The AI Agent Builder Who Declared "SaaS Is Dead"
So there's this guy, Bora Celik. Software engineer, thirty years in the game. Last year he tells his investors something crazy: SaaS is dead. Not dying. Dead. And here's the thing - he might be right. See, while everyone's playing with ChatGPT, asking it questions, Bora's building these things called AI agents. They don't just answer questions. They do the work. Like, actually reach out to influencers, negotiate deals, send products, follow up. No humans involved. One of his clients, Harney & Sons Tea, they've got agents running their entire influencer program. The agents find people, check their engagement rates, send emails, track who posts. Everything. And the wildest part? The CEOs of these $100 million companies are building their own agents now. Using tools anyone can access. Today, Bora shows us exactly how.
Swym - Wishlists, Back in Stock alerts, & more
getswym.com/kurt
Cleverific - Smart order editing for Shopify
cleverific.com
Zipify - Build high-converting sales funnels
zipify.com/KURT
Apply for Shopify Help
ethercycle.com/apply
See Our Results
ethercycle.com/work
Free Newsletter
kurtelster.com
The Unofficial Shopify Podcast is hosted by Kurt Elster and explores the stories behind successful Shopify stores. Get actionable insights, practical strategies, and proven tactics from entrepreneurs who've built thriving ecommerce businesses.
Kurt Elster • 00:00.001
This episode is sponsored in part by Swim. Okay, here's a depressing stat. 70% of shoppers who want your products never actually buy them. They browse, they consider, then they forget. That's revenue walking out the door. Swim Wishless Plus turns browsers into buyers. Customers save products they want, get notified when prices drop, or items restock. You can also engage them in personalized fashion through your marketing or sales outreach. It's like having a personal shopper reminding them to come back and buy from you instead of your competitors. And 45,000 stores already use it, and it only takes five minutes to install. You could try it free today for 14 days. Go to get swim. com slash curt. That's swimwithay. com slash curt. Turn those maybe later into sales today. Get swim. com The thing that confuses me, my friends, uh, about AI, is what the hell? What what's the difference between AI and an AI agent? Right, I hear about agentic AI and and AI agents. I don't actually know what it means. I think it means you're giving it some level of autonomy, you know, just like, hey, run an automation, go out, do this, I guess. But As usual, I have someone who has experience here, who is smarter than me, who will explain it to us. My guest today is Boris Selleck. who does only agentic AI builds. He's actually building custom bots for clients, it seems. Well, we'll get into it. But I met him. through a mutual client. Harney and Sons T work with them for years, love them, and they have had Borin and his team assist with some really clever automation. And the one that we worked with one that was, you know, reporting um analytics in Slack, which was cool. I'm like, okay, I think I could figure that out. But where it got wild is product seeding. They had fully automated a product seeding campaign. You know where you like find an influencer, reach out to them. Hey, we love what you're doing. Could we just send you some product? You send them the product and then you know follow up and hope they post about it. And you know, for someone like a tea company, product seating is is very practical. And I was so impressed by this. And over, you know, the last several, several months, you know, through the course of the year, I've been reaching out to Bora with my occasional AI questions, because this guy is is deep in it, you know, with stuff I like I don't even get. You know, explain it to me. I'm like, I actually don't know what you just said. But hopefully we could we could dumb it down for for lay persons. Uh Boris Sellick, I'm so happy to have you. Thank you for joining me. How you doing today? Thanks for having me, Kurt. I'm doing well. How about you? Good. So all right, tell me uh what what your business is, what you do.
Bora Celik • 03:02.440
I'm the founder of a company named Agentic, uh, which is the pivoted version after collabs at IO, once I realized uh that AI agents are gonna take over and traditional SaaS businesses are just not going to um be exciting slash maybe survive even. And then uh last year, about 14 months ago, September of 2024, I sent an email to our investors and said, the first line said SAS is dead. Quite the claim. Everything that we're doing, all the collaboration campaigns uh that we're doing for brands. AI agents can't do a lot of this work. It's just uh there's a lot of tasks to do. So then I started uh kind of like converting all of our systems into more agentic systems to have AI agents do the work instead of humans logging into software and clicking buttons and looking at pages and things like that. Um and it ended up being uh an end-to-end autonomous system Um, and just the first one, which is the example that you gave about Hornian Sons, is uh finding uh influencers doing matches uh using vector search technology and the agents are doing this not people so like normally they employ people to go Do keyword searches, hashtag searches, try to find their emails, email them, reply to them, send them product, check to see if they posted. two dozen tasks that need to be completed until you find an influencer, make sure they are the right one, reach out, um, and then, you know, go back and forth. They always have a lot of questions, whether it's a gifting campaign or a paid campaign. Um, so that's how we started. Uh, and then every time you start working with a brand and they see what's possible now that wasn't possible, you know, a couple of years ago. They're like, oh, it's just like the light bulbs go off in their heads and they go, can we also have AI agents do this other thing? Our focus was mostly on marketing related. I mean it's still mostly on marketing related, influencer outreach related, influencer campaign related. But you kind of follow where the demand is, and then you know, we ended up building like a series of AI agent automations that are kind of e-commerce brand specific. Um that do I I kind of like uh break them into two separate uh categories. One of them is what I call the high volume agents that just do a lot of stuff, like one per minute, 24-7, 30 days a month. And the second one is what I call kind of like the data analyst agents, where it's not high volume, but it's high smarts. Um, so they kind of look at the data. and just do data analysis and then provide insights. So those are the kinds of two kinds of agents that we build. And you know, focused mostly on uh e-commerce operations right now.
