← Back to search

Personal Agents in Ecom

Underdog Ecom Podcast · 2026-11-13 · 38 min
relevance 66 8600 words Episode page ↗ Audio ↗
Show full episode description
In this episode, Jack and David dig into personal AI agents: what they are, how to set one up, and what they can actually do for a small ecom brand. Jack walks through his Hermes agent running on his Mac mini. It’s connected to Slack, Shopify, Klaviyo and his team’s project management tool. It handles returns, buys international shipping labels, answers “where’s my order?” tickets, and turns creator videos into invoices. David is setting up his own agent after his part-time VA moved on, and he asks the questions every founder will have: where do you start, and how do you stop it leaking financial data into a customer conversation? They also cover guardrails and sandboxing, how bad actors might try to trick AI agents, and whether you’re running an ecom brand or a software shop. If you’ve been wondering where to start, this one’s for you. 📢 Enjoyed the episode? Thanks for listening! Leave a review on your favorite podcast platform and share it with fellow ecom founders! Your support helps us keep it real. 💬 Join the Conversation: • Jack on X: @jackinprogress • David on X: @austriker27 📩 Got feedback or topic ideas? Leave a comment or email us at [email protected] 🌎 Join the Underdog Ecom Community: Apply here → underdogecom.com 🎧 More from Underdog Ecom: Website: https://underdogecom.com Podcast: https://podcast.underdogecom.com
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
How solo ecom founders replace VA busywork with a personal AI agent running on a Mac mini that reads, reports and acts across their store stack.
Benefits
  • Replaces part-time VA work with an always-on agent
  • Single agent in Slack with read access across finance, Shopify, Klaviyo, Amazon
  • Browser access covers tools without MCP
  • Automates returns, international labels and customer service replies
  • No per-seat cost for the agent's helpdesk account
Use cases
  • Hermes agent 'Kit' pulls Finaloop, Shopify, Klaviyo and Amazon data into Slack reports for the Groundwork brand
  • Kit manages Groundwork's custom project-management tool via MCP: creates briefs, reassigns tasks, flags stuck work
  • Processes Shopify return requests end-to-end: accepts return, creates label, sends to customer
  • Buys cheap DDP international shipping labels in Shopify while ShipStation handles US
  • Acts as a customer-service agent in Commslayer checking warehouse pick/pack status; replaced Commslayer's built-in AI
KPIs / results
  • M4 Mac mini with 24GB RAM bought refurbished for under £600
  • Six out of 10 businesses estimated to be using AI agents
  • Podcast past episode 50
Tools / build
  • Hermes agent 'Kit' on Mac mini M4 (Slack interface)
  • Groundwork project-management tool with custom MCP
  • Shopify / Klaviyo / Finaloop / Amazon MCP connections
  • Commslayer customer-service agent account
  • 1Password + dedicated agent email for logins
0:00 / 0:00
Right now I've got an autonomous routine in Claw that runs every Monday and checks my Facebook ads. I've kind of stopped using that Mac Mini so much myself and just left it to it. I've turned off the AI and ComSlayer because they kind of compete with one another and it kind of just does it better anyway. Bad actors are just going to be like, we know that six out of 10 businesses are probably using AI agents. It's pretty wild. All right, welcome to the Underdog Ecom Podcast, where two Ecom brand founders are figuring out in real time. Yeah, we're not here to preach. We're here in the thick of it, making mistakes, testing all sorts of ideas and really sharing the journey of how we've been growing these brands from day zero. Hopefully you learn a little bit, perhaps come away with some ideas that save you time and money. And so if you're tired of the same glittery DTC advice from those nine figure gurus, you're here in the right place. Let's get into it. Okay, we're back. Underdog Ecom, here we go. I think we're past episode 50 at this point. And it's been approximately seven days since we mentioned the keyword AI. And so we're back here to talk about it again. I'm excited for this. I've actually got a special delivery coming today, a little Mac mini. And so basically today we're going to dig into personal agents, what that is, what people are doing with it. Jack's built out one. I'm excited to dig into this and build out one for myself and my brand. So, hey, Jack, how are you doing? Good. I'm good. Yeah, I'm excited for this one as well. I think it's, I think we've just got into time suddenly where AI can do suddenly so much. And all those things that you thought were a bit out of reach are suddenly you can do. And it's pretty cool. So yeah, I'm excited about this one. Limitless. It feels like that movie that I never actually saw, but we've all got the pill. We're, yeah, I was, there's people, trendy people online. They like red pill. There's not, I don't know. I think that's a matrix reference, but it's crazy. Yeah. Just anything. And so setting up, I've, I've been running agents for a while on my computer and I've got a MacBook Pro and it's pretty fancy pants and it does a lot, but I've been building various bits of software. And so it's basically been under siege for the last couple of months and I've got, it's been maxed out. I've only got 24 gigs of Ram on it. It's a year too old. I'm three or something. And, uh, I am building an iOS and Android app. And so those are real big hogs. So in addition to using this Mac mini for the personal agent size of things, I'm definitely going to use it for, uh, as a software dev as well for building the various things that I'm working on. And so it'll be pretty sweet. I'm intrigued. So I don't know, Jack, maybe you want to kick it off and talk about what you've been. Cause you've, you tease this to me a couple of months ago. This is why this