How Nate Herk's AI Agent Is Revolutionizing Lead Response Times [With Human In The Loop]

Let's do this. Hey, it's Max, the original Flowgrammer, and this is the studio, The show where I share the stories of Flowgrammers, in traditional programming. When a function runs it's deterministic. That means if you have the same input, you'll always expect the same output. an LLM is also kind of like a function. It expects an input that gives an output. The difference with L LMS is that they're probabilistic by nature. That means for the same input, you're gonna expect a bell curve of responses. Some of those responses are gonna be desirable and some are gonna fall into the undesirable range. if you've got a background in statistics and probabilities, you're probably gonna guess what's happening next, but. If I start chaining multiple LM steps together, and this one's let's say 75% likely to be effective, and this one's 80% the resulting probability of success is lower than either 75 or 80. it would be 0. 75 multiplied by 0. 8. Now, I probably should have calculated that before I started filming, but it's lower. some of the most powerful things that we can get LLMs or AI agents, or AI generally to do for are multi-step processes. that's why today we're talking about hittle human in the loop. What is that? It's a fancy way for saying The human can give feedback on the previous AI step, And so if we have a trusted party at the helm of that step, we could assume a hundred percent probability for that step. If you take into account that human step could also correct mistakes from the previous AI step, you can essentially reset that probability of success and so basically, if you implement human in the loop thoughtfully into multi-step automated processes, you can get a far more reliable experience. and. Put a lot more use cases into production today. And so this is a perfect segue, which, which I obviously pre-planned, to an amazing interview I had with the esteemed Nate Herk. Nate is a power user in the end-to-end community. Once again, he's a professional automator. Anyone that watches a show, you know, I really like interviewing professional automators ' cause all the things they show me are validated by paid invoices. And he showed me a human in the loop use case that a lot of his clients are actually asking for. and it's basically an agentic workflow that can respond to new leads with the initial message. if you're in sales, you know this very well, but. What's really important when you get a hot new lead in is the time to lead. That is how quickly you reply to that lead. There's even some statistics around the fact that if you reply within minutes sometimes, or within the first hour, it can greatly increase how likely someone is to reply, which greatly increases the chance that they, uh, give you money at some point, right? So obviously people care about that. and Nate's gonna explain a lot better than I, but I think the really cool thing is how he shows a pattern where human in the loop can give feedback and then it loops back to the review agent. and he can do this n number of times, right? So theoretically he could give a hundred feedbacks. But I think the really cool thing here is that it's not that complicated of a pattern. The reason I wanna show it to you is if you've never seen this pattern in NN it might take you a while to figure that out. or maybe you didn't even realize that it's possible in NN. this is a proof of life, for a really nice, human in the loop pattern in NN He's using it with Gmail, but again, we got him in the loop for a lot of different things. I'm sure there's gonna be a little screen showing all the different human in the loop things we have. and pro tip. If you need something kind of custom N N's, wait node has a wait for webhook response. So you could actually create a kind of human in the loop or third party system in the loop, with whatever you're seeing today. So it's not limited to the native integrations you see in N eight N today in any case. As I mentioned, probably like two minutes ago, Nate's gonna explain this a lot better, so let's hear from the man himself.

