Check out the conversation on Apple, Spotify, and YouTube.
Why n8n was the hottest AI tool of 2025 (01:30)
Aakash: n8n was the hottest tool in my newsletter last year. I wrote about it over 25 times. You can use it to learn the basics of AI, like RAG and fine-tuning. You can use it to automate your workflows. You can use it to create AI agents with its visual workflow builder. And then Claude Cowork came out, and then there was the rise of Claude Code.
Today we have on Jan Oberhauser, the CEO and founder of n8n. We’re going to get to the bottom of what n8n can do that Claude Code and Cowork can’t. What is the role for n8n this year, in 2026 and 2027? How should you think about using n8n versus these other tools? We’ll also talk about how they build product at n8n, what their latest metrics are, and much more. Jan, thanks for being on the podcast.
Jan: Thanks for having me. Excited to be here.
Do you still need n8n in the age of Claude Code (02:28)
Aakash: So, n8n was literally the hottest AI tool in the world in 2025. Now people are declaring you all dead. I looked at the Google search trends, and there might be something behind it. If you look at these trends here, and then this tweet went viral from John Ennis: “Remember n8n went irrelevant pretty quickly, huh?” Do you need n8n anymore?
Jan: I think there’s a lot to unpack there. First, it’s probably important to call out that we have been called that probably a thousand times. I don’t remember half of the things that killed us over the years. We’re still here. So I think that probably doesn’t hold true this time either.
It’s also quite interesting. If you check out half of the other tools out there and type them into Google Trends, you see very similar trajectories. There’s obviously a hype cycle where something gets a bit overhyped, and then it goes back into a more normal framing again. I think that’s where we are with most products.
But maybe first, let’s talk a bit more about Claude Code and the others. I think it’s an amazing tool and really deserves to be out there. But the most important thing is that n8n and Claude Code are very, very different products, and in the end you need both. Claude Code is more of an agentic tool. It runs Anthropic models in your terminal. n8n you can see more as an orchestration layer that connects your tools, your LLMs, and your data sources, and it offers you a visual canvas for systems to run reliably and securely. This is especially important for business-critical use cases where technical and non-technical people collaborate.
I genuinely think n8n is more important than ever for these business-critical use cases, where reliability, security, and auditability really matter. You want to be 100% sure what is running. You can inspect what’s running, and you can see what did run. You cannot just rely on it working 95% of the time. You have to be 100% sure that it really works. That’s where our canvas really shines, because it shows exactly how it works and what it’s done.
It also matters when you work with other people and talk about what is actually running. If you have something like Claude Code, it generates literally 10,000 lines of code that nobody can inspect, and nobody knows if it’s actually doing the right thing.
We also see a lot of people using them together. They very often prototype with Claude Code, because you can get something started very, very fast, but then they migrate it afterwards to something like n8n for the reasons I already mentioned. Especially for auditability, when you want to handle large or complex data processing pipelines and define more precisely what should happen, and where self-hostability really matters, because people care more and more about data privacy and security.
One important thing to be aware of is that multiple model companies are using n8n as well. That makes sense, because in the end you want to use the right tool for the use case. They’re using us for things like security or compliance use cases, because that’s where we really shine.
And if you look at our stats online, you still see crazy growth. We crossed 200,000 GitHub stars recently. We have one and a half million active users. We have over 1,200 enterprise customers, and I’m not talking about enterprises using n8n. I’m talking about 1,200 enterprise customers using our enterprise solution. We have 300 ambassadors out there. There’s literally one n8n community event worldwide every day. In September alone, we have over 50 events happening worldwide in a single month. We increased our revenues 10x over the last year. The median number of enterprise instance users more than doubled over the last months. So we definitely still see a lot of growth.
One last thing to call out. There are still two very critical things that make me very confident about the future of n8n. One is model flexibility. What people care more and more about is being able to switch models, for the simple reason that new models appear literally daily, all with different capabilities and different prices. People really value the flexibility of n8n to use the right model for their use case. We see models and intelligence becoming more and more commoditized, and that means the value shifts more in our direction. It also matters when you want to connect multiple models together and say, for this use case I want an OpenAI model, for that one I want an Anthropic one, and here I want an open source one.
Moving from the 0.1% of early adopters to the 99% (09:16)
Jan: And the last one is really the community and accessibility. We talked about the hype cycle before. n8n obviously started with very early adopters, the 0.1% of people. Now we’re moving into the 99%, away from the early adopters and toward the people who actually want to get things done, and we’re lowering the entry barrier more and more. We want to make sure that not just technical people can build with n8n, but literally anybody who can use a computer. Our mission is literally to give everybody who uses a computer technical superpowers, and that’s what we’re striving for these days. It generally feels like we’re just getting started. There’s so much opportunity out there, so I’m really excited.
Aakash: Wow. So I’m going to be paying attention when you show this to us in a little bit, around privacy, security, auditability, and reliability, to understand what those words mean. Sometimes when we hear those words in the abstract, it’s hard to understand what they mean in the product. So I’m going to be on the lookout for that.
