How AI is Revolutionizing Communication with Jacob Bank & Relay.app

for small business owners like take a very structured view of everything you are currently doing or wish you would be doing. Imagine you you had unlimited budget to hire a really bright college intern to help you with that thing. How would you give that person instructions such that they could not fail at the job? Hi, my name is Dmitri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI Agents podcast brought to you by Jot Form and featuring our very own CEO and founder Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hey there, it's another episode of the AA agents podcast and in this episode we are going to be talking to Jacob Bank from relay.app. How you doing, Jacob? Hey, Dimmitri. I'm good.

How are you? I'm living the dream. Uh, I was curious, you know, we we've we've known each other for a little while, so I think this conversation is going to be uh one of the one of the better ones we've had. How how have you been, man? Like what's what's new with relay.app? Tell everyone what relay.app is and all the cool stuff you're doing. Yeah, I'll I'll give you the story and then I'll tell you how we're doing how we're doing right now. Um, so when we started relay.app, uh, this was about a year before catch came out. Um, so it was a very different landscape in terms of tool building and what all of our expectations were from AI. And so we didn't know exactly what the right product was to build. But here was the observation that we had. Um, when any one

of us is doing our real job, whether it's, you know, turning this recording into a good podcast or sending outreach emails to potential clients or meeting with customers, we all have to use a bunch of tools to get to get that worked out. I mean, it sounds obvious, of course. Um, sure. But individual tool builders aren't really wired to think about the full job that a person is trying to accomplish. They're wired to think about the capabilities of their tool. So before starting Relay, I was at Google for many years where I was the product lead for Gmail and for Google Calendar and for a few other tools. And I remember in Gmail, we would agonize agonize about every little pixel of what the Gmail tool could do, what the reply button looked like, what the archive button looked like, how you set up a

filter, how you set up a label. And then we'd go talk to customers and they'd be like, "Yeah, yeah, whatever. The reply button's fine. Like, I can reply to emails. It's fine. That's not my problem. My problem is I need to get stuff from Gmail into ClickUp that I can send send a report out over Slack every week." And we say, "Ah, good luck. Use the API." Um, and so that was sort of the the original idea behind the company. could we build this sort of relay across multiple different tools and multiple different people that would be a you know that would actually help you get your real work done. Um, that was that was the original idea. And what that means now, now that we have the language to describe it as such, that means now that we have an AI agent and workflow

tool where you specify a job that you want accomplished, like every time Dmitri posts a new video on his YouTube channel, I want to automatically get a summary of it and a draft of a LinkedIn post idea because usually he posts good stuff and I want to riff off of it. Um, I would create an AI workflow or an AI agent to do that using a tool like related app. And how are things going right now? things are good. Um, you know, part of it is because I think we have a pretty decent product and part of it is there's just such a wave of interest and demand. I mean, this is called the AI agent podcast. Like, there's such a wave of interest and demand that we can all sort of ride we can also s sort of ride that wave. The key is

we just have to make sure that we're delivering something that's actually useful, not just something that has a shiny demo. Um, so yeah, that's that's where we're at right now. We get a lot of a lot of new users coming in, a lot of new customers coming in every day and I'm just meeting with as many of them as possible to try to figure out where AI workflows and agents can actually be useful for them. Yeah, that's that's an interesting thing. So, I mean, obviously starting off kind of in the in the realm of more an automation company. What has it been um working on the fly to essentially fit into the AI agentic messaging and not only the messaging but like the performance in the product? Yeah. Um you know when we first spoke maybe two years ago at the time we were a

no code we framed ourselves as a no code automation tool. Exactly. Same category as Zapier as Make.com. Um, and the the spirit of no code automation tools at the time and still now is you can set up these if this then that structured workflows. When a new email comes in, automatically extract some information from it and put it into into a spreadsheet. And what we found was that no code automation was not communicating the right message to the right audience um for a couple reasons. One is by saying no code. Yes, you're trying to differentiate yourself from code by adding the word no in there, but just the fact that the word code is there at all is automatically disqualifying to 99% of people. Just like the fact that you even need to mention that it's a no code tool and that code may be