Hey, Nate, how's it going, mate? Hey, Max. Great to see you. I'm doing well. Very glad to hear. I'm so excited to have you on the show. I was looking through my list of guests I booked. I was like, whoa, how haven't I had Nate on yet? one of the, I would say stellar contributors in the internet community. So yeah, welcome to the show. So happy to have you. Awesome. very excited to be here. so Nate, I hear you've got a really cool use case to show us and something that your clients are actually asking you a lot about, but before we jump into that, for the folks that don't know you, would you mind giving quick introduction? Absolutely. Yeah. So my name's Nate. I have been playing around with end to end for about six months now. as soon as I found out about it, I just jumped into it. But from there, launched an AI agency where we're delivering solutions to clients with code tools, primarily end to end, of course. And I'm also doing YouTube videos where I show demos of how this stuff works. My main goal is to make the barrier to entry very low for anyone that doesn't come from a technical background. then I also run some school communities where I'm teaching people how to build this stuff if they are running an agency. Very cool. You know, I get a lot of DMs, as you can imagine, people asking, Hey, look, I'm learning and, and who could I learn from? And you are always one of those guys. I'm like, look, do yourself a favor, go watch some of Nate's videos. I think what underscores your videos, a lot of the folks I have on the show is, you're working with clients, you're seeing real problems to get paid, you have to create a real solution. Right. and so then you can kind of export those learnings and those insights and into the content that you make. you see it in the work that you have, which I think is a perfect segue, to the use case that you're going to show me today. could you just explain to folks what you're showing to me today and why you built this all stems from two things. The first one is common themes that we're seeing from clients. And then the second one was when you guys dropped the human in the loop notes, which is just awesome. One of the biggest things that we're seeing from our clients, real problems that they want solved are sort of sales processes, whether that means, generating quotes for clients or generating quotes from leads or writing sales briefs for their teams internally. But the idea here is just. the importance of the speed to lead. this workflow that we're looking over is utilizing both of those elements of, let's say we're getting a new form submission on a website. We're having a sales agent look over that form, look over past projects, and then write an initial email, personalized email, for that. From There we're getting feedback on the email from the human, because we obviously want to make sure that whatever we're sending out to our hot leads is going to be a high quality email. So then we're able to provide feedback and we can make as many unlimited revisions as we want. until we're finally good to go and say, Hey. you can send that off. And then the agent will take care of the rest from there. Very cool, and I think this is a great example of human in the loop, elite coming in. This is a hand raiser, they've literally said, Hey, I want your service. This is, like in sales, a very important, sort of individual to get back to fast and Nate, why is speed to lead important for sales folks? for those of us who aren't sort of in the sales profession, when people are looking for some sort of service, they usually don't go just for one business. They're going to shop around a little bit and they're looking for people that can get back quick. have results. ultimately if they send out five, Form submissions on the day, whoever gets back to them first might be the one they end up giving their business to. So speed delete is super important, but not just speed, but also the quality and the responses. So you need to make sure you can trust that what your agents are sending off is actually, relevant to your business and something that you can live up to. makes a ton of sense. And I think that's the perfect segue. to show us this, the solution. So would you mind sharing your screen and perhaps running other solutions so we can get the context of this and then walking us through your workflow. Absolutely. Yeah. So let's do it. Here's the workflow we're looking at today, and you may notice it doesn't even look too complicated. It's just a total of about seven nodes. that's really cool because we can obviously scale it up and customize it based on whatever the client's looking for. If this was a client bill, anyways, we have a lead form submission which is triggering the workflow. so we're just throwing that in our table. So I'll hop into this real quick so we can see an example lead, Which would be the name, company, what they're looking for, their budget, a product description, and a timeline once that form gets submitted, we'll hit run and we'll see that it's going to trigger the sales agent, which is going to look within our project database to say okay, what projects have we done that are similar to this lead And Then it's going to be writing a very personalized, and flavored email Now what we have is it's finished up that email and it sent that off to us for human feedback, as you can see right here. if we hop over to our email, give it a refresh, we can see, we got a new email right here, which is approval required for a new lead, Robert, California. So in here we get a quick description of what this lead is looking for. Customized solution budget of 10 to 15 K and an immediate timeline of one to two weeks, And then it asks us to provide feedback on the message that's generated by Jim, our AI sales. agent. So what I'm going to do before we read this email is just scroll to the bottom and hit respond. We're just going to pull up that new window where we're able to actually give our feedback. And So