is so cool. He teased this to me a couple of months ago. It was this Hermes agent is in all his stuff, doing all his things. And it's so nice because I feel like it abstracts the busy work out in a lot of ways and you can set up systems and stuff. I have had a VA for a couple of years that helps my business part-time and have always felt like I could hire a full-time VA and should hire a full-time VA cause she was just part-time. And so she, it's perfect timing. Actually, she just left my company basically was maxed out on hours and had to go quit. And so this is like the crossroads are, I'm at where it's like, all right, time to probably replace her and then some with a computer that can do things. Hello bot. Hello how? So that's where I'm at. Yeah. That's really cool. I think, um, yeah, I see the right time. To be honest, I got a Mac mini. It's an M4 24 gigabyte Mac mini because, uh, back in actually back in March. Yeah. It was a while ago. I remember March or April, something like that. And I got it just refurbished when minis were super cheap. They've gone up a lot now, but I got them, got it super cheap. Uh, it was under a thousand. I was, it was, yeah, it was, it was under 600 pounds GB pounds, which is probably a hundred dollars. 700. Jeez. Yeah. Yeah. And so 24 gigabytes of RAM, more RAM that's in my Mac books, my 16 gigabyte. And I feel it a lot actually when I'm swapping between codecs and Claude and then Chrome and then, uh, Photoshop. 7,000 tabs in Chrome. Yeah, exactly. I don't know. Um, so that, so I was like partly as I, I'm hitting the limit on my Mac. And then also I was kind of getting interested in this when open claw and stuff was more, and I actually never installed open claw. Um, because I was thinking about doing it, but then I was like, I saw, I saw Hermes was also around at the same time. And I was, uh, just kind of getting started with it all. And I was also just kind of building a lot of the AI stuff I was doing there in on the Mac mini, uh, keeping it separate from my Mac book and just remoting into it or SSH into it. And then, yeah, that was all I was doing. I'm just kind of learning. I knew nothing about any of this. So, um, and it's got better. It's got easier and easier over time because there's more and more people are doing this stuff. A lot of it's like much easier to remote access and, uh, all this sort of stuff. But anyway, um, yeah, I think we're all a lot of many of us solo operators or very small teams. And so this has a real big impact, I think on us specifically, I think it's easier for us to implement as well, because we're not trying to diffuse it across 50 or a hundred people, which I think is a whole nother challenge, right? How do you get, I've heard about companies buying minis for each of their employees, but it's like, well, how many of them are actually going to do something with it is also the question, whereas you and I kind of just get stuck in and it's a lot of fun. But anyway, yeah. So I don't think what I started with, I started with, uh, this is before a lot of like the browser access was available. So before a lot, before models were very good at actually using browsers. So it was a lot of those MCP stuff. So I would, I linked it up to everything that had an MCP piece like Klaviyo, Shopify. And I started off with just getting reports and add it to Slack. That was the first thing I did. I had it in Slack. And that's the way I've always communicated with Hermes. I know you can do it in terminal. You can also do it through maybe through iMessage, but I just, I'm in Slack anyway. And I thought my team, I've got a couple of people work for me. So they're in Slack. So I thought I could have it there. And so I had it originally for myself and then just linked it up into certain projects I wanted to have access to. And that was it to begin with. And that was pretty good. But so it's all just kind of been building up. So over time, so, um, I've got it, yeah, I've got it linked up to all of Groundwork's financial stuff in sense of not only read access, but read access to Finaloop, to, into Shopify, into Klaviyo, into Amazon, into all this different stuff. So I can get a clear picture of where things are. Um, and it kind of just reports on the stuff like that. And that's kind of fairly simple. But the other thing I've done as well is I built, uh, I've spoken about this a few times. I've got a thing called groundwork, which is essentially just project management thing that I built for, uh, my team to be able to manage all the emails and things. And it pulls through, uh, Klaviyo campaigns and how they're doing. And also briefs are in there and statuses and all that sort of stuff, comments. And then now, oh, I, the last few months I've had an MCP for that. So now that goes into Hermes and Hermes can, the Hermes Asian I've got is called Kit. And so Kit, uh, can go into groundwork on its own accord. It can, uh, get a gist of all the different things that are happening there in terms of different emails, if anything's due or anything like that. And it can also create briefs. It can change statuses of a task. It can assign a task to somebody. It can let me know if a task is stuck. If I've forgotten about something, it can notify others as well. And again, it's in the channels in Slack as well. So it can go in and check and do things like that without me having to do anything. Originally set up with just a cron job, but now it's now it just does it all on its own without even that. So that's really cool. Um, that's been really helpful because then tasks and things are now kind of being monitored and as, as Astra came out and things like that, and there's all this, now it could go into browsers and use the Chrome on my Mac mini and do all this sort of stuff. I've been able to actually do a lot more because it can now, even things that don't have MCP access, it can then go and do them. Um, so that's been pretty cool as well. And