AI Agents Built My Business While I Was at Church

I had a new business idea. Of course, I do not have time to implement another business. So I put AI agents in charge of starting this business and see if I could get it all spun up, get AI running it in just one evening. So my idea was a Google review response business. Three AI agents found 145 Austin restaurants that had unanswered reviews. People left reviews for their restaurants and they hadn't responded to these reviews. Of course, businesses are busy. If you're getting lots of reviews, you're a busy restaurant, you don't have time to go through and answer each review. But answering those reviews, both the positive and the negative, can actually really help your business. But again, people just don't have time to do that. So let's put AI in charge of answering those businesses and then we can sell that service to these restaurants to help them grow their business. So I wrote AI agents that found businesses that had unanswered reviews. It wrote 777 responses to these reviews. And then it sent the completed work to these businesses and kind of pitched a service. So again, this is a business that I just put AI in charge of. I put these AI agents in charge of running and it ran away with it. So again, always with business, you're looking to solve a problem. So the problem is 93% of customers read reviews before visiting a business. But most restaurants can't keep up with the responding to these reviews. And every silent review kind of tells potential customers that you don't care about feedback. And that can hurt the number of people coming to your restaurant. So I built three AI agents using Twin. So this is an easy way to build AI agents where you don't have to know how to code. You don't have to download cloud code. You can do this all online. But agent one that I built finds businesses that have a lot of unanswered reviews. Agent two writes responses to their unanswered reviews. And then agent three sends that completed work to that business and pitches them, hey, you should pay. I haven't even decided on a price totally, but you should pay $150 a month for me to handle your review management. So again, I used Twin to build this. So if you saw, I did a Minecraft video where, you know, these twin agents run 24 seven, they can fix things when they break, no coding required. Anybody can do this. This is different than I build my own agents on cloud code, which is still pretty easy, but there's a, you know, a little technical knowledge you need to download cloud code and to get the agents going. Twin kind of takes that step out of it. So this is great for people who want to build agents, but don't know how to get started. So the best part is these agents run even while you sleep. So you can have work going in a business going while you're asleep at night. I love it. I love automation. So I want to kind of walk you through how I built this again. The goal is not for you to copy this business idea completely, although you could go use it in a different area. I did Austin already have reached out to a lot of these companies. This could be a potential business. Go do it in your local area or come up with your own business idea or use twin to help you come up with your own business idea. What's a problem? How can you solve it? And then you connect the problem with the solution and that's a business. So first I went into twin and I had this business idea already. You could work with twin and say, Hey, I want to start a business. What are some ideas you have and talk through it with that? I came with an existing idea, but I said, I want to build a Google review response business. And the first thing I want to do is search Google maps for restaurants in Austin, Texas that have 50 or more total reviews for each restaurant. I want the exact business name, the address, the phone number, the website, total review count, and count how many reviews have no owner response. And then flag any business with over 20 unanswered reviews as high priority and put this all into a Google sheet. So that is a task that I gave to an agent. Again, this is all stuff I could do manually. Like if I wanted that, I mean that would take me probably a week, but if I wanted to, I could spend hours and hours and hours, go do all this myself. But again, busy, busy boy. So I'll have an AI agent. I'll tell twin, Hey, go make an agent that does this. So first off it flatters me. This is a smart business idea. Thank you. And then it maps out what it's going to do before it starts building. So you can take a look at it if you want and say, is this kind of the plan that you're looking at? So it came up with a plan to look at Google maps. Again, I could do it manually, but there's twin knew about this. There's a tool called Appify that can go scrape Google maps. So you don't have to go through it manually with your eyes. You can do it with code. You can do it online. So they're going to use Google maps scraper to pull restaurants in Austin, Texas with 50 plus reviews for each restaurant, extract out the business name, the address, phone website, total review count, what I asked for, and then also scrape to get count reviews with no owner response, and then flag the businesses with 20 or more unanswered reviews as high priority, and then put it all into a Google sheet with fields and a priority flag. So it gave me this plan and then I clicked build it and it took, it took a while. I'd say it took probably about 20, 30 minutes for it to go through and do all this. But again, what I love about AI agents, they're not faster necessarily than humans. Maybe, maybe they are, maybe it's a little faster, but I maybe could have done this manually, or I could have probably written the script myself. You know, I know Python. I could have written my own script that went and did this, maybe done it a little faster, but the beauty of AI agents is not that they're faster, but that I can get it going. And then I went to an Easter play at a church and this was just running in the background while I was there. I didn't give any thought to it. I just came home and the work was done. So this is kind of the lead pipeline. So this will give me a list of