Did Claude Cowork and Claude Code hurt n8n (10:17)
Aakash: But I want to dig in a bit more on the rise of Claude Cowork and Claude Code specifically. I had on the founder of Lindy. I used Lindy personally for some of the same things I would use n8n for, like automating agentic workflows. He admitted Claude Cowork had a really significant impact on their business. Did Claude Cowork, Claude Code, and the rise of Codex have a noticeable impact on n8n’s business?
Jan: Going back to the times we got killed. At some point, OpenAI launched their own agent builder, and I think we actually had our best week ever when we got called killed by them. The nice thing about all of those moments is that people talk about it more and see more opportunity. So that was quite exciting.
I think every company these days sees people hype things up and use those products a lot. What we’re definitely seeing is that the use cases are changing. In the past, a lot of people started using n8n for personal use cases, like, write me an email summary. What’s changing more and more is that people say, these use cases are great and I need them, but where they really use n8n now is for business-critical use cases. That’s probably the biggest shift we’re seeing.
n8n’s revenue, users and enterprise customers (11:38)
Aakash: So I was trying to find some numbers on n8n. Just in May, you announced your strategic investment from SAP, where you mentioned you’d hit a $5.2 billion valuation. Startup riders put you at $100 million ARR earlier this year. You just mentioned that you had 10x’d your ARR over the last year. What can you tell us about your users and revenue now? Feel free to break any news.
Jan: We’re not showing many more numbers there. I can say we are across $100 million by now, still growing strongly. The user numbers I already shared, one and a half million users, are also growing strongly. But we’re not sharing any additional numbers publicly right now.
Aakash: I think you had said 1,200 enterprise accounts.
Jan: Exactly. The exciting thing is seeing what kind of enterprises are coming in. Take SAP again. They didn’t just make the investment you talked about at the $5.2 billion valuation. They also added n8n to their product. So n8n will be available to literally any SAP customer out of the box. They can use it without installing anything, without setting anything up, without adding billing information. It becomes the AI layer inside SAP, where you can build your agent automations inside the product. We also have customers like Mercedes, who we’ll probably talk about, and how they’re using n8n and seeing the company transformed through it.
Those are the really exciting things. A lot of the numbers people are sharing are a lot of hype. What’s actually interesting is when you see real usage of a product, and especially real ROI. In the past, revenue was great because revenue was very close to real value. But a lot of decoupling has happened there. What our customers see very strongly is that we deliver real ROI for them. They’re not just burning money. They actually see something come back.
Another thing that’s important to call out about n8n, and why that’s happening, is the combination of AI with deterministic logic and human in the loop. AI is amazing and offers so many possibilities, but at the same time it’s not the solution for everything. What you really want is to link AI with deterministic logic, because a simple if statement is much cheaper, much faster, and 100% reliable. And then you still need human in the loop. I’ll demo later how that works inside n8n, because for certain use cases you always want to make sure there’s a human still in the loop.
That’s what people should look out for on LinkedIn. Not the numbers people share, but the real users behind them and the ROI that gets delivered for those customers.
What n8n does that a Claude Code agent can’t (14:44)
Aakash: So I asked Claude to go through my newsletter archive, and it said I had written about n8n 25 times in the last 12 months, which surprised even me. It also found that when I did my AI tool ranking a couple of months back, we put it in the A tier, above Zapier and Make.
Now, my audience has been hearing me talk a lot about PM operating systems in Claude Code. What they typically do is open Claude Code, connect some MCP servers, and have a working agent in a couple of minutes. What does n8n do for that person that they can’t do themselves?
Jan: Sure. It’s probably easiest to simply show you how it works. This is the starting screen for all users. The first important thing to call out is what you see here, our AI assistant. The idea behind it goes back to what I said about giving everyone who uses a computer technical superpowers and lowering the entry barrier. That’s exactly what it’s doing. You can do the same thing you do in Claude Code here. You can just describe literally what you want built, and it’s going to build it for you.
I have something prepared, but before I show you how it got built, let me show you the outcome, to give you a better understanding of how n8n looks and feels.
Here you can see an AI assistant that works with your Google Calendar and your Gmail account. Generally, how n8n works is you have a trigger node, something that starts a workflow. In this case, you have an agent, and it gets a response. You also have different AI models. It’s important to call out that here we use Claude Sonnet by default, and it runs via our own gateway, so you don’t have to sign up for separate accounts if you don’t want to. If you want to, you can obviously still use your own. We even have a fallback model, because we know these providers are not always very reliable, so it’s always good to have one in case they don’t respond.
Then you have a lot of different tools. In this case, list emails. You can get a single email, you can write emails, get information from your calendar, and so on. Here you can also see further actions, like sending an email. That’s something I don’t want the AI model to do by itself without asking me. So we have this human-in-the-loop step, where you can define that a tool only executes if you got approval first. You can define how it should be approved and what it displays to you.
Let me give you a fast demo of how it works. You can use it externally, or you can test inside of n8n. So I can just ask, what is my next call? And now you can also see the auditability piece. You’re not just seeing how the whole workflow was built. You can also see how it actually executes. You can see the agent ran. It used a model. It called certain tools here. It also had a memory. And you can go through literally each step and see what information got sent in and what information came out of it. And you can see my next call is tomorrow, Thursday, at 11:00 a.m. to 11:30.