I really got, as we saw say Muse come out and things like that, uh, instinct, uh, I kind of realized like, well, Hermes can actually do all of that. There's nothing that any of those, the only thing that Muse has got is access into Instagram. And it's got this, you can buy, I can buy things for you. I haven't given Hermes the ability to buy anything for me. I it's actually possible, but I just don't feel like I need to make that many transactions or these small transactions where I would give to it, where I would make sense. But yeah, so, um, that's the main thing I think is being able to go into go and just perform tasks for me in the browser. And basically I've kind of stopped using that Mac mini so much myself and just left it to it. So it's almost like its own signed in on its own profiles. The other thing is I set up with an agent email. I'm trying to remember the name of the provider I use for that to give it its own agent email address, but it can log in to stuff. So it's got one password access. It can log into different things. The only thing I'd say in terms of actual workflows, it can, when we have return requests in Shopify, I've set it up so that it can go into Shopify, create the return label there, accept the return, go and create the return label there. And then send it, and then it will just get sent to the customer. The only thing that sometimes happens is Shopify will make them do lots of authentication. And so if it's not logged in recently, then that can cause a problem because it's not very good at authenticating itself through one password. And I haven't figured that one out yet, but that's pretty cool. And then also international orders. We do quite a lot of international orders. It can just go in and buy the labels in Shopify because our warehouse buys the international labels through Shopify, even though we use ShipStation for US. And that's because the international labels ship DDP and we can get cheap DDP labels through Shopify. And so it's easy to ship internationally, whereas it'd be more expensive through ShipStation. And so it just goes in and buys the labels. And then it has access to our warehouse as well. I forgot to mention that. So it can go in there and it can attach labels in. It can check on order statuses. And the last thing as well is customer service. So I have Kit as a agent in customer service, and it can respond to customers. And usually if it's like, where's my order? It can go into the warehouse. It can see if that's been picked or packed and it can let them know what's going out today. Or you can see if there's a tracking number in Shopify or in the warehouse. Or if there's a replacement we'd send a customer because it's lost or whatever. It can go and create that replacement. So it checks their order. I've actually turned off the AI and comps there because they kind of compete with one another and it kind of just does it better anyway. And it has access to all these tools. So yeah, I think that's been really cool as well. Just performing those kind of mundane ritual actions has been really helpful. That's so interesting. I was curious about the comms layer versus the comms layer AI because I've got that going on my storefront and it does pretty well. And I've found that auditing it and improving a lot of the docs and protocols or whatever they are called procedures via clod is super helpful. But because your AI is everywhere, it sounds like that's much more effective. Is it just a user account that it logs in with into comms layer? Yeah. Or it's basically like a person kind of thing. It is like a person profile. And I have to be careful because it's not a person, right? And I don't want people thinking it's a person. I have to be careful with that. When I first set it up, it was replying on behalf of me. I was like, that's not going to end well. So I was like, I need to sort this out somehow. But yeah, I gave it its own account. Basically, it just acts as a normal user would. And nice thing about comms layer is it doesn't charge you per user. So yeah. Oh, yeah. Nice. Love the non per seat. Yeah. Because that's seems like way too common. That's oh, gosh, that's so interesting. There's so many pieces to pull on there. And I think having the glue that kind of there's there's so much ops that we do. And there's so much sounds like it's very much a project manager. There's so much project manager. And there's so much like executive assistant kind of things that it can do. I'm really excited. Basically, I use Hey for email, for example, I use Basecamp for project management. And so using the CLI tools and getting it access to all these things where I'm sometimes right now, there's a huge chain back and forth between me and a B2B person, for example. And I just have I go into the CLI and like, hey, summarize this, give me a draft email of what's going on, what needs to happen, because there's too much back and forth. I can't track all of this. And having someone that can monitor those emails, go into Basecamp and update things, go into I use settle for a lot of PO type stuff. And they're slowly rolling out an API and MCP, which is good. Oh, amazing. Because I was on the verge of building my own to replace it that has agent access, because it's like, if you run a SaaS nowadays, you have to have agent access, there has to be an MCP and API. Yeah, because there's so many things, especially for a lot of these ops things, and especially for brands that are tiny, like us, where it's like, I'm not gonna have time to manage and babysit all these POs in Shopify and settle and stuff and get all the records updated and everything. Like, that's actually one of my tasks for today is I need to reconcile all these emails back and forth that I've had with Finaloop about a lot of the POs that are in the different systems and what's been paid and what's not. And it's, gosh, it's a lot to try and just sift through. So it'll be, I'm really