Kurt Elster • 06:08.120
So I want to focus on, you know, because the thing that is a a gray area for me is you know the the format that an AI agent takes, it's sort of like what its limitations are, how it works. Currently, if I think of AI, it is synonymous with ChatGPT. I type into, I presented a text prompt. It gives me an output. And then, you know, like it's smart and clever and we've all used it. But it is not autonomous. You know, it's it's not and it certainly is not performing tasks or doing very few. Right. It like it works in a very limited way. It involves, you know, me sitting over its shoulder, monitoring it, and just kind of and and doing the inputs. You go a step further that really changes things where it is directly it is no longer me saying, Hey, this is what I'm trying to do and it goes, All right, well the go do this. And then go back, oh, this is what happened. And, you know, like it's no longer the middleman. it is itself doing the work, the AI. A lot of the work you do is in e-commerce. And then within that, it's kind of like there's analytics work, the stuff that in which it interac where it's only interacting internally and then stuff where it's interacting externally. That's the wild part. Where you're just letting it go out into the world and engage with people. And I think like influencer campaign management in general Seems like such a great use case for this. So let's like walking through that as an example, what does it look like? Like let's say I have in broad strokes, you know, if I've hired you, someone like you, or built something like this. What's the starting point? What's the output? How much human is involved here?
Bora Celik • 07:44.039
So the term AI agent is kind of being used a little bit loosely right now. Um, you know. But the thing is it it does things the way you program it. So let's for example, let's talk about this uh influencer outreach system. Being done by multiple agents, right? Agents, agents have tasks, and then when they finish the task, there's an output of that. The output could be the input of another agent. So let's uh talk about the influencer finding. So you start working with a brand like Harney Ans, for example. Um, so People who might be uh interested in teas or you know, tea kind of lifestyle. So, you know, we have like a brand insights agent that kind of does a research on the brand, what people are saying, what their products are. Uh based on that, it does automatic matches using uh the creators in the vector database that uh talk about certain things so it kind of matches the similarities. What the heck is a vector database? Uh a vector database stores text in vectors, which is like a bunch of numbers, like 1024 dimensions. And then it allows you to do similarity searches not based on text, but based on something called embedding. So it converts text into Depending on what dimensions you use, I use like 1024 dimensions, depending on what you do. So like every piece of text, let's say, a caption a creator wrote in their Instagram post. So that's a a paragraph of text. So you vectorize that, then uh let's say let's get their last 25 posts. 25 of them vectorized, and then we have Harney and Sun's brand vibes vectorized. It compares and it finds the most similar vectors, not based on, oh, this is talking about T or like for instance it knows that Tattoo and ink are the same things. It knows the meaning. It's not trying to match exact words. You see what I mean?
Kurt Elster • 09:45.519
So that makes it so I had no idea such a thing existed. Okay. Incredible. The vectorization is really clever, because like currently with a an AI search tool, like perplexity, it is literally going through the same human text that someone else might. your system, Vector DB, you're loading the data into it, I imagine, and then essentially like it assigns it a a more machine readable format.
Bora Celik • 10:11.200
Yeah, so I vectorize a lot of things. Once I learned how powerful it is, so vectorization of uh influencer data is one thing, vectorization of brand data is another thing So now uh what we're working on now vectorization of customer messages, vectorization of creator messages. So once we do like creator gifting campaign, creator's right back with feedback about the product. Say, oh, like I like this, I don't like this, the taste is like that, blah blah. So we vectorize all of those messages and then give a uh like a chat interface to the brands and you can actually ask questions to see what they are saying. Like what are they complaining about? What are the top five complaints? Or vectorization of customer reviews. So that allows you to kind of do searches based on the meanings, sentiments, and things like that. So uh once you have something like this running in the background and tie this as like uh tools that agents can use, uh that it allows with you know an autonomous system end-to-end.