we can see we have the subject of our email and then we can actually read it. I'll go high level over what this email is saying. Thank you for reaching out to us with your inquiry about a custom AI powered customer support and sales optimization system for Sabre. It's now going to be looking at the requirements and this is the aspect of personalizing it and saying, Hey, we've done some other things that are similar. this is going to increase the lead's confidence that we are the right choice. for you. So we can see here that we have developed an AI driven system that handles omni channel So you can see here that it says based on your requirements, I'm thrilled to share that we've recently completed a similar project for an e commerce client. We developed an AI driven system that handles omni channel interactions, Integrating seamlessly with multiple platforms, blah, blah. And then we can see some actual, proof. And it also resulted in a 23 percent reduction in response time and 142 percent increase in sales conversion. just once again, giving more confidence to the lead And then obviously it's a sales agent. So it's going to end with some sort of call to action which is basically saying hey based on your timeline and your budget. We know we can do this for you. Would You like to book in another call? So at this point we're able to actually provide feedback And I don't know about you, Max, but this email seems a little too wordy for me. So what I'm going to do is first of all, just ask it to make it more concise, and then from there we can give it another read, and see if there's any more specific feedback we want to give, but I'm going to start off by just saying, make it more. concise. you gave me, so you've already got me. I'm yeah, that's a great point. I'm sure he doesn't want to read through a chunk of text. He wants to get some bullet points that he can scan through quickly. I'm going to hit submit, and then we're going to hop back into the workflow and you can see that now it's continuing. It has hit the revision agent, which is looking at the original email that was written by the sales agent. And it's looking at the feedback that we gave it. And you can see it's already waiting on us again for more feedback. So I'm going to hop back into the email, give it a refresh, and we'll see we have another approval required. We will quickly give it a. scan, we can see that it's already more concise, but let's click on respond to open it up in a new window in order to provide more feedback. It keeps all of the same content. So it keeps the personalization, It keeps the actual results. And now it has a new email that's a bit shorter for us. Let's say in this case, we want to say, Hey, Let's not mention the reduction in response time. And let's also say, let's meet on Thursday. So that's exactly what I'm going to say in my feedback. I'm going to say, don't mention the 23 percent response time, I need to make sure this is actually a percent sign. And let's see if he wants to meet on Thursday. So this is the feedback that we're now giving because we maybe checked our calendar and we realized we want to meet on Thursday with him. So let's submit, go back into the workflow and we can see once again, it's checking for feedback. and then it sent it off to the revision agent and their vision agent looking at the most recent version of email. not the original one that we had sent off. So it already knows to make more concise and then it takes that email that was The second version of email and then implements. our feedback once again. I think the cool thing as well is, that feedback message you wrote is the someone in sales writing, in between meetings or like on the go, real quick, just two, three words go. And then, that unblocks that next revision, maybe in between the next meeting, they can take a quick peek. It's good. And they can send it, it can slide into the busy workflows of the folks that need to do this work. Exactly. So we're still kicking that can down the road while keeping that personalization touch in there. so now back in the email, we can see we have another approval required and hopefully this is the last one we have, but we will once again, click into respond and we'll take a look at the actual email. the first thing that we'll notice is that when we get to the section about here's what we did for a previous client, it no longer includes that 23 percent reduction in response time that was previously here. It only is mentioning the sales conversion increase. We also have in the call to action, we believe, we can craft a tailor made solution that meets your timeline and budget while exceeding expectations. Would you like to schedule a call this Thursday to discuss the vision and deliverables in detail? Which makes this a little more, actionable for the lead, rather than just leaving it open for them to maybe provide some days. This basically just helps us say, okay, Thursday, let's get it done. So now at this point, let's say we're good to go. And We're ready for Jim to send this off. We will just say, Good to go. Hit submit, switch back into the workflow, and it should go up into the approved branch. And now we have that final email sent off. we can just quickly check back into our email, and make sure that it did get sent off. And here it is. It's a lot more concise. easy to read. and We have a clear call to action at the end, which is scheduling a call on Thursday. there's a series of example projects, but, we see this at end right now, as the sales teams grow, we're getting so much of this data, existing proposals that people did, they sometimes take a long time for someone to thoughtfully create. and you may not know as a new person in the sales team of some proposal from nine months ago from some different team that is relevant here, But an example like this, you can still leverage that investment that your colleague made nine months ago a different, sales squad and use it to help win, This project and have