Then you can send off, for example, please create another meeting with my VP of sales tomorrow at 3:00 p.m. And now you see each node executing again. If something goes wrong, you’d see it in red, and you can very easily debug and say, I’ll fix that one thing. You can see what it’s actually doing and how it’s executing. And now you can see that if you go over here, it’s waiting. It’s not doing anything further. It says, I’m about to create a calendar event with certain information, am I actually allowed to do it? I can either cancel it or say add event. If I say add event, it goes in, creates the event, and it’s done. That’s generally how n8n works.
Obviously it looks quite technical now, and non-technical people probably just want to know how they can actually create that. Again, that’s where the AI assistant comes in, and I can show you how it was built. Here you can see the prompt, generate me a workflow agent for a personal project. You can see what it was supposed to do. Get emails, write an email, search for events, and so on. It gets prompted similarly to how you would prompt Claude Code or other solutions. Then it starts thinking and asks some clarifying questions, like which model should be used. It thought for eight minutes, again very similar to Claude Code, and after eight minutes it came back with what it built. And you can see here, that’s literally the workflow we were just working on. It also gives you more information about the tools it used and where it used the approval step.
Then you can make additional adjustments, and maybe that’s something else to demo here. Let’s say, that’s great, but I very often want to schedule one-on-ones with people on my team. So I’ll tell it, please extend that workflow. Make sure I can schedule these one-on-one meetings. They should be 30 minutes long. Always add a Google Meet link. The agenda should always be in the description, and I should only pass what’s required. And it should get information from Google Contacts.
Now it’s going to keep thinking for a little bit and come back to us. In the meantime, let’s go back up here, just to see that while it’s thinking, you can still interact with the workflow. You can still ask questions and keep working, and it keeps working in the background. You can go back anytime and see what it’s doing while it’s figuring things out. It’s probably going to take a few minutes. I’m not sure we have enough time. But at the end, what you’ll see is that it adds additional tools to that workflow. It will add another tool to get contacts from Google Contacts, and another tool to add specific calendar events for my one-on-ones.
I think that shows the power of n8n, especially if you imagine that I can very easily hand this over to somebody else. Even if you’re not technical, you can go in here and very easily understand it. Here’s an agent. You can see the system prompt that’s been defined. You can see what model it uses and how the tools are defined. You can see it creates a draft of an email. You can see what gets auto-defined. All of that is simply impossible in code, because it’s just too much output. Or if it’s done 100% by AI, it’s literally a black box. It can do things for you, but you have no guarantee that the thing that worked yesterday is still going to work tomorrow.
That’s why we’re great for anything where compliance is important or where anything is really business-critical for you. You can really rely on those workflows because of the combination of AI with deterministic logic and human-in-the-loop steps.
What reliability actually means in n8n (22:46)
Aakash: So I want to make sure I define those key terms for everybody watching. Starting with reliability. One element you just talked about and showed us is that Claude barely has four nines of uptime right now, so you can put a backstop model in place. If Claude isn’t available, use this other model. Is there any other component of reliability people need to understand?
Jan: There are probably multiple ones. Another one to call out is that n8n runs on your own infrastructure. You can literally self-host it. So if you want it close to your own data, where no internet problems can cause issues, you can run it there.
Also, all the tools, like this get email tool, are literally code we created, tested, and maintain. So even if the API changes, all we have to do is change the code once, and it gets changed for everybody. It’s a reliable piece of tooling that works for everybody, not something every person worldwide has to reimplement each time, forgetting certain edge cases. We can think very deeply about the quality there.
On top of that, there’s the whole platform. You don’t just create workflows, you also deploy them on it. If something goes wrong, you have retries. You know where the data gets stored, and so on. It takes out a lot of the thoughts you normally have to have when you deploy something yourself or get code written for you by another solution.
What auditability actually means in n8n (24:38)
Aakash: Another word we talked about was auditability. The way I’m seeing it, although I’m curious whether this is technically accurate, is that when you build these things in Claude Code, the fundamental unit, the source of truth, is code. It’s a bunch of files living in some Python script and some random markdown. It’s kind of built on a brittle house. Whereas here, the way I experience the product as a power user, the core is this workflow editor, and everything is thought of in terms of more stable components. Each of these lines and nodes in the workflow is something you really put your foot behind as stable. It is working, and it’s something somebody can see, versus just code.
Jan: Exactly, especially the auditability piece. There are multiple things to it. Next to seeing what’s been built, the workflow itself, you can also see how it ran in the past. Here are the last executions we went through, for example. You can see literally step by step how it actually ran. It started here. It called an agent. It called those different tools. You can see the information that went in and out of the product. You can literally identify everything the AI really did for you.
Again, with code you see the input and the output of the whole thing, but what really happened in there, you have no idea. Why did it take certain decisions? For example, we could add an additional node here, between the output of the agent and the output, with deterministic tendency logic that says, if it outputs something smaller than 50 or larger than 50, do something very different. You can be 100% sure it’s always doing this. And afterwards you can inspect why it took a certain path, because that number was the output, so it went there. If it didn’t go to a certain thing, you know exactly how to debug it and change it. Then you can literally rerun half of the workflow from that data point, from that point in time, and get it to the state and quality you actually require.