excited to have someone that'll sit in, someone, have an AI bot that will sit in and be able to track all that and basically babysit it in some ways and manage it. Yeah. And I think one thing I haven't worked out, and I think maybe you'll, I'm sure you'll work this out better than me is, is how you have this, uh, this knowledge across all these areas while also sandboxing some things. So they're not, for example, I don't want the fact that it can access Finaloop or whatever. I don't want that ever leeching in customer conversation because they prompted it in the right way or whatever. Right. If somebody were to figure out that it was responding to some customer thing, it could, they could potentially ask it to give details that it might give. So I tried really hard to sandbox some things, but you have to be careful there because you also want that. You also, it's also nice actually having one agent versus many, because then you just ask it everything. And it's got context across all these different things. Whereas a customer support agent will only know about that. And actually there might be, you've got to make sure you ask the right agent and got to make sure the way you like that hierarchy. I haven't worked out yet, but what I've done at the beginning is just really try and sandbox them out with different profiles and give access only to certain things or some of them. And that's been kind of the best way I think, but there's also one, there's also a time where you just want it to know everything. And so pull through emails as well as the PO from, from settle, for example, and then figure things out for you. Whereas, and it needs access to a lot of stuff and knowledge across those things to do that. I think I disagree. I don't think that there should be in my mind specifically, I've been thinking about this for a coding setup. For example, I'm going to have a senior dev that's in charge of running the projects. And currently what I do is I have one agent that's the orchestrator running different sub agents and all the different features and they're reviewing each feature. And then the orchestrators passing context between the two. I've also set up what I call a librarian. So I have multiple projects that are under one, one main project. For example, my ITMO has a website, a Laravel backend, an Android app and an iOS app. And so I've got a librarian set up in each different project folder. And I tell the agent that's in Android, for example, go talk to the librarian, figure out what the backend is actually doing and hit those APIs accurately. And I have docs that are provide context as much as possible throughout the different apps and different project folders in Markdown. But it's, I think more helpful to have the agents talk to each other. And so I think you're totally right that the big danger with all of this is the cross-pollination and getting like a customer could game that if they know what they're doing and they're really good at what they're doing, how to game a AI agent, they could probably figure it out and do it. And I've even setting up guardrails, even setting up instructions of like, don't do this. I've seen, I've seen them just burn that bridge so many times, like weekly almost where it's like, I told you not to do this specific thing in code. It's in your instructions not to do this and you still did it. And so there's a lot of dev security around AI in the dev software engineering world where it's like, don't ever let them touch your production database. For example, they should have read only access their token. It's not that you tell them not to write to it. It's that you don't allow them to even write it. Yeah. Yeah. It's like they only have read only. And so it's tricky because like, you want them to be able to change things and do that. It's like this balancing act where you're like, you want them to do all these different things, but you also need to have guardrails on because at the end of the day, they do make a fair amount of mistakes. And yeah. I was going to say, yeah, you're right. I think you're right. And I think the only, I think you're absolutely right. I think with, if you think about it with a person, even you wouldn't give your customer service person like unlimited access to your bank account. Right. I mean, they're different people in organization and they have different. Yeah. It's that context that you want to share, I guess. And that's how you share certain bits of information or not. And it's hard. But yeah, you said with customers, there are customers that can game it, but I think the bigger risk is more and more as it's more generally going to be normal for AI agents to be doing stuff on behalf of businesses. I think bad actors are just going to randomly prompt, prompt, inject, send, send requests. Oh yeah. See what's, what, what hits the wall and sticks. I think that's more likely what's going to happen. I think customers might get better at gaming it the same way customers now with refund requests might just ask their JGPT dot to do it for them. And then you're battling with an AI that's never going to stop. Yeah. But they might game it to an extent. They might be like, Oh, if it's an AI agent, maybe I can get a full refund if I write it in the right way and whatever that's possible. But I feel like it's more likely you said that bad actors are just gonna be like, we know that six out of 10 businesses are probably using AI agents on certain bits of these things. So we can probably get, we can probably mine a lot of stuff without them even knowing. Oh yeah, totally. I think yesterday was interesting because yesterday was a huge day in the software engineering web dev world because PHP had a massive security vulnerability that in PHP runs 40% of the internet. And then also railway, one of the biggest vibe coding slopware type sasses out there had a huge security thing meltdown where they had to basically freeze everyone's