that like collective brainpower of the whole organization behind you, Very cool exactly. So maybe your use case isn't that you want an agent to write your outreach email. Maybe you just want it to do research on past projects like you mentioned. Based on a huge vector database or whatever it is. And then it can say, Hey, here's the lead. And then on a silver platter, here's all the research that you need in order to write them a really high converting email. Absolutely. I think another really cool observation here is you're sending the email directly, but it's a trivial change to say, Hey, in our organization, we want to have it send a draft first. And then the human still going to send the button, another human in the loop, all very trivial tweaks to do the workflow that you have running there. but Nate, I think it'd be really interesting for. folks, if you could walk us through how this workflow works, especially things like system prompts. I see you've got a text classifier there. it's a really neat pattern, in human in the loop. could you switch to the workflow and just basically walk us through what happened that execution? absolutely. So we can see that this workflow is being triggered by our air table form. that's capturing the data that we saw earlier about Robert California, his company and what they're looking to do. So first I'll open up the sales agent and we can see that in the user message, what we're giving it is basically all the information from the lead form. So we're giving the name, the email, intent, the budget, the company name, and the product description, as well as the timeline. So this is the stuff that uses to search through that product database and make this email. very personalized. From here we'll take a look at the actual system message, which I tried to keep as brief as possible. always like to start off by giving an overview. So I said, you're an expert salesperson for an agency that delivers AI solutions. Your job is to respond to incoming leads by adjusting their needs in a professional manner. You will receive information like the leads, product description, timeline, et cetera, and your end goal is to convince them that we are the best AI agency on the market. So this really sets the role and makes the agent understand what it should be doing. From there, I like to give it an idea of the tools it has access to. In this case, it's only one. which is called Projects database, and it uses this tool to search through previous projects that we have done. From there, I just gave it three main rules, which is obviously it's going to receive information, because it's really important to have your agents understand the context of what they're getting and then what they're trying to do. We told it its main objective is to convince a lead to book in a second call. Your job is to make them believe that we can deliver the project they've described and exceed their expectations. And in order to do that, it needs to retrieve information about previous projects to share with the lead to prove that our team is capable of handling their project. So I said, find a similar project we have done, to share with the lead, and share the result of how we helped the client. final notes, sign off as Jim, from Dunder AI, and then I gave it the current date and time, just in case. then you can see that we're actually giving it access to that database, where it's going to be searching through. in this case, I have three projects in here as you can see just to prove that it's going to look through and actually grab one that's relevant based on the lead form product description and then as you can see over here, we have it connected up to a GPT for a model as well as a structured output parser, where we're saying, hey, you need to output two things, a subject and an email. That way it's easier to map into the human feedback node later. okay. So you've got your revision agent and that first agent. Having the same basically structured output parser so that you can assume the same schema between these loops, because you got that loop going there. is that why? Exactly. Yes. And you don't have to hook them up to the same one, but just to ensure consistency, we want both of them to output a subject and an email. from there, what we move into is a really important step that may be overlooked, which is setting the email. And I didn't actually have this in here until I had multiple failures. And so the idea here is that we have to set the email that's coming in from either the sales agent or the revision agent. so we can reference this node later. Otherwise, when we're trying to make revisions, it's only going to be looking at the first initial draft, so it won't know how to make multiple revisions on top of each other. Similarly with the actual send final email node, let's say we had the revision agent make three revisions, but then this. email node is looking for the output of the sales agent. There'd be no point in making these revisions if the final one that's going to get sent off is just the first email anyways. So this is a really, really important step, This makes sense. and see this across the internet and community is when you have this loop, most of the time when you have a loop, if there's data that you rely on within that loop, there is typically this strategy. We're using edit field and snowed like this to sort of set Some dependable schema that you know you'll be able to access no matter if you're going from like your initial case or your loop case or whatever it might be. a hundred percent. And if that didn't make sense yet, once we look at like the revision agent or the email node, it will. anyways, from there, the next step is to send it off to the human for approval. So if I click into this node, we can see that what's going on is. We have basically hard coded in the subject line, which is always going to say approval required in all caps. So we can get the attention and then we'll list the actual name of