How to hand an n8n workflow to a teammate (29:40)
Aakash: Another thing you mentioned that’s important about n8n is this idea of being able to pass it to somebody. Can you show us, if I wanted to give this to somebody, let’s say I’m a manager and I want somebody on my team to also have this workflow, how would I do that?
Jan: There are a few pieces there. You can literally share your workflow with other people. You just invite them, and they can access it and change it themselves. You also have version history, so you can see how it changed over time. The person can go through and see who made changes and what went on there. You can even publish certain versions and describe what changed between them, similar to Git.
Another important one is that you can very easily export them, so you can literally send another person a workflow via email if you want to. And something that’s not in this version yet, because I think it’s an older version, is review steps. You can say, I made this change, please review that workflow and see if it does the right thing for you. Then you can approve it, and only then does it get published.
The use cases n8n is uniquely built for (30:58)
Aakash: Okay. What is making enterprises like the model companies themselves use it? Here we’ve shown the email and productivity assistant. I feel like other products could also do this. What are the use cases that n8n is so uniquely equipped for?
Jan: The more important reliability and security are, the more it shines. We have quite a few companies using it for use cases like security orchestration. Every time an email arrives with an attachment, scan it, then archive it. If something is going wrong, inform certain people or make sure something gets started. Or anything that relies on more sensitive information. It could be employment data, or even things like onboarding and offboarding employees, where you really don’t want anything to go wrong. You don’t want a half-onboarded employee, and even less a half-offboarded employee.
We have people using us for DevOps use cases. There’s a wide variety. Anything where you want to be 100% sure that it really gets done, and not just half done. So more or less anything that is not a private use case. That’s very often why people normally choose n8n over other solutions.
What PMs should build in n8n after their first agent (32:21)
Aakash: I feel like you have so much power that sometimes it’s almost hard for people to figure out, at least when I talk to them about it, what the specific thing is that they should go do next. We’ve showed them the email and productivity assistant. We’ve given them a preview that it’s amazing for security and compliance use cases. If you’re a product manager and you’ve built the email productivity assistant, what’s the second, third, or fourth thing you should build in n8n?
Jan: That’s probably more a question for the person, because you can do everything in n8n. You want to do the thing that’s really most impactful for you. In the end, you want people to think about what they’re doing literally every day. Where are you using multiple applications? Where are you very often copy-pasting things between applications? Maybe you get information from one place, put it into ChatGPT or Claude, and then do something with that. Create a PDF report. Anything where you think, I feel like I’m wasting time. That’s where you can get started for your personal use cases.
For company-wide use cases, we normally suggest people start with something small. We have a lot of companies that want to start with something huge, like, how can I totally transform X. Then they spend literally weeks building and building and building. That’s a very hard thing, because at some point, if you don’t have experience yet, you may build things the wrong way. What we’ve seen is that sometimes workflows that literally contain an agent and a few other nodes are actually very often the most impactful ones.
And again, you don’t always need AI for everything. AI is one tool of very many. We have one company that literally used n8n to do password resets for employees. They have a lot of employees, many of whom forget their passwords regularly, and they were locked out for literally hours. They cut that time down immensely, and it saves them literally multiple full-time employees a year in time saved.
So it’s not about the craziest one. It’s about where you can get started easily and what you think is the most impactful one, and then building your way up from there. The simple ones especially are great to show off to other people. Look what I built, see what’s possible. I spent literally two hours on that, and that’s the impact. That gets other people involved. And the interesting thing, especially in an org, is that the more you talk with other people who actually build, the more ideas you get. Ah, I built this. Yeah, that’s a very similar use case I have as well, and you adopt it.
On top of that, we have our template library online. There are over 10,000 workflows in there. You can go in and say, I’m using tools X, Y, and Z, or, I’m in sales, and for each use case you literally see hundreds of workflows and agents that other people have built. You can either use them to get inspiration, or literally take them as they are, add your own credentials, and get started.
Inside the n8n template library (35:42)
Aakash: Can you show us the template library and maybe give us some inside information about some of the popular ones and the ones PMs should be looking out for?
Jan: Yeah. On the templates page, as I mentioned before, you can get started by saying, I want to use Google Sheets. In this case, for example, you see 4,000 workflows that use Google Sheets. We can then say, I’m in sales, and now you see all the sales use cases that use a Google Sheet. What we see very often is a lot of people using n8n for web scraping. Get information from web pages, get information from different data sources, enrich it in a certain way, save it to a Google Sheet in this case, and then send that information to Slack.
Or we can remove everything here and go to, for example, IT ops. The nice thing about n8n, going back to the reliability piece, is that if something goes wrong, you can get informed in whatever way you want. You can get informed by email, and in this case, here’s an example of getting informed via Telegram. Here we see somebody who used n8n for monitoring. They track whether the SSL certificates on their server are still up to date, and it alerts them on Discord. It sends it also to Notion. There are a million use cases out there that you can use it for.
We definitely see a lot of monitoring use cases in there as well, very often in combination with AI, because that’s where you want to react very fast and where the reliability piece matters. You can say, by default just inform me, and then maybe try to auto-fix it with AI. You can very clearly define the different steps that should be taken first, and make sure, for example, that the informing piece always happens 100% of the time, while the auto-fixing part with AI only happens later. That’s again one of the things where n8n really shines. I’d advise anybody to look through there and get some inspiration from what people are actually building with it.
