. One of the biggest reasons in my mind that blew up besides being easy to use and visual is because it was open- source. I think they prefer to actually use the term code available or something like that, but meaning you can host it in the cloud, you can host it on your own private server, you can host it locally on your own computer, and you can have full control over all of the data, especially if you're locally hosting your LLM as well. You know where everything's going. Agent Kit, on the other hand, is deployed and managed entirely in OpenAI's cloud, so users don't have to worry about the technical setup, which is good. Everything's hosted for them, but also OpenAI has full control over where the data and agents are living. And once again, if your company and your workflows are already completely baked into OpenAI's environment, then that's probably not a huge deal for you, and you could easily spin up some really helpful automations with OpenAI's Agent Kit. But Naden, once again, you can have full control, although sometimes that does require a little bit more technical setup or expertise at the beginning. So for deployment and control, agent kit gets a 7 out of 10. Naden gets a 10 out of 10, moving the total to a final agent kit 40 and NADN 51, which means that NADN is our winner today. But once again, this is very relative. If I was approaching this from a standpoint of a different type of consumer, maybe not as much of a developer standpoint, I would probably enjoy Agent Kit more. So anyways, there's a few more miscellaneous things I wanted to hit on real quick about the two platforms. So like pricing, evaluation tools, and community support. This is still not a finished product by Agent Kit, and they're capable of doing a ton of cool stuff here. So their pricing is not fully finalized right now. You can just basically access it for free, but I believe you'd be paying for the usage of the AI chat models. Agent Kit also has some really cool evaluation tools, but there's also some aspects of the evaluation that I don't love. So, let me show you what I mean by that real quick. Okay, so this is one of OpenAI's templates. And what I want to do real quick is just preview it and just do a quick conversation. So, I first of all say hello. You can see what's going on over here. The triage agent is going to figure out what to output. And then it basically sends me back some information and it says to get started, could you share the following? You can see I responded to that. You can see we saw the path that this workflow took. And now the launch helper agent is basically creating us a guide on how to launch this. So regardless of the output, what I don't love here is I can't see how the data moved through each of these steps. So if I close out of this preview and I click on, you know, that we want to see what happened in this condition, it's kind of hard to see what data just passed through. And if I click on evaluate, this is where we could actually kind of, you know, see those logs, but it's a little bit confusing. Like you can see the timing of each of these steps. You can see what happened. If I click on this triage agent, we can see that it used GPT5. We can see the system prompts. We can see the input right here. And we can see the reasoning and the actions, but it's just a lot harder to see the data moving through the nodes in the way that NIDAN makes it a lot easier to do because in NIN with this workflow, you can see everything's green. You can see how many items are moving through. When you click into a node, you can see the input on the left, you can see the configuration in the middle, and you can see the output on the right. So, it's really easy to follow the data trail and understand exactly where something went wrong and what you need to change. But in my mind, in this OpenAI agent kit environment, it's just hard for me to understand what's coming in and what's going out because you can see here we have a variable which is input output parsed blah blah blah. But I don't exactly know where that variable is coming from. I would have to come into here and I would have to try to figure out, okay, output format, we have JSON, we have this response schema, and is this where it's outputting the information? So anyways, not trying to get too technical here, but I really like in Nitn how you can see the outputs of the nodes. You can see data flow. You can see the input configuration and the output. And you can also click on executions and you can go see all of the different runs very clearly what happened in a much more visual interface than going into the agent builder, looking at the evaluations, and looking at all of the logs and having to kind of look through it like this. It's just a bit more confusing, which surprises me because it's so much easier to use if I was a complete beginner. But they also do have some really cool evaluation capabilities with data sets, trace grading, and prompt optimization. As you can see here, there's different ways that you can grade your runs and then like pass prompts through to see if they're passing or failing. You can optimize your prompts, and of course, you can trace grade. The last thing I wanted to hit on here was community support. OpenAI obviously just released their agent kit yesterday, and we all know what that company is capable of. They're it's just really cool what they're working on. So, I imagine we're going to see Agent Kit continue to evolve and evolve. But because Eniden's been around for 6 years and has blown up in the past year, there's so much content, there's so many courses, there's so many free templates, like 5,000 plus templates, whereas in Open AI's Agent Kit, you're a little bit more on your own right now. Anyways, overall, we have Agent Kit at 40, we have Naden at 51. And I wanted to talk about real quick when you would choose each of these tools. So, I would choose OpenAI agent kit if I wanted to prioritize rapid deployment with minimal technical complexity. If I wanted a polished chat user interface or some widgets that could be dynamic, if I'm looking to do some really quick and easy comprehensive agent evaluation and testing. And of course, if I have everything that I know about my business already in OpenAI's ecosystem, it's going to make it really easy to plug in these custom workflows. Now, Nen, they definitely were targeting more of the developers. Flexibility across multiple AI providers, complex workflow automation beyond simple chat, cost control through self-hosting, complete data and infrastructure control. You obviously have the ability to connect to literally anything, trigger your workflows by literally anything, and do any type of data manipulation or transformation that you need. But making this video was really fun for me because I think a lot of people tend to get caught up on tools. You know, what's going to drop? What's going to be relevant? What do I learn? Am I taking a huge risk or gamble? The answer is no. as long as you're approaching it from the mindset of I want to learn when AI is valuable, when it's not, how to save time, and how to solve a core problem that is actually facing the business. Because if you're trying to work with businesses, they're probably not going to choose you because of your specific tool. They're going to choose you because you know how to deliver them results that save them money, save them time, and increase their focus. So, real quick, the mindset should be become tool agnostic. Solve what the problem is at its core. Doesn't really matter how you get there. So, if I apologize for the ramble there, but hopefully you guys understand where I'm coming from. That's the type of stuff I love to talk about in this space because there is a lot of hype. And if you're looking to connect with a ton of other people who enjoy talking about that kind of stuff as well, then definitely check out my plus community. The link for this is down in the description. We just hit 3,000 members, which is super exciting, but it's full of people that are building with NAND every day or different platform. It's a community full of AI problem solvers. Tons of people in here are building businesses with AI or using AI to help their own businesses. It's a really cool space to be. And we also have three full courses right now. Agent zero which is the foundations of AI automation for beginners. We have 10 hours to 10 seconds where we dive into naden and identifying how to save time with automations. And then we have a new course for our annual members called one person AI automation agency where we lay the foundation for building a scalable AI automation business. I also run one live call a week in here which have been really fun lately. Been getting a lot of cool questions and cool discussions. So I'd love to see you guys in this community. But that's going to do it for today's video. If you enjoyed you learned something new, please give it a like. It definitely helps me out a ton. And as always, I appreciate you guys making it to the end of the video. I'll see you on the next one.