Kurt Elster • 11:11.000
Okay. So how much I want to know how much human intervention is involved here? Yeah, how mu like if I s build the system and say go, could it run like obviously I can add checks into it where it's like I have someone has to approve a list. you know, a spreadsheet, whatever. But could this be set up to just run autonomously, continuously?
Bora Celik • 11:31.579
So what you're describing is a pattern called H-I-T-L, human in the loop pattern. So during certain points We start with introducing a human in the loop pattern where this is like a sensitive area like for instance sending an email, right? It writes the email. So let's say we found the creator. We did a match, created a match report, over 75 score. Great. We're gonna email this creator, find their email. And then personalize an email based on the stuff that they've been talking about and the what the brand is about. So it says, look, you know, we're a good match because of this, this, this, and this reasons, because you're talking about this, this, this, and our brand is about this, this. This email is created. So now before sending that email, we had a human in the loop pattern where a human looks at all these emails generated and clicks the send button for edit and send. So that's an intervention area. So what what we do is if you actually go through like 100 or 200 of these emails without changing anything, it builds trust. So one once we start building AI agents, the most important thing is people are like, uh, I don't know, like is this gonna hallucinate? Is it just gonna say weird stuff?
Kurt Elster • 12:48.560
Yeah, it's speaking on your behalf. So before you let it loose, you want to make sure it's not going to do something weird.
Bora Celik • 12:54.160
Yeah. So anything that we do, it needs to build trust, right? So Is it writing good emails that we can be proud of? It won't embarrass the brand. Or when it comes to things like data analysis, is the data accurate or is it making stuff up? Because, you know Uh if it's not engineered correctly, it can because LMs are known to do that. But there there are ways to prevent that from happening, right? So um After a while, people stop reviewing and saying, you know what, let's which I call re go from H-I-T-L mode to YOLO mode, where you say YOLO and just send the email. I don't have to review it anymore. Uh and then right now actually with you know uh high sons and in the beginning we used to review like just somebody would review hundreds of emails. And it's just it's just good. Like it sends tens of thousands of emails out and not a single one coming back saying that like why did you send me this weird thing? You know, instead people are like, oh, thanks so much for uh you know reviewing the stuff that I'm doing. I feel seen, you know, things like that. You know, like because normally A lot of brand brands do outreach with um templated email saying, hey, uh we love your content. I listened to podcast where you spoke with guest name. Yeah, I get a lot of those. It's just uh I actually we actually built some AI agents on the creator end, so I was able to see what their inbox looks like. Oh my god. It's just so ridiculous spam coming through from brands trying to get them to promote their products. And then it's hard to stand out. So uh another intervention area could be uh product sending, right? So for brands who have multiple products, flavors, colors. uh tastes or different things, sizes. You know, we work with a shoe company, you know, they have kids' shoes, different sizes, different colors. Creators ask questions, or like they're allergic. Oh, does it contain so and so? Uh if it if it does, I can't use that. Can you send me another one? Things like that, right? So th the these are the areas where uh people kind of still need to get involved. But what happens is over time you build this history of AI able to learn from what kind of questions are asked and what kind of answers are given. So if we get a similar question in the future, I can say this. So it kind of like learns, it increases its learnings about the interactions about that brand and the kinds of questions. I mean, you know, there's all kinds of questions, but even if you have two dozen questions, it's finite. It's not like unlimited number of questions coming in. Um so that that's actually currently still in a draft mode. AI writes draft replies and then somebody kind of hits send or edit and send button. But it just like still increases uh the speed so much.
Kurt Elster • 15:51.140
Even with like humans kind of eyeing it, editing maybe Well, there's so much research and legwork that's done and so much like mental overhead that goes away. It's very easy to look at a finished result, like an email. And you go yes, no, you know, send this or not.
Bora Celik • 16:07.620
Yeah, I mean uh replying to people on behalf of other brands, uh the there's there's a lot of work involved in there. Um and uh but it it does remove a lot of kind of um grunt work uh out of the plates of people. So people can mostly focus on more creative things or strategic things. So like AI agents are just doing a lot of boring tasks.
Kurt Elster • 16:30.019
Well that yeah, that's the stuff I want to get off my plate, right? The stuff I don't like. For sure. Like the things you do enjoy, don't outsource or automate those. If you enjoy them, keep doing them. Man, those order change emails suck. They're always like, I ordered the wrong size. I meant to change the shipping address. Oops, I was drunk. Instead of making customers email you that stuff. Let them fix it themselves with Cleverific. Cleverific is a self-service portal that lets customers edit their orders without bugging you. That means you get fewer headaches, customers get faster fulfillment, and there's fewer returns overall. Everybody wins. Peter Manning, New York. Cut their support tickets by 99% with Cleverific. You want in on that? Get 50% off the Pro Plan, just $49 a month. exclusively for listeners of the unofficial Shopify podcast. Go to Cleverific. com slash unofficial and use promo code KERT50. Done. Problem solved. That's Cleverific. I want to know what this interface looks like. You know, like chat GPT text interface. An agent, what does that interface look like? Especially for you, we're these are custom tools.