potential customers that have the problem that I'm looking to provide the solution for. Now, just on a, on a business note, this business is going to be a little tricky. I don't imagine that customers are going to just pour in because this might be a problem they don't know that they have. The best businesses are where you're solving a problem that they know that they have. It's a pain point that they feel and they're actively looking for a solution. This is closer to, you know, I talk about painkillers and vitamins. This is closer to a vitamin because it is a problem, but they might not understand that they have this problem. So you have to convince them first, you have a problem and I have the solution. That's all, that's a harder ask. And I'm getting ahead of myself a little bit, but what you can do is phrase it around a problem that they do know they have. Do you want more sales? Every business owner, that's a problem. That's a pain point they have is, Oh, I want more sales. I want more customers. I don't know that I know any businesses that are like, Oh, I've got as many customers as I ever want. It's probably not a growing business, probably not a business that's going to last if that's their attitude. So instead of saying, Hey, you have unheard Google reviews that haven't been responded to can say, Hey, do you want more customers? And that again, attaches your solution to a pain point they already feel, but I'm getting ahead of myself. So the plan is you get a list of leads, and then agent two is going to be an outreach email generator. So it takes each of those leads. It generates an email for each of those leads. And then agent three is an automated email sender. So then it contacts those businesses and says, Hey, look, I already provided this solution to your problem. So, and when this agent got done, it says the summary is a bit unexpected. It looked like the builder got ahead of itself and built both agent one and agent two in one shot. So I had kind of intended to building this as three different agents, kind of one at a time. That's how my brain thinks. But AI, again, sometimes has a mind of its own. And if you don't want it to do something, you have to be pretty specific and direct. Hey, don't build agent two. But AI thought, well, why not just build the email along with the leads list? So it built agent one and two in kind of one shot there. So this agent went to work. It did 439 things. It's actually, I'm going to take it back. This would have taken me a lot longer than 30 minutes to do all this, but it went through, did everything I asked and it found 144 high priority restaurants. So 144 restaurants in Austin that had more than 20 unanswered reviews and it put them into a spreadsheet. So, and it made the executive decision. So it put business address phone exactly. Well, like I asked, put the website total reviews. Now it made an executive decision, which I'm okay with. AI does this sometimes though. It decided if there was more than a hundred to, it's not worth counting all of them. If you're going to count them all, you have to go page by page. And it just is going to add a lot of time. It's going to add a lot of credits. And they, AI decided that if there's more than a hundred, we don't care if there's 140, 127, stop at a hundred. So again, AI will make some decisions for you if you are not verbose in the instructions you give it. In this case, that was a good decision. I'm glad it took that shortcut. I don't actually care if there's 101, 147, a thousand and 47. If there's more than a hundred, I'm okay with it. Just saying there was a hundred unanswered replies. If there's more than a hundred unanswered reviews, you can just kind of assume this business does not respond to reviews, which means it's leaving money on the table. So actually I skipped a step there. Step one, it went through and found businesses that had enough reviews. You know, if there's one or two reviews, this is not a good target for our business. Step two is then it ran again, looking for unanswered reviews. So step one was just getting all the businesses in total reviews. Step two was it got the number of unanswered reviews and then it put, if there was more than 20, it says it's high priority. So like Panera didn't count because they didn't have any unanswered. Well, shout out to Panera, 672 reviews and they were all replied to. Maybe they have AI doing it. I don't know. So then this AI agent also did another thing. It wrote reviews for all of the unanswered reviews for these businesses and put them all into their dedicated spreadsheets. So, and it did this by spinning up another agent. So that's what you have going on in AI these days is an agent created another agent. So it spun up another agent that's going to go do this work. And this agent loaded the leads that we had, found the high priority restaurants. So it went through the ones that unanswered reviews that our first agent got. It scraped the reviews. So it loaded up all the reviews and actually it stopped at a hundred. Then it sent the reviews to an LLM. So to like chat GPT. I mean, this was the prompt. So AI created the prompt that it sent to chat GPT to generate the response to these reviews. So you are a professional restaurant owner writing genuine human responses. I mean, there's, you know, let me know ethically. This is where ethics kind of gray areas here a little bit, but always right. And first person is the business owner. Keep responses under 50 words, sound warm and human, never corporate or robotic for, and they gave instructions. So this was AI generated instructions. I could give it my own instructions. I could say, no, I want to handle reviews differently, but for four and five star reviews, thank them for one to two star reviews, apologize. And then never use like these cliche terms. We value your feedback or thank you for your review. AI determined that was too generic. So again, this is pretty good instructions. Again, if I wanted my own instructions, I wanted to handle it a different way. I could have interjected myself, but this was just AI's take. And actually AI did all this again while I was away. I could have monitored on my phone or something, but part of the point of