a possibility is like puts it outside the realm of of a typical non technical person. And the second is that I don't think automation is the right word. automation to people. It sounds mechanical. It sounds wrote. It sounds lowquality. It sounds like it's going to be a lot of work to to set up. Like I don't Let's say I I want to have an AI agent that creates a a synthesized report of all of my customer meetings every week. Automation isn't the word people use to describe that because it has much more intelligence and much more vibrancy and value than what people think of when they think of automation. So about 5 months ago, we changed our primary positioning from a new generation of no code automation tools. We didn't we never had the word no code in our tagline, but the no code part

was implied. We said, you know, we were the next generation of automation to a platform to create AI agents that work for you. And it turns out just making that positioning shift from a no code automation tool to uh a tool where you can create AI agents that work for you um sort of 10xed our demand immediately overnight. And that was without even necessarily making any product changes because for that example, a weekly synthesis of customer insights, yeah, you can set that up in a no code automation tool. You run a weekly trigger. You fetch all your recent transcripts. You pass them through it through an AI step and then you dump it out to Slack or Google Docs or or wherever you want to. But the the framing of an AI agent did such a better job of capturing for people this is a

real a really valuable task that requires intelligence that I want help with and I want AI to help me do it. Um and so yeah that's that's a bit of our sort of journey from automation tool to AI agent building platform which has primarily been a positioning and messaging change. Of course we've also changed the product al along the way but I think you know we've done this this new framing does a much much better job of communicating to our target audience what what the thing is for and how you use it. Okay. So with with AI agents in 2025, you know, obviously this podcast, like you said, AI agents podcast, Idkin was uh nearly prophetic with the idea. I think he had the idea a uh maybe like July of last year, which is in the world we live in pretty early, you know,

to be less than a year to be six months early you're early. Um what do you find is some of the most common use cases that uh companies end up trying to add to their workflows? Um maybe two would be good. So first I think it's important to note that I'll use AI automation, AI workflow and AI agent interchangeably. I know that they're technically different and we can talk about all the technical differences between them and none of the use cases I'm going to mention technically require a true AI agent that does runtime tool calling based on based on a goal. Um, I think what's really important about this class of use cases is that the AI works for you behind the scenes. You don't have to go to a chatbot experience to talk to it. It just does the work for you automatically and

it does something that has, you know, material material intelligence that previously would have required a human. My two favorite classes of use case right now are synthesis and research. A synthesis example is the one I just mentioned. Every week, send me a summary of all my customer calls. Every day, send me a summary of all my support tickets. Every day, send me a summary of all of the new AI model announcements that have come out. Um, every week send me a synthesis of all of the highlights from my customer feedback. Send me synthesis from the low lightss of my customer feedback. This is something that would have been nearly impossible to do without AI because um, and I've, you know, I've worked in teams where someone's whole job was to do synthesis tasks like this, like look at all the transcripts of customer interviews and

write up reports based on them. But those programs never really stuck cuz they were way too much work for a person to do to synthesize all that information. And now you give 20 transcripts to your favorite your your friendly neighborhood LLM and it comes up with something very good in 10 seconds. Um, so I love those sorts of synthesis use cases because those for me are not just reducing the mechanical friction of something we've always done, but unlocking totally new use cases that we just weren't feasibly able to do before. Second class of use cases I really like are research use cases. Look up a person before I meet them. Look up a company that signs up in our funnel. Um look up, you know, a particular industry or job title. And again, these are the kind of things where yes, there's always been like

lead enrichment tools that would find someone's LinkedIn profile or do some some scoring, but often to do good research, you need to do multiple Google searches. You need to look at multiple websites. you need to browse for multiple data sources and having an AI agent that can do that automatically before you meet someone rather than spending the 20 minutes doing it yourself is is pretty magical. You know, that's a good that's a good point. I I I see a lot of what is going on in the AI agent space and um what I'm noticing from you specifically is that you've grown your company obviously through the messaging of your own product which I find very intriguing but I've also noticed that even on an individual level you know sure you're using it for your business in in this essence but I've noticed on an individual