Bora Celik • 17:39.460
So there's no interface. I actually don't no longer believe in web interfaces. Like I stopped building any kind of web apps. Um so for instance, Hanya Sons, there's no interface for them to look at. There's nothing to log into. Uh they they just want a certain result like yesterday they messaged they they have like a new campaign, holiday campaign they want to get started beginning of December. And that's what's gonna happen, and that's just gonna happen. Asians are gonna do it. Do you just kinda instruct the agents to hey like you use this product instead of this one? Uh and find me also now uh new type of creators who are also interested in books and tea, not just tea, you know, because you know they have like a new collab going on related to uh a writer. And then um So interfaces, the agent building interfaces. There's the interface to build agents. You either code write code, you know. So for me, I use uh land graph cloud platform. uh for high volume agents. I also use N8N. I prototype a lot of things in N8N and this is something anybody can use. So I actually started almost like encouraging people to learn how to do it. I started making content. uh even like teaching people look you should do some of these things yourself. I believe that uh everybody should learn how to build agents, how to orchestrate agents. I think it's the job of the future. Um, you know, people are kind of afraid and worried they're gonna lose their jobs. Uh I believe what's gonna happen now is like the tasks are getting replaced. Right, the tasks. But when you kinda think about it, why were we doing those tasks in the first place? Is this a task for a human to do? Should I be doing something else? Right? So coordinating, orchestrating, building, architecting these agents, you don't have to be technical either. Well, you know, like you don't have to be a coder. Uh you kind of have to have a systems thinking mind. Uh and you have to be analytical, but there's stuff to learn, and I think people should learn this, you know. Uh and the people should learn this and build these agents, connect them to each other using probably easier to start with visual interface tools like N8N. Uh it's you know easy to understand. You connect things to each other. You know, a lot of people probably use already automations in HubSpot or Zapier, so it's similar. Or Clavio. Uh or Clavio.
Kurt Elster • 19:58.419
Shopify flow. Yeah, I'm looking at N8N and like the The editor, the way the templates look, immediately I'm like, oh, this is like Familiar, right? Yeah, familiar.
Bora Celik • 20:08.559
Uh it's just that There's like an AI agent you add to the flow. And once you add an AI agent, it it changes things because it becomes like prior to AI agent, everything is deterministic. Like a HubSpot flow would be Uh if this happens, then go and do this and send this email or move this deal from here to there. Else if the scenario is this. So like you basically program a flow just like you would program code, a lot of if else then kind of logic. Um well with the when there's an AI agent, you don't program it deterministically like that. You let the agent decide what to do. Yeah. So you still have to instruct the agent. you know, like what you expect to happen. Um and then there's like MCP tools you give the agent so that because agent itself doesn't know anything. Right? There's an agent like in in an A10 flow, it gets triggered via either a cron job that runs on a like hourly or once a day or every minute or every five minutes uh or if an email has been received or if I get a Slack message or something like that. Right? And then uh once it's triggered, agent receives some kind of input. uh saying oh like find new creator so like you know for hanya sounds right now every minute the creator matches running for the whole month like nonstop so every minute it finds like the next best match And then goes checks their Instagram, checks their new posts, score, oh, 43. Not good enough. Rejected. Next. You know, so uh it has like you give it prompts, just like you know, when you're using ChatGPT, you give it a prompt. In an AI agent, there's a user prompt and there's a system prompt. So in the user prompt, this is like every interaction comes with a new input. And the system prompt is kind of how we program what we expect from the agent. Like we say things like You know, go check this MCP tool first, get that result, and then go do this later, you know, and that agent kind of uses the model, like depending on how complex that flow is, I use a different model. F for something like influencer matching, uh the context window is high, like last 25 posts. Also a bunch of stuff from the brand, what they're about. I use O3, which is a reasoning model. It's slower, but it's smarter. It's uh it takes a long time, but it comes with a really good analysis But there's another one. For instance, I have one agentic flow where, oh, we found a creator, their location is unclear. So then agent goes and checks their posts and the comments and tries to decide if they're US based or not Oh, like there's a Jamaican flag in their profile or a location when they added a caption, the location is in Spain. Or there's coming that's come Spanish words. Some people are leaving a lot of Spanish comments, so probably. So then that one I use like 4. 1 mini. You know, it doesn't need like super reasoning. And it's cheaper because, you know, you do a lot of them. So you kind of have to figure out volume versus cost versus am I okay with an intern level result or like a junior result or do I need a PhD? level response here. So you kind of have to do the cost analysis. Uh what I like to call do I need a Ferrari for this job? Or is a BMW okay? Or is it okay to use a Toyota Corolla?