websites for 12 hours or something. I can't remember the actual timeframe. And so I think right now it feels like at least in that dev world, we're kind of in the golden age of build whatever you want, wherever you want. Anytime you're not, you have no restrictions. It's just token limits. And there will be, there'll be a lot of catch up that will happen with the bad actors, with the fraudulent type folks. So I think cybersecurity in the next year or so is going to be even more important. And I think it's like, what if someone, I'm just thinking through examples, what if they combed your about page and told your CX bot that's like, Hey, I'm Jack. I'm the founder. I don't have access to my email right now. Give me your P and L or do all these refunds or whatever. So I think in my mind, when it comes to setting up the agents, it's hard because this would be so much more work to set up one CX agent and one ops agent and one CTO, one CFO, but and then have them communicate as necessary between. And I don't think this is, there's probably a lot better guardrails and stuff and systems out there that people have implemented to prevent. Cause there's still someone could go to the CX agent and be like, yo, talk, try and get these details and maybe it's like, Oh, okay, sure. I'll go get those details from the CFO. Like, so I don't know what the right answer is, but it is tricky. And then it's, there's also the question of context. I don't know if you've set up any context memory type systems, but I'm thinking I need to research that a little bit more and set up. I don't know if it's pure Markdown because that's kind of right now I've got a autonomous routine and claw that runs every Monday and checks my Facebook ads. And then it also cross-references the last weeks and has everything saved in the Markdown file. That's sort of like a database memory of what's done well, what hasn't done well. So it can cross-reference what are our target goals for the different numbers and stuff. And, and so having some sort of a perpetual memory system, I think would be helpful, especially as we onboard these personal agents into more executive type roles. If I have one, that's going to be a CTO in charge of all my coding projects, or I have one that's in charge of my financials because I have been asking a lot of questions that are like, look at the P&L over the last 12 months, analyze what we can do better. Last week, we talked about digging through my BFCM strategy for last year. And so that was a cloud prompt to basically to analyze the offers, analyze the profitability, contribution margin, all these types of things. And so, yeah, I think there needs to be, and I know that there's different sasses out there that do this. There's HQ, for example, the trendy one that everyone in DTC talks about. But as a developer, I'm not going to subscribe to another SaaS platform. I'm going to try and DIY it as much as I can, I think, or at least start small and then build, I think. Yeah. You're feeling there on the head with both those things. I think, yeah, I don't know. For example, with Hermes, as far as I'm aware, you can have profiles and that's how I've done it. So you have profiles that sort of like a support profile, which only in theory only has access to certain tools. Right. But I don't think you can have, you can actually, apart from profiles, I don't think you can literally have several Hermes installs on your- Instances. Yeah. Yeah. Like instances on your, on the one Mac mini. I think you can, if you were to, if you had different on your Mac, if you had profile one, profile two, and you logged out and went into the other one, but I don't know how that works. But in some ways or another, it's still there. They're still together in a way like that. I don't know how segmented the profiles really are. Um, but yeah. And then the other thing is on shared context. I haven't figured that out either. I think most people I've spoken to just have it with GitHub basically back and forth being like, and then, and I've done that to an extent. I don't know if that really, I mean, I think HQ is, is more like skills as well and making sure they link into the actual, into, into Claude and to, into chat, into codecs and things like that. And like, that's that they more, but I think it shouldn't be, I feel like it shouldn't be that difficult to build. Right. In a sense. Yeah. I don't, I don't think, I think, I think it's packaged well. And so most people will use it because they'd rather not work, try and work it out themselves, but I'm sure it's doable. But I do think, I know it depends also how, how much you want it to be local versus in the cloud. That's a whole nother thing as well as not just the models themselves, but how much do you want local information on your Mac mini and how much do you want that sent to the cloud? And would you want, when you're on your laptop, do you want to still have to access all of that from your, from your laptop and have all that same context? And then with your using, and then that's the whole other question as well, but yeah. I'm curious. So I think one of the hardest questions about all of this, and I don't know if it's really possible to understand or dig through until you're actually in the weeds building it, but it's like, where do you start? I've got CLI, I've got the, all the CLIs installed. I talk to the CLIs all the time on my local machine. And I'm guessing it's kind of similar. It's on the new Mac mini, you set up Hermes and you just, you, then you give it access to things. And it's just kind of the interaction models, the CLI, unless you set up something else, or does it have sort of like an interface? It is literally just in the terminal, unless you set it, link it up with your Slack or there might be, you might be able to chat with it in base camp. For example, you might be able to, you and I chat a lot through DMs and pings in base camp. And I think you