this was to just get AI to run this business itself. Again, my brain's going too many different directions. AI go ahead and take over. So I'll come up with all these reviews, save them to a sheet for each one and continue until all the restaurants are processed. So what we ended up with was like a hundred Google sheets. This was one for Sushi Bang Bang, but it has all the reviews, their rating, the review text, and the generated response of what we would want to say back to them. So it made a hundred of these spreadsheets. Again, that's, I'm not going to do that. So agent finished that. Then the next step, this is where the business actually starts making money. Let me think through the outreach agent before we build. So now it's going to build an agent. We have all these responses. Now we want to contact the businesses and say, Hey, look, we made responses for all your Google reviews. Did you know you're leaving money on the table? Are you interested in our service? And again, if I'm going to turn this into a real business, which let me know, comment, if you want me to turn this into a real business, probably need a website rather than just emailing people and saying, Hey, I'm Jay Pelton, 722 at gmail.com. Are you interested in this? Probably will work better if I make a domain name, make a website, get an email for that domain name. And again, AI could help me out with all that. If I decide I want to move forward with this. So, but this agent too, it's going to look for the high priority restaurants, craft a personalized cold email, mentioning their business name, how many unanswered reviews they have and the risk of ignoring them. This is where if I'm going to do it again, I probably would change the direction of agent to a little bit to focus more on how much potential money they're leaving out or how much more potential money they could make. If they responded to their reviews, you can maybe look at studies. You have to maybe do a little work there or have an agent do a little bit of work there, but just in my brain, it might not connect to businesses that not answering reviews is a bad thing and low lost trust and lower ranking. I don't know that maybe that's compelling for people. We'll have to see some of this might be testing. Do businesses respond to that? Businesses are very busy. If you're going to pitch something, especially something that's going to cost them money, you better have a pretty good pitch. I do like this solution where if you're going to cold pitch someone, you provide the value first. You don't say, hey, there's this problem and I could do some work for you, but we're actually going to send them the sheet of here's all the reviews that haven't been responded to. Here's the response already given to you. We're actually giving them the solution before we've even asked for any money or anything like that. Then the ask is, hey, if you don't want to go through and actually post all these responses or create your own agent that goes through and post all these responses, I can handle that for you. And again, I don't know the price value of that. I probably could do some research. The AI agent said it's worth $200 a month. I don't know. I don't know if people are willing to pay that businesses are pretty much worth what people are willing to pay for it. So the agent lead with free value, say keep it as a gift so they can keep the spreadsheet if they even if they don't work with us and it's hyper-personalized. So this is not just this mass outreach cold template, but it's like work has been done for each of these businesses already. So, and I have my email connected in jpelton722 at gmail.com. Shoot me an email if you want it, but it's writing the emails from jpelton722 at gmail.com. So, and it's starting with just 10. I appreciate that. I'm glad. And again, I probably should have maybe specified this, but I'm glad AI was smart enough to think about this. I don't want to just send out, you know, 300 emails all at once. Google's probably not going to like that. I'll get my Gmail banned again. It's been banned before for different AI things that I've done, but it's starting with just 10. So that's a good way just to check it out, you know, make sure that it's working the way that we expect. So again, this agent reading 10 uncontacted high priority emails, it actually looks on their website for the real contact email, gets the review Google sheet that the first agent created, crafts a personalized email for them, sends to the restaurant, real email, logs every contact, and then updates the master sheet with the contacted date. So then this ran and there was a little bit of a problem. So it was only able to find three of the emails. And then one thing I noticed is like IHOP, Taco Bell, some of these are big companies where it's going to be probably pretty hard to like get in touch with them directly. So again, I might want to hone this agent a little bit because out of the 10, it actually only sent three emails, but these more local businesses where you out on the website, you actually have the email address of the owner. Those are probably better leads. Anyways, I'm going to have a hard time pitching this to IHOP or Taco Bell or big city bagels, but curry and hurry, which is an in Austin, it's got a Gmail address in, you know, six 20 Austin. Those are probably pretty good leads. Those are real people. They'll probably see these emails. So again, that might be one thing I hone them, but that might be one thing that I think AI did poorly that I could improve on, you know, and if I told AI, go ahead and improve the quality of the emails you're sending, it might come up with that its own, but again, my human brain can come in and say, Hey, don't send to business. Don't send to franchises. That might be an easy way to get past some of this. So, and again, my main agent, kind of the boss agent, he created these, this email outreach agent. So I kind of have three different agents here now, and this is the, it's, it's kind of nice. Cause you can look into that specific agent and give it specific instructions. So the email sending agent read the master sheet checks for duplicates smart. I might've forgotten that, but it makes sure not