Kurt Elster • 23:49.260
I'll just take a Civic. It's fine. You know. Help me understand I want to understand cost at scale. Like of you know, of of all these campaigns or the agents that you've built. What's the one that has the high you know the the biggest output? Like deliverable wise, you know, what's got the biggest impact?
Bora Celik • 24:07.179
Well the the biggest one is the influencer matchmaker agent uh uses the O three and the the input is big and the output is big. W Uh and the second one is We have a video scoring agent that's right before an ad is created, you know, brands get these testimonial videos. Well, you know, if if the outreach is successful, hopefully creators generate a ton of content and the brands are swimming in Tons of content. Like one of our customers gets fifty thousand videos every month on TikTok.
Kurt Elster • 24:37.279
Ah, yeah, that's what I was looking for. 50,000?
Bora Celik • 24:40.159
Yeah, yeah. They've built a machine. Uh and it took them a while to build the machine. Uh and they're one of the highest uh revenue generating brands on uh TikTok shop. And uh so if you get 50,000 videos every month and they have actually a big team, like maybe a couple dozen people, just doing this influencer operations Uh but they could only look at five thousand of them. So they kind of leave on you know like ninety-ninety five percent uh on the table. Like they haven't been able to review them for two years. So they have like going back two years, hundreds of thousands of unreviewed videos that are sitting there that it could just content created for them.
Kurt Elster • 25:20.039
Um so the agent AI can review the video and make a subjective decision on good bad or like rate it one through ten. What's the criteria for that? So we even have like two layers.
Bora Celik • 25:31.179
So we we use like one uh model that does like a cheap review. Uh in this case it's uh Google's uh Gemini has a video understanding uh and that's pretty cheap, right? It allows us like you know if you use big open AIs, uh they have like a computer vision. image understanding. That's kind of expensive. Gemini's video understanding, that allows you, but it doesn't give you a lot of detail. Like it gives you like just enough one small paragraph of like what's happening in the video then it allows us to feel as possible. You just have to kind of Play around with it. You know, it's uh I it's a new thing that I learned in the past few months as well. So uh just models are changing all the time, Kurt, like new ones are coming out. You kind of have to play around and see Can I use this for the brand, for this job, for this task? Um, so then you automate it. So because we want to do so many That's running on like every minute it's reviewing two videos now because we have to also go back uh and then uh it reviews them And then there's the usage rights outreach agent that kind of uh looks at it, say, hey, like this reviewed over 50 score. Let's email this creator, ask for usage rights, hey, here's our contract. Give us the rights and you know we'll do like a Rev share kind of a deal. Um and then you get them to sign the agreement, you know, like I even kind of like work with their legal to remove that manual docu sign that they had in place, which kind of slows everything down. So the creators can automatically kind of click something and then it triggers the next agent. So that's a trigger right there. Because the creator agrees to usage rights, it triggers the next agent that does vectorizing. Now now they have rights to all of the videos this creator did for them. So now it opens up. Alright, so like let's add them to our video library. This is a video that was out there but we couldn't use it before because we didn't have usage rights. But now we have usage rights. So now it triggers the again scoring, but vectorizes scene by scene, frame by frame, visual and audio transcript. What is being said, what is being shown. uh breaks up into scenes so now it's making it ready for the creative department to kind of see can we use these scenes A woman putting uh skin uh cream or like uh eye cream under her eye or somebody talking about transformation, like what what is being said? You can do audio searches, but the agents can also do the searches, right? So like you can kind of write a uh hey like I need uh uh you can kind of storyboard it I need a scene that happens, something like this, but of course this is looking for hundreds of thousands of videos, millions of scenes, because every 30-second video has like five, six scenes in it. So just multiply that, right? Uh so then that analysis, there's another scoring there that kind of scores scene by scene. That's a little bit more expensive, but You know, at this point the volume of videos is down to thousands instead of hundreds of thousands. So and then we want to make sure that this is compliant. Meta is not going to reject it if you upload it as an ad. The you know, the person is not saying stuff that's uh not against FDC or something.
Kurt Elster • 28:52.140
Yeah, they're not making like unreasonable claims.