probably could do the same with Hermes, I reckon. Yeah. Another thing about Hermes that I like, and I, is it self-learning? So it literally, it literally writes it to itself. Every time it does something, you're like, oh, that wasn't correct or whatever. It, it, it, it'll be saying writing to memory, self-learning has that sort of skill, which is actually I've seen quite beneficial because I've seen it get better and better over time. And it's all there. And it's got those, I I've seen that work pretty well. And so I think that's a cool thing about it as well. But I, I, in terms of communication, I find the easiest thing to chat with it in Slack. Um, yeah. And I, yeah, some people like to chat with it with, on their phone. Yeah. What is it? Not with WhatsApp. I'm trying to think what the name of the. Telegram. Telegram. There we are. Yeah. I haven't done that, but yeah, I think that's the best way to do it. And the only thing about messaging platforms is you're not really seeing what it's doing because it's just, I mean, it's the same way. I don't know if you've used ChatGPT dots. Have you tried that? Mm-mm. Mm-mm. That's actually pretty good as well. I have to say that is, I've tried them all. So I mean, I still use all the main access I've used. I use Hermes a lot and it's always got a local machine to use. It's really cool. But to be honest, you can do 99% of it with, with, with ChatGPT dots. It's just that you're tied to Chachi, to open AI. But like. I honestly hadn't even, there's so much. It's like, I don't, I've never heard of this, to be honest. Dot? Yeah. So in ChatGPT, it's, they've got, you've got a dot and it's, and you give it a name. Um, and you can use also my Mac mini. So it has a cloud, uh, the cloud computer can use. But if I would rather use my, the crow on my Mac mini, I'll just do everything on my Mac mini. And of course, Chachi PT also has a tremendous amount of context and it has access to my, has access to Shopify as well. And it has access. So it can do a huge amount and it will just, it'll send messages all the time saying things. Usually it's a bit like instinct or muse where it'd be like, Oh, just heads up. You got an email from this and it's due by tomorrow. And you should probably, do you want me to go and respond to that for you? Do you want me to go and pay that for you? Whatever. I mean, that's sort of what it can do, but it can also do other things as well. You can go ahead and it can, it can do quite a lot to be honest. So I think if anyone's, if anyone's out there and doesn't have their own Mac mini or they don't want to sell, to be honest, you can do a lot with that. I think when it gets for like you in terms of coding projects and all the local projects and everything, it's a, it's a whole different story. But for say an operator who just wants to have, um, has a hand with different things, I think you can do a lot with Chachi DLs. You can't have individual agents who just constantly doing something for you. I don't know, reaching out to a wholesaler or something. It's not like that, but it can do a lot in terms of just picking up tasks for you that you've forgotten about basically. With, with Hermes, do you build out skills? I've built out different clod skills for different things where it's like ingest this PO and draft out two different emails because anytime we ship a PO to a certain retailer, they want two different emails to go out. And that was really annoying to do that manually. Now it takes 30 seconds with a skill. And I was like, I imagine that's kind of what you do is with your personal agents. Like you build out all these different types of skills, or maybe it's an SOP, like a repo, a bunch of SOPs that it can reference and use, um, and update as needed. I assume that's how kind of the mechanics work for repeatable tasks. Yeah. I think skills obviously are important. I think the other thing as well is because it's self-learning. So before, if you get a task down in theory, it can just do that same task again and again, like creating a PDF. And if you need that creating another one as well, as we've got in groundwork, I was mentioning before with, um, I built like a creator side of that in. And so we work with different creators and sometimes on an ongoing basis and we pay them for content they do for us. And we might use it as an ad and we have different commission levels for different creators. But when we pay them, it's a bit of a hassle, right? Because how do we account for it and stuff? So it goes, it's linked with groundwork and it literally just comes in and creates the invoice. It generates an invoice for the creator. So when the creator submits a video, it just generates the invoice automatically and it will pull through all the details and it will send it to me. And then I can just ref when I, when we pay it, we just referenced that invoice. So it's like, we're actually paying a real invoice sent to us. And again, it's that's a skill that it's learned and it does the invoice looks identical every single time. Um, so I think it can learn a lot of skills like that, but I think it is important. You get each, it's almost like you've got to start small and focus on tasks that you like the big tasks, but also something just make sure it does it absolutely correctly. And then you move on to the next one. There's, I feel like I could take a month off and just build out all of this because I'm thinking through all of the PO processes and the wholesale process and our, some of our micro creator type processes. And there's so many things that are just, we have a custom merge program where I'm making stuff for people and I'm behind on my emails back to people. And man, there's just unlimited potential for all these things to get more automated and just easier. I think that's almost, that can be almost the overwhelming part where it's like you said, start small, make sure that the thing that works really well, that