level it's been very beneficial to you uh for those of you that don't know uh Jacob has a very nice LinkedIn to follow. I obviously have known him for a little while, so maybe I'm a little biased, but I really enjoy what's going on with your LinkedIn. So, I'm curious when it comes to communicating, you know, growing in and public in a sort of way as as you're kind of doing on LinkedIn, what necessarily does it look like for you to to create content? Because to my understanding, it's gotten easier and easier as you've done this through your own products uh capability. Yeah, totally. This So, first, this did not come naturally to me at all. like I am not a natural social media communicator. That's like the last thing I ever would have imagined that would be an important part of my job. But

I think there's two things that have been really important for me. The first is figuring out how, you know, what class of content resonates and how to get ideas for it. And the second is like mechanically executing on on those posts efficiently because I'm, you know, I'm a solo I'm the solo marketer at the company and I've got a lot of other things going on. And so the opportunity I saw was that there are tons of people posting very um polished shiny demo videos of things that don't actually work um and don't have any real usage. That was observation number one. And then number two, people are posting lots of hot takes about models and tools, but again not actually using them themselves. And so every single thing I post on LinkedIn is either something I'm using myself actually and getting value from it or

something one of our customers is actually using and getting value from. And so my LinkedIn posts, they're not they don't have any fancy hook format. They're not like I didn't go to any LinkedIn academy. I just write like here's something I'm doing with AI right now. Here's how it works. Here's why it's valuable for me. Like let me know if it could be valuable to you to to you too. Um, and because that format is so practical, idea generation is very easy. I spend a good chunk of my day building AI workflows and agents in our own product. I spend a very good chunk of my day talking to customers who are doing that. I probably have 10 customer meetings a day. And so at the end of the day, um, I have many, many more ideas of things that would be useful to people

than I can post. And that's where, you know, part two is I have some AI to help me out that reviews my customer calls every day and says, "Here are three use cases that might be good to post on LinkedIn, and here are some uh some uh here's maybe even a draft of a post in your voice and style based on what you talked about with with this customer." And so that's pretty much my whole LinkedIn. About three or four times a week, I post a specific use case of something I'm doing our customer is doing with AI. I ask people if they're interested. If they are, I make a template, a video. Sometimes I do a live session about it and then the it just kind of kind of rinse and repeat that there's no I wish I could say there's some magic formula

or science to it. But and and sometimes I'm shocked by which things perform well on LinkedIn and which ones don't. I still totally don't understand either what people want or how the algorithm works. Sometimes I'll have a post that I think is going to be a total dud and they get its hundreds of thousands of impressions. And sometimes I have one that I'm really proud of and I think is awesome and be useful to a ton of people and it gets like 2,000 impressions. Um, but that's, you know, that's the the the name of the game in any of these social media platforms, right? No, absolutely. It's kind of frustrating how it works like that. Um, what have you seen from I think maybe more of a fun side for for early founders? Uh, what what kind of gains have you seen from a efficiency

standpoint at your own company? For example, um you obviously run an AI agents company, but this podcast still being about AI agents. A lot of uh businesses are concerned with how do I save time in my own business? And you know, one of the key things that a lot of people talk about is scaling, right? Scaling is really hard. Um so how have you felt that this your own company has actually helped you with uh the scaling issues uh that you know a growing company face? We're a nine person team. If we didn't have AI helping us to achieve what we're achieving now, we'd probably need to be more like a team. Um, wow. Can you say that one more time? Do 15. You you you'd need Wait, say that one more time just so you get I think if we didn't have AI helping

us to achieve the same things we're achieving now, we would need to be 15 people instead of nine people. we would need to be like that's a pretty whatever hefty difference 40% bigger um I have AI agents so so the way our company works is we have an engineering team five engineers we have two designers we have a product person and they of course use AI in the product development process as well but my job is I am marketing sales customer success much of customer support HR finance and operations that's all me there's no other virtual assistant There's no other chief of staff. There's no marketer. There's no saleserson. There's no other magical person. It's literally all me. Um me and my team of AI agents. And so in marketing alone, I have about 70, I think, AI agents that do everything from automatically drafting