Bora Celik • 28:54.660
Yeah, you like completely you know fixed my cancer or something like that. You know, you can't say stuff like that. Yeah. So there's instructions about what cannot be said, but there's like it's either either like not compliant with AFTC, but also Just not a good ad, uh not a good thing to say. And then it also like makes recommendations. Hey, like remove this sentence, then it will be okay. So now the next level of agents is like doing edits. Like the interfa Yeah. And it's in Google Sheets, right? So the interface is like the agents are inserted into the Google Sheet the team has been using already. So there's no like web UI that you have to log into, but it's like, hey, this is the video. And then it says this is the final analysis. This is the score. It says, well, it can be used for an ad, but you need to remove this text that's kind of like not compliant. And then the person says, all right, just remove text. And then just giving it like like five or so text edits. Remove text, move this text to the front so it's a better hook. So it's good for hook testing, or increase volume by 30%, or speed it up by 15%, or add captions, like things like that. Like normally prior to the agent, the team has been using things like CapCut. Um, you know, they have to kind of like edit each video like one by one, then turn it into an ad, then upload the meta. But now they're kind of like doing it at scale, just the Google Sheet has hundreds of them, thousands of them. Every day it adds like a hundred new videos in there. Uh and then they're kind of going in there, oh like this is good score, 75, one small change, just remove that thing about talking about the neck or something like that. Remove that and then the thing that she's saying at the end is actually better as a hook. Move that sentence to the front. And then And now our agent actually knows everything about this video. It also recommends ad copy, four or five different kinds, short, medium, long, things like that. So like all of this stuff is kind of removing these kinds of work from people's plates so that they're kind of focused more on creative stuff in and the agents are doing just the grunt work of these meaning little tasks.
Kurt Elster • 31:11.519
Here's a word from our sponsors at Zippify. Black Friday's coming, and so's the squeeze. Every year, the same thing. Ad cost balloon, customers want 40% off everything, and your margins get crushed. But there's a fix. Boost your average order value before the chaos hits. One-click upsell does exactly that, increasing AOV by up to 30%, with AI-powered upsells across your entire customer journey. Launch up sales for every product in one click. Customers see personalized offers and spend more automatically. Unlike other apps, you only pay for results, not views, so you keep more profit. Takes under two minutes to set up and starts working immediately. Don't let this traffic season slip by without maximizing every sale. Go to zipify. com slash Kurt for your 30-day free trial. That's zi-ip IFY. com slash K-U-R-T. Turn this into your most profitable Black Friday yet. You're taking the hard part and automating it so that The people who have experience doing this could do the valuable part. Um I love and especially like a bulk operation where we have tens of thousands of videos. It just is not realistic for A person to go through all that and make meaningful decisions.
Bora Celik • 32:28.980
Like they it's not even it hasn't replaced the work that was being done. That work was not being done at all. Yep. Because it was not reasonable or cost efficient for a brand They would have spent millions of dollars trying to review hundreds of thousands of videos with like humans what sitting there and watching each video one by one deciding. Well, I mean that's that was the previous process, right? Watching Uh but they decided to only watch like five percent of the videos because that's that's even too much.
Kurt Elster • 32:57.200
So okay. I get I I you have opened my eyes to like the next level power that is possible here. For sure it you said, hey, uh your advice was learn. a a automate an AI automation platform like N8N first is like that's a good starting point and then you know hit the limits of that and then move to more advanced custom tooling like you're doing, which makes sense. I mean, I I tell new Shopify merchants a similar thing. I'm like, hey, you know, take the theme as far as you can go in the theme editor on your own so that you understand it and can talk the language when you're hiring someone to do it for you, you'll get a better result. That part makes sense to me. Um I want to know, you know, beyond influencer management for a a Shopify store owner. What else works? Like what are some of these these agents that we could leverage for high ROI impact?
Bora Celik • 33:50.000
Well In the enterprise, the most popular use case is customer service. That's the most impact. That's the biggest market share. That's why there's a bunch of companies out there that's doing uh customer service AI. Um and it it it does a good job.
Kurt Elster • 34:06.720
It just uh you know, any in particular you think is decent?
Bora Celik • 34:11.360
I haven't used them specifically, you know, um myself because You know, I I I build these things. You know, I just don't buy or use software anymore. Like I canceled our hubspot in uh just octo beginning of October. It was uh we had an annual contract It's like $26,000 a year or something like that. I don't need this. Just all these like internal tooling that we can build. Um and then you you can kind of cancel stuff like this. So like I'm a believer in build over buy now. And I encourage people, everybody fights me on it. Oh no, you know, people don't want to build. They just want like what's ready. Oh no, well you can customize agents so much that does exactly what you need. And then you don't rely on this Or vendor. So, you know, I'm a big believer in people learning how to build these kinds of things. And you don't need like a big team. You just need like a couple of people internally who are like super AI native. uh understands how workflow automations work and just like pick one thing at a time. You know, I recommend people to you have these SOPs like standard operating procedures or training videos that you're training your you know, either virtual assistants or interns or offshore team to kind of do stuff for you. Take those as input. You're using Chat GPT or Claude already. upload them and say that look here's an SOP that uh this process uses find me the areas That we can automate with AI agents. I mean kind of clever cla Claude does a great job. I'm sure ChatGPT5 will also do a great job.