Checking the AI assistant’s extended workflow (38:27)
Aakash: Yes, a lot of scraping use cases. I’ve also used n8n a lot to teach AI. It’s a really good place where you can very easily fine-tune a model, set up a RAG system, or set up a vector database. So if you want to learn the fundamentals of AI, I think it’s really good for that too, outside of just workflows.
So, one of the areas we talked about was ease of use. Can we see how the AI assistant has done on extending our personal productivity workflow?
Jan: Okay, here we see the prompt and how it thought about it, and then it extended it here. It also asked a question in between. Now we can go into the workflow and see it on the side. Let me extend it a bit more. Now you can see how it extended it. What it actually added is, for example, this lookup, so it can now look up contacts on Google. You can also see it actually executed, because part of what it’s doing is already trying to test the workflow for you and run it.
You can also see this approval for the one-on-one. If you want to book a one-on-one, it says the same thing again. I’m about to book a 30-minute one-on-one. It will include a Google Meet link. Are you okay if it gets added to the Google Calendar? And only then does it do that.
Aakash: Okay. So it took five or six minutes, it looks like, but in the end you get an updated workflow that actually makes sense and works with your tools.
Jan: Exactly.
n8n vs Zapier (39:42)
Aakash: So your episode is coming right after Wade Foster, CEO and founder of Zapier. So we have to ask the obligatory question. Zapier versus n8n. What can n8n do that Zapier can’t?
Jan: The great thing about n8n is that we focused from the very beginning on power and flexibility. You get the most value out of it when you really don’t want to be locked into an automation. You built something today, and as I mentioned before, you want to start with something simple and then build on top of it and build it up. That’s where n8n shines very strongly, because we have that power and flexibility built in. You have things like code nodes, where people can fall back to code any time. You can build very complex agents, not just basic ones. You can literally add your own memory. You can add output parsers to make sure the output is always in a certain format. As I mentioned before, you can have different models and fallback models. We have the human-in-the-loop steps.
A lot of that matters when things become a bit more complex rather than simple use cases. Wade is a great guy. I always enjoy talking to him. Zapier is a great product. But as I mentioned before, there’s always the right use case for the right product. Especially where you need power and flexibility and a very deep integration with AI, that’s when n8n really, really shines. And obviously, probably worth mentioning, if you want to self-host it, that’s where Zapier can’t compete, considering theirs is a SaaS solution.
Aakash: That’s an important differentiator too.
Why sprinkling AI on top only gets you 10 to 30% (41:44)
Aakash: So I want to learn more about your growth, because you have a pretty fascinating history. We talked about how many times people have called you obsolete. You actually existed before ChatGPT. n8n has been around, and I think that might have been one of the first times. You talked on a podcast about how you were scared a little bit. But sprinkling AI on top would only give you 10 to 30% growth, not the 10x growth you’ve been seeing year after year since. How did you make that call? And can you break that down for us a bit more? What does it mean to sprinkle AI on top versus make it core?
Jan: Sprinkling AI on top is what I see when somebody tells you to add AI, and you add an AI button somewhere that does something with AI. What we did instead was really think about how we could become part of the value chain. How can we not just add AI to the product, but make sure people actually build with n8n? The idea was that when people think about building an agent, they should want to build that agent with n8n. So we’re actually seeing that they get real value out there.
When people get asked to add AI to the product or build an AI agent, we want to be the solution that helps them do that, because that’s where the real value lies. Otherwise you see 10% or 30% growth. Last year we literally grew 10x, exactly for that reason, because we were part of the value chain, we empowered people, and we provided real value for them.
Where 80% AI agent adoption came from (43:29)
Aakash: I think a pretty crazy stat you released is that 80% of workflows on n8n now use AI agents. Obviously that would have been zero at the beginning of 2023. So is that your 2022 users adopting AI? Is this a new crowd? If so, what happened to the old crowd?
Jan: It’s actually people adopting. As I mentioned before, our users are tinkerers. They’re very interested in checking out the latest tech and really making their day-to-day more efficient. And obviously AI is the tool to do that. All of them were super interested in adopting AI and making it part of their daily usage. That’s how we became so successful, because we made it super simple to get people started with AI, much faster and easier than other solutions out there.
Why n8n killed its lead gen target and per-seat pricing (44:23)
Aakash: And how you’ve grown this thing is very different. The average company doing this type of growth would embrace per-seat pricing. They would have lead gen targets. You killed your lead gen target and you refused per-seat pricing. Why?
Jan: The thing is, we’re in this very lucky situation where we can think very long term. Most AI companies out there may grow ARR very strongly, but at the same time they’re losing a lot of money. So you have to show a lot of growth to get more money from investors. That forces them to think very short term. How can I get more ARR now to get more money tomorrow? And then the whole cycle repeats.
Because of the way n8n is built, we’re actually sustainable. Right now we’re actually creating a profit. So we don’t rely on more investor money anytime soon, or at all if we keep doing what we’re doing right now. Instead, we can think about what sets us up for success in the long term. Right now, that means capturing a lot of the usage, and not thinking about how to increase revenues as fast as possible, but how to create the best product out there, how to create a lot of value for our customers, and how to capture as much of the opportunity out there as possible. That’s a much easier decision for us than for many other AI companies right now.