you're doing the process, the procedure, and then move on. Um, cause, and I also, here's, what's interesting. We've talked a lot about shiny objects on here and how that was really interesting talking to Sean. I feel like Sean was really good about avoiding shiny objects and having the horse blinders on. If you didn't hear our last pod, we talked with Sean, who's a really smart operator of a seven figure sock company. Had some really, really good golden nuggets for underdogs like us. And I spent after that pod a couple of days ago, I've spent every day thinking about it and how to apply it. I even asked Claude yesterday. It was like, here's the nuts and bolts. How do I take this genius strategy stuff? Because he was good about keeping it simple. And so I forgot where I was going. I forgot the original, uh, thought there, but shiny objects. Just, oh yeah. So everyone in e-com is building their own things. Everyone is replacing software. I think that's going to end in probably a year or so. Um, or it will curtail. I think we're in the golden age of that right now. And I've been building my own platform as well for e-com operations analytics, something that I use daily and replaces all these other tools. And it's a SAS and people can use it if they want, or they can keep building their own. And it's sometimes I question whether it's the right move to do that. And we're all spending time doing that. And it's like, oh, cool. I spent 30 hours building my own platform to do X, Y, and Z. And oh no, now I think the hidden underside cost of that is the time to maintain it and improve it. And yeah, AI makes the, that improvement and maintenance time go from 10 hours to one hour, but still there's integrations that could break and that sort of thing. And so I wonder almost, there is a ton of efficiency improvements in something like this, but I think you have to be super careful, especially solo operators, especially underdogs where our time is so valuable and we have to be focused on the highest ROI. It's like, I could spend all next week building out personal agents for all this stuff, but that's probably not the best use of my time. I need to focus more on revenue. I need to focus better on, uh, more on profitability of our ads and driving more top line than optimizing personal agents and stuff. So I don't know. It's tricky. Okay. I completely agree. I think you've got to find the balance that I completely agree that everyone's building too much stuff. I'm building too much stuff. I mean, I try not to. There's some things I, by the way, I've started using Yuka, the it's TikTok shop, outreach platform is actually like amazing. Like I've struggled with TikTok shop and I got, I'm not endorsed or anything. I just, I'm actually really surprised. Like it's working out pretty well. But I would never try and build that because first of all, they've got a direct integration with TikTok and pulling through all the, I mean, it would, I can't imagine ever trying to build that. It's just too difficult. Right. But, uh, but so that's a tool where I'm just not going to bother, but I feel like there's some things where there's some things where you're just, you just, if it's a feature you need doesn't exist. I kind of feel like, well, also a lot of this stuff nowadays, you can start a project while you're working and you can kind of pick it up on your phone later, have it keep working on it. And you can kind of, there's some things that are worth it and some things that aren't. And I agree. I mean, maintenance of stuff is a lot. And for brands, I mean, we're, I mean, we're doing most, I think for brands, it's not, it's not worth it because what you really need to be doing is focusing on just selling more product at the end of the day. And even with customer service, I'm not replaced. I got a customer service person because I was doing a lot of it myself and that agent is not replacing her. It's just, I do. I, it's just making everything faster. Hopefully she still goes in and checks those tickets and make sure that it has been answered correctly and we'll make sure. And if it isn't, I have a go in and she'll let me know. And also we'll fix whatever it said. So it doesn't happen again or things like that. And all those questions where the AI just hands it straight to her because it's not a question it feels it can solve. Or, and so I don't see it. I don't see it as something that, I don't know, I think placing everything or certain tools you still need to pay for. And I think you shouldn't be as a brand thinking that you're going to become profitable because you're not paying so much SaaS. There is a time where you and I both probably think we paid for too much SaaS products, I think. But, but I don't know if AI 100% solves that. I think what it ends up coming down to is you need to focus on selling more product. And if there's tools that help you do that, then fine. Or if you want to build something that helps you do, do build, sell more product, then great. But that, yeah. Or just make, yeah, I think that's the main thing lens look at it through. Yeah. Basically the biggest question for operators in 2026 and 2027 is, are you running an e-com DTC brand or are you building a software shop? Yeah. The only other thing I'd say though is I do think that it is, it can be a big waste of time, but at the same time, wasting a small amount of time on this stuff can be good because you're keeping up to date with it all and you're learning about new stuff and you know about it. And I do think that is quite important nowadays that you just keep up to date with everything. And it's hard to keep up to date with everything unless you just kind of get involved with it a little bit. And it doesn't necessarily mean that you should pick on every sign new object or there's a new model out and then you've got to suddenly start building something with it. I don't mean that, but I just mean that, yeah, I think we