Kurt Elster • 35:43.280
Yeah, I I prefer Claude too.
Bora Celik • 35:44.800
I I I like Claude very much. Uh I like them both. I kind of like interchangeably use them for different things.
Kurt Elster • 35:50.320
I refer to Claude as our our French developer. He's overseas. Yeah.
Bora Celik • 35:54.080
Yeah, Claude is is really good for programming, architecting, things like that. Um, I like it a lot too.
Kurt Elster • 36:00.480
Well, I want to know what the limitations are. Like right now it sounds like you get this thing to do just about anything. Uh what What can't your agents do? Like what's the thing if I asked you for help with, you'd be like, I can't do it. Like, listen, I need it to perform brain surgery. Obviously not. You know, where where's the limit?
Bora Celik • 36:16.140
Yeah. I don't know. Like for our kind of work you know, just internal operations of an e-commerce company. I haven't seen anything so complex and advanced and difficult than that AI agents won't be able to do for this kind of work. Like I I don't work with airports, flight control, doctor surgeries, or, you know, the kind of security you need transferring somebody's money from one bank bank account to another. Like those are the kinds of sensitive areas where Scary things might happen. In our line of work, I always think about like yesterday I commented on somebody's LinkedIn, he was talking about it. Like, what's the worst thing that could happen? Like if this goes wrong. Right. Um so you kind of have to start from that. What's the worst thing that could happen? Uh you know, like in the case of emailing agent says something like really weird, embarrasses the brand, you know, like that would be a bad thing to happen. Um or like if the video is scored incorrectly, it's not a big deal, you know, scoring every once in a while. Like If it's like 1%, 5%, you kind of compare it to humans, right? So like sometimes people compare AI agents to, oh, this needs to be godlike. 100% correct all the time. I'm like, well, no, let's compare it to our coworkers who are 70% correct and 30% wrong. You know, uh the people that we work.
Kurt Elster • 37:44.720
Exactly. And that works there's no question.
Bora Celik • 37:47.040
I'm okay with them making a mistake, right? So there's just a lot of purists out there. They're like, oh, look, look, ha ha ha. Look at this screenshot. The AIA just messed up. I'm like, well that's perfect solution fallacy.
Kurt Elster • 37:57.000
Where like if it's not perfect a hundred percent of the time, don't use it. It it doesn't make sense.
Bora Celik • 38:01.800
Yeah. Well that's what I meant, Kurt, by it needs to earn our trust, where you kind of get a feeling that you know what, like it's it's good enough for this job.
Kurt Elster • 38:11.240
If I'm getting started with this, what are the tools I should explore? Like I'm a Shopify merchant. You know, my automation experience is probably Shopify Flow or Clavio, uh, and I'm familiar with ChatGPT. then you know I'm gonna try, you know, maybe I use uh Zapier. And it sounds like N8N, and you know, N the number 8N is a good starting place. It is.
Bora Celik • 38:35.859
What I do is, first off, one annoying task that you do every day. You know, one annoying thing. It keeps it like really small, like atomic task. And then sign up with NADN. You drag things on the screen. Start with like a manual or just like you can you can automate it, run every hour or something like that. And then there's drag an agent, drag a model, use like OpenAI GPT 4. 1 or something like that, or Claude, you know, you just put in your API key in there. And then at this point, you you know you're gonna need to kind of connect some tools for the agent to do stuff. So we didn't really talk about MCP that much, which I believe uh MCP is the future. I think it's going to replace the web. So everything that I build now are MCP servers and MCP server tools. I don't build APIs, I don't build websites, web apps. Um and even the agents, I feel like agents are kind of moving up to, you know, the consumer level simplicity now where you can build the agents. But you need to connect to some kind of tools and MCP tools for the agent to be smart and just do its work.
Kurt Elster • 39:51.840
Like And MCP, we didn't define it, but MCP model context protocol.
Bora Celik • 39:55.900
It's a context protocol.
Kurt Elster • 39:57.500
A standardized for agents to use. Yeah, for AIs to talk to each other.
Bora Celik • 40:01.420
You know, I they kind of explain it as it's like a use USB-C connector. So it lets you connect to things and Yeah, just let's give you like one specific example, right? So now I'm I started a series of build your own customer service agent, right? Um I made a video, then I uh you see this board behind me, this new glass board that I got that allows me to Make this glassboard uh neon marker videos now. The next one that I'm gonna do is I'm gonna make it real. Customer service agent. What does a customer service agent have to do? Let's do very simple things. Check my message history. Check the order status in Shopify. Draft a reply. Maybe somebody's asking, what's where's my stuff? How do you respond to that? Right? Check the message history and MCP2. Check the order status using the shop Shopify as an MCP2 now. So you can just drag Shopify's MCP there. Um and then draft a reply. That's a very simple thing you can do today. That's pretty good.