the new lead. So it's a nice, short subjects, but we know what's going on. And then in the message, what we're doing is we're mapping in fields from the actual air table trigger. So if I just closed down all the way to air table sugar, it was just simple as dragging in. Okay. The name from this company is looking for this intent with this budget. And I think you get the point there. then we were able to really quickly scan and get a full feel for what this leads looking for And then just a message about, okay, hey, here's our subject. Here's our email. Could you please provide feedback? for the folks at home where that set node was coming in that set email node the top left here, as you can see, when we're referencing the air table trigger, we're using an absolute reference to that node, Because it's the first node. but by the time we get to the email at the bottom here, we're using dollar Jason. This is a relative reference. and that's the crux of using the set node. between these two different loops, because it doesn't matter if it comes from like the first agent or from the, the looped agent, we're always referencing that set node with relative notation, i. e. dollar Jason. And so it works irrespective of if it's coming from that first case or from, the first revision, the hundredth revision, this is gonna work all day. That's exactly right. And it's a great point to bring up. Whenever you see dollars on Jason, you know that you'll be looking for whatever came immediately before the know that you're looking in. cool. So the last thing I want to touch on within this node is something that I was really excited by, because by default, this is going to be an approval message where you add an option where you can basically have a button that says approve, or maybe you can have a button. that says approve and one that says deny. But if you come in here, you can see that you can also just do free text, which is what we're doing. here. that gives us the aspect to actually type. in feedback. And then as you can see on the right hand side of the output, we're going to see our actual feedback. if you remember the demo, on the first run, we said make it more concise. On the second run, we said don't mention the 23 percent response time. And then finally, we said good to go. now we have actual free text feedback that can be a lot more specific. Rather than just saying, yes or no. after we get the feedback, we're moving on to a really cool Text classifier node, which is basically a switch node and an AI node wrapped into one because it can create these different branches based on the categories that we give it to classify. what we're doing in here is we are classifying the actual feedback that came back from us, the human. As you can see right here, we're looking at good to go and we're using the JSON because we're referencing the previous node. In this case, we only gave it two categories, so it's going to be looking at this text and saying, Okay, is this an approved message, or is this a declines message? in each of these, you have to give a brief description of what that looks like. in here, you can see for approval, it says the email has been reviewed and accepted as is. the human explicitly or implicitly expresses approval, indicating that no changes are needed. And then I went ahead and gave it some example phrases of something that could be an approval message. And then we do the exact same thing for a declines message, where we're saying the email has been reviewed, but it needs edits or modifications, removing parts, stuff like that. And then, once again, I gave it some example phrases that could be considered a denial message. And this is really important because if a text comes in and it Can't understand is this approved or declined, it will just output nothing and then the automation will not continue. So being very specific about these descriptions is how you can make sure to improve the consistency and the quality of this text classifier note. And once again, if we look at the runs, we can see run one, run two, went to the decline branch, and then finally run three, was good to go, so it got approved. once an item gets declined, it's going to go to the revision agent. and this agent's goal is just to obviously make revisions. So first of all, let's take a look at the user message. all we're giving the agent here is. Most recent email, and this is where we're referencing that set email node, as you can see, because this makes sure that whatever was the most recent version is what's getting passed along. So it's getting the most recent email, as well as recent feedback that just came back from us, the human. it's taking these two elements, and then it uses its system prompt to create a new one, as you can see the system prompt is super, super concise. All we're saying is you're an expert email writer. Your job is to take an incoming email and revise it based on the feedback that the human submitted, and then the exact same final rules which was sign off as Jim. and here's the current date and time. as you were explaining this, I remember you saying, Hey, look, this is gonna be a relatively simple flow, and obviously there's some complexity here, but I love the elegance of some of these system prompts. I mean. We got, what, a couple sentences here, and as we saw live, those revisions were working and were improving with each sort of human in the loop, run. Very cool. a hundred percent. And something else to take into consideration too, is you also have the control of switching over the output models. So let's say maybe you want a super powerful reasoning model to write that first initial draft, and then you could have one that's a little less powerful, maybe a little cheaper to make the revision. you can customize all of that stuff too. I think it was an interesting one. Noticing here that using like Google Gemini Flash, for the text classification, because I think it's a much easier job, write a compelling thing that