to send, you know, you might, for some reason or another, you might end up with two on the sheet that are the same and you don't want to send them to emails. That's very bad. It gets the response sheet, finds contact email, which we could maybe put more work into this. It just did a web search. Like what's the Google search? What's the email address for this business? We could maybe tell it, you know, scrape the website. Oh, it did try scraping the website. Okay. Nevermind. Good job AI. Then it got into my Gmail, which I gave it permission to do that. I could have just said, give me a spreadsheet of the emails to send and I could do it manually. A lot of times what I do with AI is at the beginning, I do more manually just to make sure I know it's working and that it looks good. And then once I'm confident that we have a good system down with AI, then I'll give it more automated access. This is actually very foolish for me to do it from my main email address. I definitely should have spun up like a new email address to do this. This is kind of risky. So, but then it keeps track of each email that it sent and then it updates the spreadsheet here. So it added these fields, contacted and what day it was contacted. So you can see the first 10, it did go ahead and try to contact. So then what's next? It gave me actually some options, which again, I love how AI agents do this as it kind of says here, what do you want to do next? So the first thing we could do is schedule both agents to run automatically. So just get this going, like have it run overnight. Again, I want to be a little careful because this is coming from my main email address. If I'm sending a hundred emails, Google's going to start seeing that as some spam. So I want to be careful with that. Another suggestion was build agent, another agent that is a follow-up sender. So if you send people one kind of cold email, they might not read it. But what if after three days you sent them a follow-up? You could expand to a new city. I did Austin, Texas. It just popped into my head, but it said, why not do Lincoln? Your home turf. Number four, it could improve email finding. So you could scrape websites and look at Yelp. Maybe Yelp would be better. They could build a dashboard to track your leads, your outreach, your status and replies. I could do all of these things. I could have an agent. I could say, have an agent do one, have an agent do two, have an agent do three. And again, let me know in the comments, if you want me to keep going with this. Happy to. But what I really wanted to show you is that I had a business idea. I typed in one thing. This is the only thing I actually typed in, was here's my business idea, get going on it. And by the time I got home, then I started this agent on it on twin. And then I went to an Easter pageant at a church here in Lincoln. Then I went to a friend's house. We ate dinner. We played some games. When I got home in the evening, this is what was here. This was all done already. So I just want you to see the power of that. I can go from a business idea to a business. I mean, this is the start of a business. Again, I don't know if it's going to be a successful business. I always say you start 10 businesses. Probably six are going to fail miserably. Nobody's interested. Two or three are going to probably be like, oh, this is decent. There's a couple people interested. I have to hone the idea a little bit. And then there's maybe one that's going to turn into a real business. But we can start businesses faster than ever before. I just want to show you the power. If you're not using AI agents, link down below to twin. Go ahead and check it out. You can start for free and just try it out. You can ask twin for a business idea. Maybe you didn't have a business idea like this. I've been, it's a blessing and a curse, but just business ideas pop into my head constantly. But I love that I can get an agent going on it. Before this would have just killed me. This business idea probably would have been in my head. And I would have thought, oh, I've got to at least try it. I got to see if there's something there. Are people interested in this? It would have eaten away at me, the untapped potential there. But now I can get AI going on it. And at least I can fail early. I might send out, 10's not enough for a test. And actually it only sent out three. So I need to probably send maybe 50. But if I send out 50 of these and I get all nos or no responses, it's probably more likely, well, maybe I need to either try something else or maybe this wasn't a good idea. Maybe this is not something anyone's interested in. I know in the comments, just some business ideas you have and just get AI going on them. I know you're busy. Well, we all have to-do lists. My to-do list never gets finished. Okay. I have a to-do list every day. I get usually about 70% of the way through it. This is one of the things that would probably not have been done today if it weren't for AI agents. So, and again, if you don't, you can build your own, you can download Claude code. I'll leave a link to Claude code and go build your own agents. But it's nice to have a service like twin. It's just online. It's just like, it's just like talking to chat GPT, except for it's creating agents to build your business and not just build a business. You can use this to just improve different areas of your life. If you're an employee, you know, think about the workflows you have in your existing job. And it's like, well, maybe some AI agents could take this part of my job or take this part of my job. So go check out twin. Let me know in the comments, if you want me to complete this business, I'm happy to kind of show you the results, probably hone in a little bit, get my, get a different domain. So I'm not risking my J pelton seven, two, two at gmail.com, which I've had for like 30 years. But let me know if you'd like to please like, and subscribe again, 90% of people who watch these videos aren't subscribed. So if you're not subscribed, please hit subscribe. It just, it makes me feel good about myself. So appreciate you very much links all down below. Shout out to twin. God bless you guys and have a great rest of your day.