descriptions for my YouTube videos to automatically drafting LinkedIn posts for those videos to automatically drafting LinkedIn posts for webinar promotions to automatically replying to comments on my LinkedIn posts to automatically tracking leads that come through LinkedIn and sending them personalized outreach messages to automatically creating creating email newsletter drafts to our audience to automatically research and enriching every new user who signs up for the platform. The list goes on and on and on and on. Um, and there have been various times, I think over the course of the company, we've hired seven or eight different marketing contractors just specifically for marketing to do various things, whether it's SEO content or video creation or email marketing or um, positioning and messaging. And every single time I've realized, and this is not a knock against any of those individuals, they were all good. But it turned out that

just me plus AI was cheaper and better. Um because I have a lot of the context of the company in my mind. I'm the one meeting with all of the customers. I'm the one kind of interacting with all of the different elements of the product we're producing and business we're creating. And so it's very very efficient for me to say we need to create some content about this category. And in the preAI world, that would have been like, ah, find a blog writer, give them a brief, have them go off and work on that brief for a week, have them send back a version of the blog post. Might be a few thousand words. You do a round of edits, they take another three days, you do another round of edits, they take another round of four days, and then eventually you get something that

you're almost happy with, but you're just you're just exhausted by the process, and you don't want to go back to the human again. Um, and so in two weeks and probably an aggregate, you know, two hours of my time going back and forth on this thing, you get something decent. In the AOI world is you get you create a brief, you have a live conversation with CHBT or Claude or pick your favorite. You've given it through a project, a bunch of your historical context of blog posts that you've written. So it knows your tone, knows your voice, and that whole iteration process is done with within an hour. And that's both obviously the wall clock time is dramatically lower from 2 weeks to 1 hour, but even the active time is lower from 2 hours to 1 hour because um you're in such direct control

of the guidance that you give to the AI. And so one of the things that I saw that I see more and more people posting about right now, which really resonates with me, I'm curious if it resonates with you too. Um the most important skill, professional skill in the next several years will be how well can you structure your own thoughts and articulate what you want. Um because it will no longer be important to do every little bit of execution along the way to that. We'll have many many different types of AI tools and products to help us with that. But if you cannot structure in your own mind what you want and articulate that clearly, the AI will have no hope of achieving it. And so, uh, my hot take is that I think like philosophy majors are going to be the best equipped

to succeed because if you've studied logic, if you've studied really crisp, articulate, well ststructured communication, you are going to thrive. You are going to thrive in the AI agent era. What I what I have to say about it is honestly I do I do agree with you there there's something interesting happening in a couple of different places. So one I had a thought that I I did want to touch on. So I'm I'm trying to make sure I don't forget it. But when you were talking I remember a specific conversation that ended up occurring on your LinkedIn. It was a mixture of contentious and kind uh responses to your post. I think I know exactly which one you're talking about. This was a funny one. Yeah. Yeah. Could you actually touch on this cuz I want to cuz I want to talk about it and

follow up on your thought on like the whole who can like synthesize their thoughts thing because I do think some of the people in that conversation may have actually missed the key component which is the initial human like u inputs because they they kind of went off on you. It was pretty interesting but I would love for you to touch on this. The post I wrote on LinkedIn was how I use AI to help me write LinkedIn posts and I basically told it like how I use AI. I basically I write down an idea that I have. Then I have AI which has knowledge of all my previous posts, my voice, my style, my structure. I have AI create a draft for me and then I post it and then I edit it and then I post it. And the way a lot of people

interpreted that was uh this guy just creates a bunch of AI generated slop and spam and like not only is this going to be lowquality content but it's downright evil to publish AI AI generated content on LinkedIn. And my response to that was um well first again the way I wrote it down was the original idea was mine. The point I wanted to communicate came from an original idea from me. I had the AI do some drafting for me based on my voice, my tone to take away some of the writer's block of getting it down on paper. And then I took it over the line to edit it to make sure it met my standard. And so first, I think I don't know why people interpreted that as that I was just like autopublishing AI generated slot because I think I was pretty clear