will convince the human to buy something versus classify, approve, deny, I think a lot of people in the space right now, are enamored by a really neat model. and perhaps in dev, it's fine to use like an expensive model, but Just because a flamethrower will light a candle doesn't mean in prod you should be using a flamethrower most likely, right? That's a great analogy. Yeah. I love that. Definitely important to understand the strengths and weaknesses of different models and when to plug and play different aspects. we've come out of the revision agent. We've got the revision draft It's looping back around into the set email node And then basically it's the loop again, It'll send another human feedback. You'll provide feedback or not we go through that text classifier, and then we ultimately send the email once you approve. could you show us that step just for completeness for everyone? Yeah, of course. So Once again, in here, the really important thing to do is to reference the actual email. We want to make sure we're sending off the most recent version. So that's, once again, the importance of setting the email over here to make sure that we can grab that from there and send it off to the lead. So the only other things that we needed to map here The actual, who is this going to, and we're able to get that from the original air table trigger that started this whole workflow to make sure the emails are going to the actual lead who submitted the form. And then of course, our subject In this workflow, we have one subject. We're not having the actual subjects be revised on. we're grabbing this from the sales agent output, which was that first agent that created the email for us. Very cool. I could imagine, especially, for a lot of sales teams, sometimes someone posts something on LinkedIn and there's a bit of a spike and you get 50 inbounds it's precisely in amazing moments like that. I give me for a lot of business owners waking up to 50 inbound leads. it's an amazing opportunity, but at the same time, if you reply slowly to all of those, you'll miss out on that. It's probably like really frustrating as well. being able go faster in this way, or I can imagine as well, with all the information you have in that, lead, form, it's very easy to segment this as well, right? If someone comes in for a quarter million dollar ticket value, you might have, maybe that goes, to your senior. Guy to quickly hand write that the top priority. But then for all your, small and medium ones, you're using, this process. But either way you have that flexibility. I'm curious, Nate, have you got this set up for some clients already? Yeah, we definitely have stuff like this, very similar where lead comes in. We're having them search through full database of previous projects, because a lot of things that take time are quote generation or writing those briefs and doing that research about the actual lead. one thing to keep in mind here is how do we build systems that help you get 80 or 90 percent of the way there? Because we always still want that human element, especially in something as important as sales. Now, Nate, off camera, you were showing me some other workflows that are basically utilizing this in a really neat way. Would you mind, giving us a little sneak peek on that? Yeah, for sure. So obviously once I played around with this workflow and I saw the power of these human in the loop nodes that could pause your workflows, I wanted to experiment. And one video that I did that did really well was personal assistant that had access to an email agent, a calendar agent, all this other type of stuff. one thing that I was seeing feedback on that video was I'd love to make sure that my email agent One piece of feedback I was seeing on that video is that We'd love to make sure that we're actually having the right contacts or making sure that we're not scheduling over events. So what I decided to do was play around with a human in the loop calendar agent. And so this is a very similar workflow as You can see. We have the aspect of an original agent up front, We set the intent, we get the human in the loop, and then we have a text classifier node to check if the feedback was approved or declined. and just doing that unlimited loop, until we finally get an approval message from the human. this one's pretty cool because it's gonna look through your contact database and say, hey, there's three emails for this person, which one do you want to send the invite to? And it's also gonna say, Hey, You asked to make an event at 5 p. m., but you already have something there. Should we push that back an hour or what do you want to do? And then finally, it gets pushed all the way up to the actual agent with the tools to take action for you. I love this example because it's, I think a lot of newbies when they're jumping in, think it's totally fine also to be a little naive when you're trying new tools, When you're building, you test and you learn to iterate. But here separating different specific jobs. AI steps are very often probabilistic, The best way to control those outcomes is to do one specific thing. Well, break it down step by step. here, that first, agent, it doesn't have access to creating events. That's not its job, Its job is to do that first step of the process, and then by the time we get to the calendar agent, that calendar agent can focus on just, crudding calendar events in Google very well, and that's all it does, but it doesn't have the context of all the other things that have to happen in this workflow. and if it did, it's likely to perform that less well on average, Based on how these systems work. Yeah, I think that's a super important point. And one of the first things we try to do when we're, architecting out and wireframing out what an agent system is going to look like is how can we break everything down? a task into subtasks and give each subtask to an agent so that it only has to focus on one very specific thing. And we found a lot more