Ilan (00:00)
David, should

we pivot and do a fitness podcast instead?

David (00:02)
What?

Ilan (00:05)
Two

guys in their ⁓ late youth, maybe very early middle age, talk about getting who are not in the best shape, talk about getting in shape.

David (00:13)
Mm-hmm.

Yeah,

exactly. Let's do away with dad bods,

Swole in circumstance.

Ilan (00:21)
Mm-hmm.

Hey everyone, welcome to prompt and circumstance. I'm Ilan

David (00:26)
And I'm David.

Ilan (00:28)
And today we're talking about N8N.

Alright, guten tag David. How are you?

David (00:46)
Yeah,

doing all right. I'm very curious about N8N when it comes to AI agents. There's just all these different ideas that I've got. And, you know, of course we hear so much in the market about AI agents and it's all the rage. ⁓ So you did quite a deep dive into N8N. I'm very curious to learn about, what you've learned.

Ilan (01:16)
going to take us all the way to Germany for today with Jan Oberhauser, who is the founder of N8n, who was a person who spent years working in visual effects or VFX. And in 2019, he decided that he wanted to build tools for the people who worked in this industry.

And the real problem that he noted was that there were workflow automation tools, like say Zapier, that were targeted at the non-technical user, or you could write code. And his thesis or his argument was that there should be something that marries these two. A workflow automation tool that allows you to quickly

build but that allows you to run code and write code and where you can actually read the code underneath the hood. ⁓ So if you're a developer, you know, can precisely tweak it ⁓ or you can have a step that runs a specific code block

David (02:11)
Hmm.

Yeah. know, lot of startups, know, the, the, the way that you would define a startup is a disagreement with the market. Right. So he disagreed with the fact that there's this ⁓ big dichotomy, I guess you can call it, right. Where there's a, there's a gap in between the two extremes. Yeah.

Ilan (02:31)
Mm-hmm.

Exactly.

So he started N8n and he adopted a fair code policy,

It's basically a happy medium between open source and completely closed code. So the fair code licensing model basically says that you can read the underlying source code. However, you cannot replicate it. You can't fork it into your own product.

David (03:04)
Mm-hmm.

Ilan (03:10)
So if you want to use that code, you have to license it from the original creator.

David (03:17)
That sounds fair to me. I'm like the other unfair code bases.

Ilan (03:23)
Now is a great time to tell you about our sponsor for this week, Querio. Querio is a data analysis tool which sits on top of your data stack and allows you to ask natural language questions about your product data. It gives you beautiful visuals and allows to really pull out threads so that you can get to the bottom of all of your product questions.

You can try it too. Just go to querio.ai and let them know David and Ilan sent you for two months free of Explore. That's a thousand dollar value for our listeners.

Ilan (03:58)
So N8n has been one of these companies on a meteoric rise since the sort of AI revolution about eight to 10 months of 2020, they raised a seed round from people like Sequoia, one of the big US-based VCs for 1.5 million US. In 2021, they raised a series A.

David (04:16)
Hmm.

Ilan (04:22)
$12 million on a sort of a hundred million dollar valuation. In March of this year, they raised their series B, which was led in Europe. It was a 55 million euro round on a $300 million, a 300 million euro valuation. And they're currently in talks for a series C round.

David (04:40)
Hmm.

Mm-hmm.

Ilan (04:51)
led by Excel for a pre-money valuation of $2.3 billion.

David (04:58)
That's an enormous

jump from March was it the previous round? Yeah, wow.

Ilan (05:02)
That's right, almost a 10x jump from

March till August, August, September of this year.

David (05:09)
That's the power of AI agents, I suppose, right? Like the potential of what people can do and what people have demonstrated that they can do on N8N.

Ilan (05:18)
That's right. And as I said, N8N really was in the right place, right time, right industry.

where because of this medium code,

ethos that they had taken. They had the ability for somebody to link into an LLM to create an agent, pass prompts, get response, really tweak the code as well as set up all of the kind of loops handling or evaluation metrics that are really vital to being able to turn

your little workflow automation into something production scale and can really run an entire production level agentic workflow in the background.

David (06:00)
Mm-hmm.

It sounds like it's the Goldilocks between Zapier and writing your own code. I'd imagine that there's already agents in production using it.

Ilan (06:16)
Absolutely. I do want to talk a little bit about their pricing and, how they've managed to grow so much. They specifically communities and

The fact that they share their code means they have a ton of people checking them out on GitHub. They, it's one of the things they flag at the top of their website and they have something like 55,000 people in their developer community

David (06:38)
Mm-hmm.