about like there's a human in the loop at the beginning and a human in the loop at the end and the AI does a bunch of the grunt work in the middle. Sure. And so my response to that was like because a few of the people who were commenting very negatively on this were like marketing executives and I was like have you ever as as a marketing executive ever had an idea for a campaign and then asked a more junior person on your team to flesh out that idea into a specific set of copy and proposals and then you review those before they actually make it out into your budget. Of course. Of course. Of course. Everyone who has ever worked in a marketing or content related production of any sort has always had assistance from a junior researcher or a junior writer or a

junior editor or sometimes all of the above. Um, and so my point was just was not it was not supposed to be like, yeah, let's just dump a bunch of AI slop onto LinkedIn. My point was that it used to be that only the very elite people could afford a researcher, a writer, and an editor. And now we all can. That's so amazing. We should all be using that like every and now now like your content is still only going to be as good as your idea and then your final packaging of the content. But man is it useful to have AI to do some of the writing, the research and the editing for you. That so that was the that was the point I was trying to make. It turned into this whole debate of like is AI content ruining everything? And it's like

and and I'm sympathetic to that point of view like we've all seen AI slop for sure. We've all seen AI slop. Oh my god. Yeah. Um, but my point was that if you have a good idea, AI can be a really really helpful tool to help you deliver that idea effectively to people. Well, I think that that that that was something I wanted to get to. It's funny. Um, there's there's a couple of layers of interesting in that uh argument that I remember seeing. It was like one getting into the immorality on uh how we post on a social media platform is kind of hilarious considering it's just business advice. Like I don't know. Um, not to get too philosophical, but you know, getting into philosophy a little bit, right? There's just like a bunch of false presuppositions that seem to have been like attacking

your point. It was like presupposition one, it's AI content, not AI augmented human content. Yeah. It's like a My point was that like it's an AI assisted content creation process, not AI content. And so, yeah, it was funny. There were a couple reactions that that maybe are worth calling out. One person wrote this whole like response post of saying, "If you take the time to read my post, I can guarantee you that every single word was written by me." And I'm like, "Okay, okay, that's like a fair perspective." The other person um who I'm actually a fan of, she's a content writer, her name is Lauren, and what she wrote was like, "All I care about as a reader of your stuff is, is it unique? Is it useful? Is it valuable? Is it engaging? I don't care. I don't care how you deliver that.

Like, if you have a two sentence idea and then AI takes it the rest of the way and comes up with something useful, valuable, and engaging, I'm going to read it. If you deliver, if you write 50% of it and have AI edit it and I find it useful, valuable, engaging, and sort of by definition, if you do something fully autogenerated without any of your own ideas or personality in it, it's very unlikely to be useful, engaging, and novel because anyone else could have had automatically write a LinkedIn post for them for them too. And so I think that's the way we should think about it. like how do you create useful engaging output which for almost every task in the world going forward will be some combination of what you know in your own mind and how you're using AI to help you get

that you know get that out to the world. Yeah. No, I I think this is this is one of the key points um that you're touching on. It's that the human element at the beginning of the process is obviously going to be always very important and an AI agent even in so an AI agent will produce content will probably even if it's fully autonomous need to actually be pulling from humanity I'd imagine in order for it to be well performing and and I didn't read I I hadn't read any of your posts that AI um and I think the funny thing is My favorite critique of the of the critique was that if it was AI slop, it wouldn't have gotten any traction. Yeah, exactly. Cuz it was funny. There there were a bunch of because I even said at the bottom like AI helped

me write this post, which it did cuz I had written like a little content idea of like, hey, it'd be cool to write a a post about how I use AI in my own content creation process. And at the bottom, I said, you know, PS, AI AI wrote 80% of this post for me. Um, and then there were all these people saying like, "I would never read a post written by AI." It's like, "Well, you just did." Yeah. So, I um I just I I love I love LinkedIn and I love uh there's like this subculture of irritation that like comes out once in a blue moon and then they like go off. Yeah. It's not as bad as on other networks. LinkedIn is generally kind of a cheerleading place. Yeah. Other thing that's that's funny is sort of like my perspective on LinkedIn is