consistency with the outputs and just higher quality outputs in general. Nate, I think this one's going to be a really powerful inductive learning tool for folks like learning how to do these. Or how to implement these patterns and strategies in their own workflows. Is this going to be available somewhere where people can basically download this and inspect it on their own? Yeah. 100%. And that's what's super cool about these templates is that you can get a good skeleton, but then from there you want to understand how it works and see how you can fit it more towards your own needs. But yeah, all of workflows that I show on my YouTube channel are all available for free download. All you have to do is join my free school community and then you'll be able to access all of those JSON files. We're going to make sure to run a little clip here so everyone knows where to go find that, but do go join this community. I know there's a lot of folks who have paid communities, which, they're providing a lot of value, but Nate, as a really just, stellar, community member, offers his open for free. so if you want to learn more about edit and, go check that out. go check out this template and a bunch of other ones that Nate has. taking something like this, it reduces some of the initial effort to build. Then being able to tweak it and modify it, apply it to your own use case. It makes it more relevant, increases motivation. and from my own experience, that's a great way to learn if you're getting new into this. And perhaps building is from scratch seems a little intimidating. Starting with something that's already 80 percent of the way there that you can tweak again, find some relevant value and basically go down the rabbit hole is a great way to learn. Yeah, I couldn't agree more. When I first started learning, I pretty much went on to NADN templates just grabbed a couple, hit run, and I would just node by node click in to see the input and the output and understand the configuration. And I definitely think it's a great place to start. Absolutely. Well, Nate, I'm curious, since you had some time with the human in the loop features, are relatively new. Do you have any feedback for them? Or perhaps put otherwise, if you had one magic wish to change the human in the loop functionality, what would it be? And I'll make sure to get that to the end team. Yeah, I think my only thing right now is just whether you're doing it on Slack or Telegram or Gmail, like we saw in this demo, being able to click respond and then have it not open up a new window. So within something like Telegram, you can just type back to it. I think that would be super powerful, more seamless, and just improve that, user experience. It makes a ton of sense, and I promise to get that to the right squad as soon as we're out of this call. I'm gonna DM the PM that's working on that one. but Nate, I just want to say. Thank you so much for your time. I know you're a busy guy. I got your YouTube channel, you got your school community, you got your agency. thanks so much for taking some time to come share this with all the flow grammars out there. I mean, if people like what you're doing, I know you've got your school community, but if you could just list off all the different ways where people can follow your awesome journey and all the great stuff you and your team are working on, First of all, I just want to say thank you for having me. It's definitely an honor to be here and it's always great to get a chance to chat with you, Max. I've got the YouTube channel, Nate Hurk. I've got LinkedIn. You can follow me, Nate Herkelman. And then my agency is called True Horizon AI. So you can go ahead and check that out if you're interested in maybe learning more about how we can actually help you out and your business. Nate, again, thank you so much. And thank you for everything that you do in the NNN community. I think you're a great example of a pretty typical NNN community member, though. You do a lot of good for the community. You're always happy to share and you're just nice and helpful, generally well rounded human. So thank you so much for everything you do in the community beyond this video. And please, next time you got a hot flow to show me, DM away. We jump on another call and I'd love to have you back anytime. Yeah, that sounds great, Max. I'm always eager to jump on a call with you and get to show you some stuff I've worked on. Cause it obviously always gets me fired up. That's fantastic. Well on that, Nate, I'm going to let you get back to the flow. So happy flowgramming and I'll catch you next time, Nate. Awesome. Appreciate it. Thanks, Max.

Thanks so much, nay. Again, I always appreciate talking with you. Such a nice, humble guy. and that signature smile. again, I think one of the really cool things there is complexity to it. There is some thoughtful design choices on how to build it. But again, I think to a lot of us that looks obtainable, that looks repeatable. That looks adaptable. and I highly, highly recommend to all your folks doing gen AI content posts online. maybe outta human in the loop step, you might save humanity and save us all from those really crappy ai generated posts. I myself am still writing my posts humanly. and my comments as well, if you ever get a comment from me, within under a minute, when you app mentioned me, it's not AI guys. I've gotten a few accusations for that. Some people even try to prompt, inject, hack my, my inbox. but yeah, keeping it all human right now, and for a lot of you, I. if you're in e-commerce, you've got 10,000 products. I understand that you're gonna explore this, but for a lot of you guys over the next years, there's going to be so much ai, lop, and content. Be human and maybe use AI to speed up your process, but add human in the loop. Add your curative lens, add that human couple percent of context any case, I'm Max. This is the studio. You are really awesome for watching this video. Thank you very much, and happy Floor Ming