A billion users with fewer than a thousand employees (45:49)
Aakash: So the lead gen target and per-seat pricing are pretty public for anybody who has studied n8n, like me. What might be some other metrics you’ve deleted, or a metric you’re going to delete, that people don’t know about?
Jan: Metrics we deleted. Maybe I can talk about an internal thing we’re focusing on very strongly. One part is what we want to achieve in certain goals regarding users and AI in the long term. The other part is how we want to build the company and how n8n should function.
We set this internal goal. We want to reach a billion users with fewer than a thousand employees. We don’t want to grow our headcount. We obviously grow it because we have to, but it’s the opposite of what we’re interested in. We want to build a very efficient org. The idea behind it is very simple. Organizations come to us to help them transform the organization, become more efficient, use AI better, automate more, and so on. And if you’re not doing it internally, you’re literally hypocrites. That’s horrible. I hate nothing more.
Especially, I think we cannot do a great job. We can only help them transform if we actually transformed internally as well, learned our own lessons, and know what worked, what didn’t work, what was impactful, and what wasn’t. For that reason, we’re not talking about how fast we’ve grown our headcount over the last years, because there’s literally nothing we think is a positive thing about that right now. We’re doing the opposite. We try to stay as lean as possible and build the organization of the future, to again help our customers and our users do exactly the same thing.
Maybe another thing to point out is what we’re not caring as much about. As a freely available product, you obviously have users that pay you and users that don’t pay you. We could always focus on, for example, how to get people who currently don’t pay us, who use our free version and self-host, onto our hosted solution, where we actually earn revenue. That’s nothing we’re doing at all, because it doesn’t matter for us. All we care about is that they’re using n8n and that we’re generating value for them. At some point, either they or the organization they work for is going to want to pay us in the future. But we don’t care whether they’re paying us right now or not. All that matters to us is that they’re happy n8n users and they keep using us in the long term.
How many people actually work at n8n (48:19)
Aakash: Fascinating, for somebody who was VP of growth at Apollo.io, where I was just focused on free-to-paid conversion and free growth. That’s very interesting and a very different way to approach those free users than people might realize.
You mentioned the employees, and I actually wanted to talk about that. At some point you had said you want to be the first billion-dollar company under 500 employees. I think your careers page puts you at around 160 employees, but on LinkedIn it seems like a bunch of people want to associate themselves with n8n, and you have over a thousand people listed as working at n8n. So how many people really are working at n8n right now?
Jan: Currently we’re around 370 people in the org, and we’re hiring. We’re probably going to be around 500 by the end of the year. You’re right, the reason the number is so inflated is that we obviously have a lot of partners and agencies working with us, and they very often appear as employees.
Also, the other thing you mentioned. What we talked about right now is a billion users with fewer than a thousand employees. Our old goal was a billion in ARR with fewer than 500 employees. We changed it for two reasons.
First, we did much more on the enterprise side of things. We realized that as long as the buying side is not done by AI, we cannot do the selling side by AI either. It’s still a people game. People don’t sign a half-million-dollar contract without talking to anybody. So that’s going to stay very people-heavy for a long time.
The second thing was that we realized what matters in the end is fast adoption. The billion in ARR was not set because we said we want to make a lot of money. It was set as a measure of being a big, impactful organization. But at the same time it still confused people internally, or even externally, into saying, all it is is just about money. We learned to be very clear that it’s not about money, it’s about adoption. That’s why switching away from ARR to actually users made much more sense for us.
Quiet growth, strategy or personality (50:22)
Aakash: I have one more question about how you grow, because it really is different from everyone else. Let’s say Wispr. I think they’re a company I’d put in a similar category, AI, and they’re just taking off like crazy. They launched their notetaker, and their entire focus was viral growth. They even bought rickshaw ads in Delhi and things like that. You have been perceived as much more quiet. Is that a strategy? Is that your personality? What’s going on there?
Jan: I think it’s probably also part of my personality. I’m not the person that has to be literally anywhere. I’m more the introvert kind. These things don’t come naturally to me, and I’m not the person that really has to be anywhere on stage. So there’s definitely a component about me there.
But generally, as an org, we’re also a European organization, and I think we are probably more the under-promise, over-deliver kind of people. So we’re not really in the business of buying any kind of viral growth. We think more long term. We didn’t become viral because we paid people to talk about us. We became viral last year because we had people being excited about n8n and getting a lot of value out of it. They wanted to share it. We had people who created their business around n8n and generated their own money that way. That obviously makes them even more interested in talking about it, because it’s this nice win-win situation. The more they talk about it, the more visible they are, and the more visible they are, the more visible we are, and then they get more contracts. So it’s a nice piece there as well.
From the very beginning, we had this very strong community focus. I already talked about having an event literally every day somewhere in the world. That’s again the more long-term thinking. We know that investing in community gives you nothing tomorrow or next week. It takes many, many years. But once you have it and once it works, it’s amazing.