often a little more AI skeptical. Like um I am not one of these people who claims that every job is going to be irrelevant in in two years or like I made the beall end all agent that does literally everything 100% for me. My whole take on the market and my whole style on LinkedIn is like we need to think very carefully about where you need a human in the loop. What is AI actually capable of today? What are the right places to apply it? And that's why I try to lean on like very practical things that I find actually useful. And so it was funny that I got sort of like lumped I got sort of like lumped in to a category of people that I don't really agree with. And I was sort of I was sort of defending that and also why

I wasn't actually saying that you should just like autopost a piece of AI generated slop every day. Yeah. Yeah, I mean so I the reason I'm asking this is that um you know I'm actually thinking about this in in a way it's like my my current business is like I make AI content or I sorry I make content and a lot of it actually is augmented by AI but it's a funny funny thing where there's like a layer that people will get irritated the u in distinction. So for me, obviously I produce content for like 20 companies. I make like 200 pieces of content a month. If y'all don't think I automate or AI part of this, you're because I don't have a team of that many people. Yeah. Eat people. So it's like where do you think this is coming from? And and when

people are none the wiser, they end up not being that irritated about it. Um so uh it's interesting. One interesting thing that I did and I'm curious to hear your your take on this is I also make videos for our for our YouTube channel. I'm not a pro like you are, but I'm I'm I'm dabbling. Um, and I tried to kind of push the limits of how much of that video production process I could do with AI. And so, certainly, um, I was able to use AI to help me get ideas for videos. You know, looking at customer requests, looking at use cases. I was able to use AI to write very good scripts for videos. Very, very good. Because most of my videos are tutorial style um, videos. And so I would basically give the AI um the use case I wanted to write

a tutorial about and then some of my previous tutorials and it would generate an excellent excellent script. The AI was decent at generating the voice over. Actually I trained a a custom voice model for myself using 11 Labs. I then also tried to create a digital twin like video avatar for for myself using hen. That did not work. I then you know in post-production I have AI automatically do a bit of editing. I have AI do generate the YouTube description. I have AI generate the YouTube thumbnail. And I basically kind of like pushed the boundaries. And what I found is that the actual voice over of the video and the me talking, it was not appropriate to use AI for those. Again, not because like I'm above using AI or I think it's immoral or anything. It just didn't create a good finished. It just

didn't create a good final product. it lost sort of the emotional resonance that people get when they see that it's actually me on screen making mistakes being human. Um but for the whole rest of the process it's amazing. Yeah. No, I I agree. I think that the key thing to to for me right it's it I totally agree. Um the content creation outside of the human element with on camera I think for the most part is pretty much AIable. Um, and what I mean by that is it still will require human input and human in the loop and and probably a writer to be able to QA it and then maturely, you know, like massage it, tweak it, when it comes to recording, that's the only part that you they're not there yet. And I'm I'm using the word yet. As somebody who runs a

content agency, I'm aware that I think it'll prove I do think that'll be part of I and I I think there will come a time when probably it's good enough and you can just write you can just write a script in text and your AI video avatar will deliver an amazing performance. Yeah. But I I my favorite thing in this whole thing is once again not to get too philosophical but everybody somehow bridged the what's that whole thing is it uh I forget which philosopher it is but you can't really get an ought from an is and everyone's just like a content is AI is bad. I'm like you don't um it's just like so funny. So I I wanted to touch on that because I figured you'd have a little interesting passionate response. Um so at the current juncture though of your business, what

do you see as some of the improvements uh that will be coming either to your product or to the AI agent industry in general that can really help businesses in uh this year or maybe even moving forward? What are some of the things you're seeing? The big thing is that no product in our category, including our own, is easy enough to use for nontechnical users. Yet, you still need to know how to set up a trigger, how to set up your AI prompt, how to, you know, wire the different components of of the agent together. our our space has not yet had the, you know, lovable or bolt or vzero moment where a totally non-technical person can just type natural language and get a high functioning AI agent out. I think it's harder in our context. This it's counterintuitively harder in our context than it