Ilan (06:49)
And their pricing is pretty fair. So they actually just revamped their pricing a couple of weeks before we're recording this. So in late August, early September, levels of plans, a starter, a pro and a business. And for each of these, the pricing is based on how many executions of your workflow automations are you doing per month? So essentially their pricing based on

the value they're generating.

David (07:18)
I like that. I like that very much. I think we're seeing a bit more of that where the pricing strategy is tying the price to the value delivered and scaling according to that.

Ilan (07:33)
That's right.

David (07:35)
And I think important to point out here is that there is also a self-hosted option, right? So if you want, you could do everything right on N8n on their cloud. Alternatively, you can self-host N8n.

Ilan (07:49)
Yeah, absolutely. Now I will say that one of the

limitations that I learned while going through the process of building my own workflow in N8n is that some features are not available in the self-hosted model.

So when you first go into N8n and create an account, you create an instance. So that's sort of the primary block that you're working out of. That's where all of your workflows live. And exactly, it's your tenant. So we're going to open up our prompt and circumstance instance.

David (08:19)
That's your tenant, basically.

Ilan (08:28)
So when you come into your instance in N8n, you see the workflows that you have built. And this is really the heart of N8n. You also on the side panel have the ability to add projects, which allows you to group those workflows into some higher level idea. But really the workflow is what you are creating when you come into N8n.

David (08:55)
All right, yeah. And so how does a workflow relate to an AI agent?

Ilan (09:02)
Yeah, it's a good question. So ⁓ we can click around into some of their pre-built agents here and we can take a look at the concepts. So let's grab their task management agent and use that as an example.

So a workflow for N8n always starts with a trigger.

that kicks off the agentic workflow. The concept of a workflow in N8n has that trigger and then multiple nodes after the trigger. And one of the nodes available is an agent node. And that agent then has instructions, a system prompt, as well as certain

models it has access to, memory it has access to, as well as tools so it can perform certain actions.

David (09:55)
Got it. Yeah. So, ⁓ when we are defining the agent here, when we're, ⁓ configuring it, we can say, look, I, I want it to use, you know, Claude 3.7,

Ilan (10:06)
All right, so if you listened to us last week, you heard me talk about Lindy AI and my frustrations with that tool. And what I wanted to build there was a workflow that when a new episode of our podcast is created in this tool we use called Transistor, then create a draft LinkedIn post and email it to me. So that's the same workflow I built in N8N Give me an apples to apples comparison.

David (10:33)
Got it. Yeah. So eliminate the effort needed create a post on social media with the LinkedIn being the start.

Ilan (10:40)
That's right.

That's right. So it's going to eliminate that step, eliminate that

I will say here, and eight N is pretty nerdy. You know, if you can imagine this is the balance between a workflow automation tool and coding. If Lindy AI is the jock in class, it's kind of like.

cool and I've got this, but you know, once you start digging in, you realize that maybe they're an inch deep and a mile wide. N8N is the nerd in the front of class who's able to answer every question. And when you go to them asking if they can help you out with your homework, they're going to push up their glasses and sit you down and get you way deeper than you thought you needed to be on this topic.

David (11:27)
Wow, I love to learn.

Ilan (11:29)
So let's talk about this workflow. N8n does not have as many pre-built connectors as Lindy So you have to kick off your workflows with something like a webhook.

And all that really means is that what an action happens in some other application, it will notify.

N8n at this particular URL It did not take many tries to get this to start functioning.

And then if it finds the transcript, then great, it goes into the rest of the flow. If it doesn't find a transcript, then I have it wait again a little bit longer so that ⁓ we have time to upload the transcript.

David (12:07)
Got it. Got it. You know, it makes me think about the joke where, you know, somebody goes and does something and says, Hey, look, I'm going to go, I'm going go do this thing. And if I'm not back in five minutes, wait longer.

Ilan (12:17)
Exactly.

right. So after the transcript is uploaded, I basically have one node, which will get the I have a code cleanup step, which is one of the really cool nodes that's available in N8n. So you have the ability to put in some specific JavaScript or Python code, and it will just run that code on whatever your inputs are.

David (12:44)
I like that there's an ask AI tab there too, where you can say, Hey, could you write some code for me?