Kurt Elster • 41:00.480
Yeah. And that, you know, that replaces a big annoyance, a headache, you know, something you don't want to deal with is these where's my order requests. Right. And so you could use this tool. That's very clever.
Bora Celik • 41:10.900
Yeah, but the thing is, Kurt, once you do this, the simplest thing, explosions will go off in your head. Because then you have been enabled now and you're like, oh my gosh, now I have a million other things I can do. I have all these ideas. So I really love this for people because people have been wanting to build stuff forever. And they're just like, oh, I need an engineer, I need a developer, they're expensive, I don't have money like that. But now they can just put their ideas to work. By kind of like engineering thinking. And with AI agents, it's it's quite easy to build. And once once you s once they build it, like these hundred million dollar brand CEOs are doing it now. They're getting so excited being able to vibe code a page. or like being able to put together their agent. They're like, oh my, like I'm gonna lead by example. I'm gonna learn this myself and I'm gonna teach my team. I'm gonna tell them to learn it. Everybody should learn it and we should be an agentic brand.
Kurt Elster • 42:04.720
Ugh, it it is fun. Like I've always loved automation of all kinds. And this just takes it to another level. And you're right, like the It's so empowering and I've learned so much working with these tools.
Bora Celik • 42:16.859
So it you've given me another Honestly, you make it makes you a happier person because a lot of people really hate their jobs. because of the menial tasks that they do, right? They kind of don't want to work. They don't want to go to work. But now like I I was kind of looking at myself. I've been working for a long time and it felt like My career was kind of getting like there's nothing new to learn. I know everything already kind of a thing. But then when there's some new things to learn, new things to do, I feel like there's a little bit of a More of a pep in my step, like when I jump out of bed in the morning, oh look, we can build this interesting, cool new thing now. It's kind of like if you have like a creator kind of a mindset that you kind of enjoy finding flow in these kinds of things. I think it kind of could turn you into a happier person because you learn more, it's more fun, and you kind of get into these new challenges.
Kurt Elster • 43:05.240
That's I agree. It's very cool. I'm I'm excited. You have some resources. Um, you've got a newsletter, YouTube channel, obviously your your own consulting business. Yeah plug it. Tell us about it. Where can we learn more from Bora?
Bora Celik • 43:20.520
Yeah. So our business is not consulting, but it is a little bit like Palantir. You know, it's not, it it has uh platform products. But they're not traditional software products. It has custom implementation, so it comes with services. So agentik, a. gentik. co, that's the website. So I like to teach people how to do these things. I created a newsletter slash YouTube channel, agenticbrand. ai is the newsletter. And on YouTube, it's at agenticbrand. Um to both of them. So I try to make one video a week trying to pick a subject, teaching how, like, or just trying to encourage people like the brands who work at brands. uh or anybody, you know, who's kind of like trying to get into AI by kind of learning how to build agents or how to think about vector databases. how to think about, you know, how influencer marketing can change or different kinds of and now like I'm doing the series on um customer service agents. Sometimes you know people come to me and say, oh like I'm using this thing, but it won't let me upload all my knowledge base. It's like only one file limit or something like that. Like well build your own. Well I don't know how to build it. Can you help me? Well, let me make some content.
Kurt Elster • 44:33.220
Smart. Very smart. Thank you so much. This has been great. Boriselic, agentic AI. Appreciate it.
Bora Celik • 44:40.740
Thanks, Gert, for having me. Appreciate that.
Kurt Elster • 44:45.580
Crowdfunding campaigns are great. You can add social proof and urgency to your product pre-orders while reducing risk of failure. But with traditional crowdfunding platforms, you're paying high fees and giving away control all while your campaign is lost in a sea of similar offers. It can be frustrating. That's why we built Crowdfunder. The Shopify app that turns your Shopify product pages into your own independent crowdfunding campaigns. We originally created Crowdfunder for our private clients. and it was so successful we turned it into an app that anyone can use. Today, merchants using Crowdfunder have raised millions collectively. With Crowdfund You'll enjoy real-time tracking, full campaign control, and direct customer engagement. And it's part of the Built for Shopify program, so you know it's easy to use. So say goodbye to high fees, and hello to successful store-based crowdfunding. Start your free trial and transform your Shopify store into a pre-order powerhouse today. Search Crowdfunder in the Shopify App Store to get started.