is for lovable. And I can talk about why I think it is harder in our context, but that moment is coming. That moment is going to come probably in the next year where anyone any small business owner will be able to create their own essentially virtual employee in the form of an AI agent. So I'll give you an example. Um one of the customers I'm working with, he runs a uh a shutters and blinds installation company in Southern California. He would not describe himself as a technical person. I mean, he's he's good in, you know, he knows his shutters and blinds, but like he's not a computer programmer by by by any means and um he's been able to succeed in our product because he has really great intellectual curiosity and he's willing to keep hammering on things until until they work. But nine out

of 10 or 99 out of a 100red shutters and blinds installers would not be able to succeed in our product today. It requires just too much of a technical mindset and technical thinking. And I think by the end of the year, we might be able to get that 99% who fail down to 70% who fail or 60% who fail or 50% who fail. And then within a couple of years, any person, no matter how non-technical, will be able to create, for example, an AI agent that automatically parses out all of their orders from Lowe's, adds the right information from Monday, picks the right job, picks the right crew to schedule, and then sends a confirmation to the customer. Like, it will be possible for anyone to create that soon. I don't know how soon is soon. Three months, six months, a year, two years, but

it's coming. Yeah, that's that's something interesting that a lot of people are probably How do I say this? People who are new to any of these things, whether it be like AI or like automation, they make a funny assumption where it's always like, "Oh, if I just type, it'll do what I want it to." Man, it takes a lot of massaging. It takes a lot of massaging. It takes a lot of knowledge. It takes a lot of things. Exactly. And that's why that's why some of these like type to agent things, they don't actually work because many many tools, Zapier among them, have had an an experience where you can type in natural language what you want your workflow to do and they don't really work because they get you 50% of the way there and then you have no intuition or ability to get

it 100% of the way because as you know, you've built a lot of these automations, you constantly have to tweak them, adjust them, test them, help them handle different cases. And so it's uh it's it's it's really tricky to to to get these things right. The other thing that I think is underrated here, this goes back to the point about the importance of of communication because not everyone has been trained to communicate in such a structured way like I studied computer science. Um, this was my my academic background was all in computer science. And I remember, you know, every time you take an introductory computer science course, the first thing the teacher does is they do this little exercise and they say, "Um, give me instructions on how to walk out the door of this classroom." And someone in the class will say like, "Oh,

just walk straight." And they'll walk straight into the blackboard. And then someone will say, "Ah, turn right." They'll turn right 270° instead of 90°. And the point of this exercise is to show when you're talking to a computer, you need to tell it exactly what you want to do with an extreme level of structure and precision. First, turn your right foot 90° to the right. Then place your left foot in front of your right foot. Then turn your eyes 90° to face the door. Then take four paces forward at a roughly, you know, three-foot stride length. And that's that's kind of like the point that the the teacher is trying to communicate to the computer science audience. And in the future of AI agents, we will not need to be that prescriptive. You will not need to literally say, you know, turn your foot 90°

to the right, then turn your other foot 90° right, then walk forward three paces. You'll be able to say something more like, you know, turn to the door and and walk through it. But you still need to have that sort of structured mindset of what is the order in which things need to happen and how do you give good enough instructions of of how to get there. So I that's why you know if you're if you're preparing for how to be effective in the AI era I would get really really good at giving people instructions like like another another thing that I think is um a good a good case study for this is like you know have you ever been driving and have you know one of your friends doing turn by turn uh like doing directions for you and they'll say like oh

look there look there and turn there and you're like well I don't know where's I'm driving right now I don't know where you're looking I don't know where you're pointing I don't know where there I I don't know what what turn means. And someone who's really good at giving directions will say like, "Okay, you're in the middle lane right now. In about four blocks, you're going to be going right. So, as you have, you know, opportunity, make your way over into the right lane and then I'll let you know with, you know, one block warning which street you want to turn right onto." Like, and you see what I mean? Um, and that is the skill set that is going to be so important because yes, as AI agents get better and better, this structure, you'll be able to give it more and more vague