Even by now, most of the enterprise usage actually comes from our community. People literally use it privately, get a lot of value out of it, and say, I actually work for this huge org, and I have another use case there, so I can automate things there. Then they do the first internal use case, and then they get other people excited. We literally have communities inside large orgs that are over a thousand people large, that have their own internal community events, inside large organizations and inside the system integrators. I think that’s just amazing. It’s nothing you can do very fast, but this is the more long-term, sustainable thing that works out very well for us.
Aakash: Very European indeed. Although if you think about a Lovable or something, they are more viral. So I think there is also a component to your personality in how you have grown this thing.
How n8n structures its product team (53:18)
Aakash: So that’s how you’ve grown this thing, which is just amazing. I want to talk a little bit about product. You said 370 employees. How do you structure your product team?
Jan: Right now we have squads of between three and five engineers, one to two PMs, and one designer. We have a VP of product. Underneath we have a director, and then underneath that our PM teams. The engineers report to an engineering leader, and the designers to theirs. I think that works actually very well for us. We have this very small unit that can work very fast and efficiently and get things pushed out quite fast without a lot of the overhead.
When Jan hired his first head of product (54:05)
Aakash: When did you hire your first head of product, and how did you make that decision?
Jan: We probably hired him around April 2021. It was around the same time we raised our Series A. We had multiple engineers already. I was obviously totally involved in the product, but I also realized it’s important to have somebody who has the experience to lead product, who can spend more time on it and go deep on the problems.
I met him, and from the very beginning, as he mentioned in the past, it’s a product he would have loved to have thought of himself. He’s literally a productivity nerd. He has a shortcut for literally anything on his computer. How he does everything, from shopping to meetings, is super efficient. He’s probably one of the smartest people I know. It was very, very clear from the very beginning that he is the right person to lead a product company like n8n on the product side of things.
What Jan handed over and what he kept (55:07)
Aakash: So what did you hand him, and what did you refuse to hand over?
Jan: Originally I stayed very close. I was still very involved in the product development side of things. I literally still merged every PR, so I still reviewed everything that was happening there. But that obviously changed more and more over time. Very early on, I owned the roadmap. But we always talked about all the things that were happening, literally every single feature in the very beginning, and discussed how we thought they should work. As the product team grew more and more, I became more removed from a lot of things.
I still have check-ins with the design team, for example. We had this case where at some point they built a feature and made certain decisions that honestly went against what I thought was the right direction to go in. Then I realized I had probably been too far away from the product at that point in time, and I said, now I have to go back there.
Still to this day we have regular meetings every week with the design team where they show me what things they’re exploring and what they’re thinking about. I can give feedback there. They can hear what I’m thinking about things. I can get feedback very early on, and it gives me a better idea of where everything is going.
Apart from that, it’s still my baby. I just want to be very close to it and understand what’s going on there, and make sure it’s always on the right level. There are obviously certain things where it matters less, and other things where it’s worth going a little bit deeper and understanding exactly how you’re thinking about things and what direction it’s moving in.
How PMs should talk to the CEO (56:53)
Aakash: So when a PM is talking to you these days at the CEO level, what are the ways they should be talking to you? What are the mistakes they might make?
Jan: It has to be on the right level, where it really matters. Obviously it depends what they’re coming to me about. Is it about a certain direction they want to be going in? That’s exactly the right kind of level. We think we want to go in that direction, and we want to explore those things. Then very early on I can give feedback and say I think it’s right, or we can say very early on that it’s not the right direction, and I can give feedback there too, and we don’t waste time and resources.
But again, everything always happens very closely with David, because he’s the product leader. I trust him literally unconditionally. He’s just a super smart guy. But I think it’s always good to bounce ideas off.
Where it’s less helpful very often is when it’s just in the detail. I still like the design details, because that’s literally where I want to be in the detail, and I think that’s important for me. But the last thing any senior leader wants is to get involved in every small decision, because most of them don’t matter. If it’s about whether we put the button here or there, or any kind of smaller things that don’t make a substantial difference for the product experience or the opportunity, that’s probably not what you want to be talking about with any senior leader.
How technical an AI PM needs to be (58:20)
Aakash: 100%. So I went and found an opening for n8n. I wanted to see what you’re looking for in your PM, and a couple of things stood out to me. Obviously it’s a very empowered job. You own the strategy and roadmap. But where I wanted to go was what you require. Some of the things you talk about here, two to three years on platform. Well, that’s because it’s a platform role. You talk a lot about technical depth. Some of the words we were just defining earlier in this podcast, reliability and scalability trade-offs, holding your own in an architecture discussion. It seems like a very technical role. How technical do you have to be to be a successful PM at n8n, or more broadly an AI PM these days?
Jan: The experience we’ve had is that the more technical the better, and honestly that’s true for almost every role we’ve hired. Our designers are very technical too. We also expect them to use Claude Code and get a good understanding there. The more technical the better, because you can have much more impact. They also understand more. Especially until recently, our user base was quite technical, so it was even more important then. And now, when we talk about power and flexibility, the thing we never want to give up is that power and flexibility. People should never feel locked in. Technical people are great for that, especially because they understand very deeply what’s possible and how much work something really takes.