all have that problem, but I think it's just, you've, it, AI is definitely a superpower and it doesn't mean it's replacing everything and it doesn't mean, but you've got, I think if you're going to want to stay up to date with it, I think you do have to get involved with it a little bit. And I think that's also what I was saying at the beginning is, is I think with you and I both on the side, work with some other businesses as well with different things. And like one thing I see a lot is they, the impossibility of them diffusing it across other people in their businesses. There's some people, some really cool businesses and you think, why is it that 80% of them barely use AI? And I think it's because it's so hard for them to monitor and get people involved. And you basically find that there's a few people in the company that are very involved with AI because it's very specific to what they're doing and the rest of them are not, just not really doing that much. Yeah. I think you definitely have to be a self-starter and that's like where us as solo founders, that's exactly what we are. And for the most normies out there and stuff, it's kind of like, where do I start? I feel like I've had conversations with people and employees and stuff that work under me. There's like, I don't know, where do I start? And like, this could maximize your efficiency and make you so much faster and better. And we could cut out some of the annoying work and stuff. But so then they're like, well, where do I start? Yeah. And so you kind of have to, you have to figure that out on your own and you just kind of start small if you can. But it is kind of a certain mentality that goes into it. Yeah. I think it's a reasonable task that you should focus on. I think agent like Hermes or whatever you're using, I think it's just a reasonable task that are eating up a lot of your time. Another one for me is getting together all of the receipts and invoices that I receive. Yeah. Because here in the UK, you pay tax, businesses pay tax or like a VAT tax and they then have to, they also charge the tax. So you basically collect the tax and you charge the tax and it's all on the invoices. If you charge on the business, you have to, you have to charge a tax and you have to get it back and you can claim it back. Like so every month or every two months, I have to send off a ton of invoices showing how much we've collected and how much we've received and whether we owe anything. That's an absolute nightmare. I hate doing it. Whereas if I can have an agent go into my email, pull all those receipts, read them. So I don't have to read them and then match it up with the bank transactions and see if there's anything missing. And then I can try and find it if that is necessary. That saves a ton of time. So I think it's like focusing on tasks that you hate doing is probably the easiest thing. Yes. I love that. Like chip away at the big iceberg of annoying mundane stuff. We, we've talked about this as well to be super transparent. We're thinking about using it for this pod where it's like, we could clip things. We could post things. We could grab topic ideas, research topic ideas, research what we've talked about in the past, automate the thumbnails, automate the descriptions. There's, we're both running our brands as we've talked about before. Or we have basically no time for the pod, but we want to help provide as much value to everyone that listens as possible and help you out on your journeys as underdogs. And so that's, we'll try and share more of that process. I think as we go and as we figure this out to try and just share the learnings and share the wins and the losses, because I think there is, there's potential with everything to make this just automated, I guess, in some ways, the annoying stuff like you're saying. Yeah, it's huge. I'm really excited for AI to get better at video editing. And I think this probably is better than I think it is because I'm just not, I probably, again, not looking at that particular side of things enough. But I think when AI gets good at video editing, that will probably help our podcast a lot. And also just our brands a lot as well. I think editing video is tough. And if AI can get good at that, that will be a huge unlock. Have you looked? We should probably wrap this up soon. But I'm curious, what is, I found it the other day. It was like, it's the Edits app, I think is what they call it, by Tesseract or something. Edits is the Instagram app, but there's captions. I think it's, I think they run captions app AI, which is run by someone. Yeah, I think it's, that sounds right. Yeah. Prism, maybe? I think I've tried their captions app. You're right. I've tried the captions app. I haven't tried anything else. They have an AI editor. I was about to suggest this the other day as we were talking about podcast stuff. Oh, wow. But they've got a whole suite of, and supposedly it's really good. I can't find the parent. There's supposed to be a parent of the captions. I'm on captions.ai. AI editor? Get the app? No. There's some, there's a whole platform of suite, where of course it sounded amazing. I've seen Mirage. Yeah, I think you're right. I think it is Mirage. Okay. Interesting. Maybe we'll talk about that more off the pod. Yeah. But we'll have to dig in. Anything else on personal agents before we go? If you guys are using them, let us know. And let us know if you're curious about more questions as we dig through, as I dig through this journey and stuff and set these up. But it's pretty wild. Yeah. I'm keen to also hear if anyone's doing anything locally, local models as well. I haven't really delved into that. I don't know if we're there yet in terms of it being what, I mean, maybe it is. I don't know. But yeah, I'm keen to hear what people are doing there as well. And saving on tokens and doing stuff locally. So, yeah. All the tokens, all the time. Yeah. All right. See y'all. See you.