instructions, but never never will the AI be able or I don't know about never, but like it's going to be a long time before the AI can just read your mind and produce exactly the output that you want. You are going to have to package up what you want it to accomplish and communicate it in some way. And that's the key skill. Yeah. No, absolutely. So I I deal with this a lot in uh tutorials. Yeah. Right. Um and actually I'm working on getting like more scripting team members and stuff like that. At the end of the day it comes down to explanation even for them. And like there's a difference between an overview and then like a tutorial. And if everybody and and by the way when I started doing a lot of tutorials on a consistent basis I got better at thinking structurally

cuz I'm like god you have to explain everything to people and then I'm like oh but that is how these things work. And that opened up my mind to automations and then it opened up to my mind to how to explain things to AI agents. And you mentioned Lovable. Um I think they actually just released a new version that claims it's like 10 times smarter or something. I mean I have to say I'm super impressed with them as a company. It's super impressed. Yeah, I'm super impressed. They're amazing. It's amazing. Yeah. So I I guess uh if you had to give any advice then to to small business owners kind of closing things out, what would it be? Would it be this structured type thinking? Um, in because I I personally think this is actually I learned how to be like effective at work, I

got into my whole like self-improvement, productivity nerd phase. Yeah. And then after making a million tutorials, I see things in uh sequential processes now which um allow me to automate AI. So yeah, what that would be my advice. That's the key for small business owners like take a very structured view of everything you are currently doing or wish you would be job? And so, for example, let's say again, let's use a trades person example. Let's say you're a contractor and you want to have someone help you out in your back office do estimations. And you would not just say to a college intern, "Eestimate how much this job is going to take. What college intern could possibly take that prompt and give a good estimation of what a construction job is going to cost? What you do is you basically say, "Here's how a construction

job cost estimation works. There's labor and then there's materials and then there's buffer because things always go wrong. Here's how you estimate labor. Here are a bunch of historical examples of similar jobs and how much labor was required both in how many hours and you know what the skills of the individual trades people were and what hourly rates they bill out for materials. Again, here are 10 past jobs and all the different materials they required to do this thing. Here are 10 past jobs and here are the items that came in kind of late or overrun or became buffer. And then your intern has a chance to succeed, right? If you've given that that that kind of structured approach like here's how we estimate a construction job. here are 10 examples of construction jobs in the past, then they're going to do a pretty good

job probably of creating an estimate based on what you give them about about about this job. And so that's that's the the thing I would think about as small business owners. It's like it's not going to be a magic thing where you just write a one-s sentence thing and like the AI does all your work for you, but the AI can do an amazing amount if you give it the right instruction. Yeah, I think I I think that's a that's a great thing to close off on because AI as much as it's going to help uh this year as we and this is my full circle call back to your uh LinkedIn post. It is only as good as what you give it. Yes. And if you give it nothing, it will give you slop. If you give it something, it will give you uh

a LinkedIn profile people don't realize are uh augmented by that is augmented by AI unless the guy admits to it then gets uh I was just defending I'm defending a buddy real quick because I was I saw them and I was like man I got to acknowledge this in some form of content so I feel like this is the natural way to do it. He and I serious you have a great follow. So it was just so funny that that that one thing they like went off and like everybody likes what he's posting. Yeah. It's funny by the way, even on that post, I got a bunch of feedback that people found it really really useful. Um, but uh and you know, my my learning there is like, you know, it's the classic social media incentive system. If you stoke a little bit of controversy,

your posts perform way better. Well, don't hate the haters or don't don't love the haters. Don't hate them. Um, so with that being said, thank you so much for listening to this episode of the AI Agents Podcast. Make sure to go check out relay.app. It is an AI agent for the future. And if you like this episode, make sure to leave us a like on Spotify, Apple Podcasts, and the works. Thank you so much for watching. See you in the next one. Bye.