Ilan (12:46)
So that's

Yeah, so that's what I was going to show next. This JavaScript code that you saw there, I didn't actually write that. I don't know how to write code. I just went in and asked AI to say, hey, I want to have some code that replaces these specific fields that come in from the transcript and ⁓ gives me the result after that.

David (13:12)
I like that.

Ilan (13:13)
Then have our old post as well as some additional guidance on the voice and language that we want to use in a file in Google Drive. So it pulls down the drive file. It extracts the text from that file. And then we have our little agentic step where we go into a model with a

system prompt that tells it what we want to do.

David (13:41)
I like that you can inject context from the current execution into that prompt.

Ilan (13:46)
That's right.

Then we have it create a file in drive in a folder where we keep our LinkedIn post drafts and, ⁓ build a link to that file. And finally, email us letting us know that, Hey, you're

draft LinkedIn post is ready and here's the link to go to.

David (14:06)
That's really cool. So what did it take to get some of the integrations set up with Google Drive and Gmail? it just, here's the API key and away you go.

Ilan (14:17)
Gmail, it was pretty simple. It was login to your Gmail account to give it access

David (14:24)
Okay, so it's one of those things where when you click on the link, it'll send you over to Gmail where Gmail will ask, Hey, N8n is asking for permissions to do this and that. Are you okay with that? And you just say, yep, sure.

Ilan (14:39)
Now on the Google Drive side, this is where it got a little nerdy. What I had to do was create an API project in Google and I had to go through a ton of steps.

to give permissions and create authorization and add my user and all of this stuff. Now I will say that I didn't know how to do any of this and documentation walked me through step by step.

you can run through a test flow so you can execute your workflow and for example, create a, a, ⁓ an episode, a draft episode in our case and upload a transcript, just to make sure that your workflow is working. And so, ⁓ that's really helpful. And it went through all the steps and, know, a couple of times I found mistakes that I had to go back and, and tweak.

⁓ One other helpful thing in here is that in your trigger, when you are testing, after you run the trigger the first time, you can pin the results so that you don't need to run your external trigger every time. It just has the results.

David (15:51)
which I'd imagine is very important for testing to speed that up.

Ilan (15:55)
Exactly. So workflow did take me probably two hours end to end to And a lot of that was the drive connection steps where I needed learn about how to create an API in Google drive and make that at the end of the day, it worked and it works.

David (16:18)
Very cool. That is the beginning of how we got to be a two person unicorn.

All right, Ilan, thank you for walking us through n8n I learned a lot. It's very inspiring to see how something like that could be put together. It does get a little bit technical, but I think that ⁓ with the assistant, it's not going to be all that bad. gives me good confidence in being able to...

create and modify AI agents on N8n. What were some of the surprises or things that you learned along the way?

Ilan (16:53)
I have to say that the level of technical know-how that I had to get to was a bit of a surprise to me. I knew that N8n was this more developer focused tool, but came in with a lot of confidence as a fairly technical person that I'd be able to do it. And I did. I was able to get to my result, but there were certain steps where I would have thought they'd have a simple tool.

Like for example, for Google drive, I would have thought they'd have that simple signup flow where I just connect to my Google account and then can point to a drive folder. That said, ⁓ their documentation and their AI assistant are really delighters. They are super thorough. They give you correct answers.

David (17:40)
Very cool. So for those who might be considering n8n for the first time, what recommendations do you have for them?

Ilan (17:47)
I would say that if you're somebody who is fairly technical, you know, you've read an API doc and are comfortable with that. Maybe you've written a Python script before. I would say you should dive right into N8n. This is probably the tool for you and you'll be able to be super powering your workflows with agents in no

David (18:14)
And it looks like they have a free trial. Is that right?

Ilan (18:17)
That's right. So they do have a free trial. It's two weeks and it's a free trial of the pro tier. that's 2,500 executions. Again, unlimited users. So I would get an idea in mind of some workflow that you want to automate. Start with a trigger in mind, right? What would make you kick off this workflow and give it a try, see if it's useful to you. And 20 euros a month.

which is for us, you know, a little under $30 Canadian a month is not too bad considering the value that this can provide.

David (18:54)
I believe it. All right. Well, thanks again for walking us through that. ⁓ Hey to our listeners. ⁓ have you tried n8n? Love to hear what your experience has been like. otherwise, ⁓ don't forget to like and subscribe and, at the socials we are at PNC podcast. We'll catch you at the next one.

Ilan (19:15)
See you next time.