Generally we always look for the rising stars. I think that has worked very well for us. We also look for really passionate builders. One of our values is literally a culture of, we are builders. We want people who are tinkerers. We have people who have home automation running at home and get very excited about those things, because again they are very close to our users. And also people who understand what it really means to build something that scales and something that is really reliable.
That’s why we want people who have not just built a basic agent, but who understand that you want to go deeper. You also want to think about evaluations. How does it scale? How do we improve it? How do we make sure it’s really reliable? If you find those traits in people, where they’re not just able to operate at a very high level but can also go deeper if they have to, that’s normally a very good sign for us, and those are good PMs. We’ve found some very amazing ones, but we’ve also always had a hard time finding the more technical PMs out there over the last years.
Who owns evals at n8n (1:01:11)
Aakash: I keep hearing this from AI CEOs, even at less technical products than yours, that they want these pretty technical PMs. One thing I don’t see on this list is evals. What do you think about evals? It’s one of the hottest topics in product. Should PMs be owning the evals, or at least defining the golden set? What is the role of evals?
Jan: We actually have our own team. It’s called the AI trust team, which owns them. They’ve literally also built an internal product around it as well, to actually make sure evals are run the right way. Especially for a complex product like n8n, where you want to test not just simple input and output, but see, for example, whether the right tools have been called. It becomes even more important not to just get something that works for half the people out there. We really need something very specific, so people have to understand evals very, very deeply.
I think people talk about evals a lot, but they don’t really use them as much as they should, because they’re probably not very fun to create. They’re very hard to create. There’s definitely still a lot of opportunity in that space as well, and honestly still a lot we could be doing much better at n8n. We also have an evals product in there as well, so you can create your own evals in n8n. I think they’re very valuable. But honestly, I would love it if more people would actually ask for them and create more evals.
When you talk about being able to run things reliably, evals are the way to actually ensure that. Especially if you want to switch a model, for example, or for whatever reason you want faster throughput or a better price. You need to get a good understanding very fast of whether the performance is the same, or how you can change the prompts very easily so it maybe works for those models as well.
How to get hired as a PM at n8n (1:03:13)
Aakash: A lot of people who are watching this podcast are seeing this role and thinking, I want this role. So if they want this role, I have a two-part question. How do they get noticed by n8n? What are you looking for to interview people? And then how are you interviewing people?
Jan: Getting noticed. I’m not directly involved at that level, but what’s definitely always helpful is if you see people being active in the space, maybe having built something like that already, and understanding what it actually means. Especially since we talked about them having to be technical, we can literally very easily see that by them having built something, or just by seeing how they present things and how they think through them. That’s definitely very helpful. And the second part, sorry. The first thing was how they get noticed.
Aakash: And then how do they succeed in the interviews? There are so many different types of AI PM interviews these days. What are you running? How are you separating the pretenders, who just use Claude Code to ship some product on GitHub but aren’t actually technical? How are you differentiating the PM in the interview process who could actually succeed here?
Jan: We found that it’s about them really being able to go deeper, and not just feeling like they’re repeating things they heard online without really understanding what it actually means. That happens a lot, and you can catch it just by asking the right questions. We’re very lucky in the sense that a lot of the things that actually become problematic at a certain point, we’ve also experienced internally already. So we know what kind of questions we can ask and what kind of answers to expect from a person who actually thought about a problem more deeply, versus the person who just repeats what they heard, and it seems like it sounds good, but actually there’s nothing behind it.
Even then, with AI, having a task for each role becomes more and more important. We’ve definitely seen in the past that with a lot of tasks, especially take-home tasks, you obviously never know whether AI was involved or other people were involved. Even pre-AI, we had cases where other people simply helped out. So something like a live session, where you just go through a problem and try to see how you would work through the problem together, actually tells us much more.
We’ve also had problems in the past where you obviously have people who worked in the same space before, and people who never heard about a problem. It’s very hard to understand sometimes whether a person who already worked on a problem before is actually at the right quality and the right level, or whether they just did better than people who never thought about it because they experienced it before. That’s just a very hard thing, honestly. Very often you still have to work with them for a month or two to really understand whether they really meet the bar.
I mentioned before that having this goal of reaching a billion users with fewer than a thousand employees obviously makes it very clear that talent density is really important. If you have only a thousand people at a certain scale, you can only have the best ones. Really ensuring that the quality is as high as possible obviously always means you have to keep the hiring bar high. But again, if something doesn’t work out, I think it’s best for both sides to let them go, because it’s going to be bad for the people to leave them in the role when they’re not set up for success. It doesn’t mean that they’re bad in any way, but they’re not the right fit for the role we need them in, in the organization.
Aakash: Wow, what a wide-ranging conversation. We covered n8n’s growth, n8n versus Claude Code and Zapier. We showed you how to use n8n to create a personal productivity assistant and get other use cases from templates, as well as how they grew. Jan, thank you for all of the alpha you dropped today.
Jan: Thank you for having me. I really enjoyed the conversation. See you.
Aakash: All right. And if people want to find you online, where should they go?
Jan: Honestly, I’m only active on LinkedIn. I left X. So only on LinkedIn.
Aakash: All right, find him on LinkedIn, obviously. Go try n8n if you haven’t. I have other videos that go deeper on n8n, which I will link right up here, and see you in the next episode.