709: AI Agents: The Only Skill That Matters in 2026 (Masterclass)

0:00Hey founder fam, before we jump in, I want to take a quick moment to talk about our sponsor OmniSend, the email marketing and SMS platform built specifically for e-commerce founders. We've been recommending OmniSend to founder students and members of our platform for a while now because it just works. So whether you're launching your first store or you're scaling it to seven figures, it really helps you automate your marketing and get results. And did you know, on average, OmniSend customers make$79 for every$1 they spend, which is an insanely good return on investment. And if you're already on another platform, here's the thing.

0:35In just five days, you could actually be paying 35 % less without doing an absolute thing. OmniSend will move every flow across, list, and template. You just show up when it's done. And finally, because you're part of the founder community, you get 50 % off your first three months with the code FOUNDER50. just head to omnisend.com forward slash founder and that's founder without the e to get started all right now let's jump in the show

1:03i don't see myself as an ai expert i don't use it nowhere near as much as i should i always feel guilty i have two agents that have been running non-stop for over 160 days it could replace any human va that i might bring in so it really replaced all the jack of all trades very quickly we'd had several project management interviews lined up for people and just cancelled all of The next one was customer service, and I'd say the next one after that was legal, believe it or not. Isn't that scary to you, though? It's so exciting to me. Like, no, fear does not play into it. Ari Meisel spent 15 years teaching founders to do less.

1:36He's worked with over 300 assistants. Now 32 AI agents run his work across six organizations, and he barely touches them. Anytime you see, like, this is the 25-word prompt that's going to 10x your business, like, it is touch utter bullshit, and it's just social media fodder. I replaced all of their customer service with an AI agent that I built. We dropped Influx, which was 16 grand a month. I have this whole document on human writing guidelines and not sounding like AI. For the first time ever in the history of the company, people are responding with like five star reviews and how the customer service are great.

2:07Every dollar I'm putting in, I'm getting like 15 out of right now. So who did you stop hiring and what role did you end up having to bring a human back to perform? So...

2:20Hear the stories, learn the proven methods, and accelerate your growth and future through entrepreneurship. Welcome to the Founder Podcast with Nathan Chan.

2:34Hey, I have a quick ask. If you're loving these episodes, leaving a quick review is honestly the best way to support us. It helps the show grow and means we can keep bringing you the founder stories and insights you tune in for. Please share it with a friend. Thank you so much. Now let's get into it. Ari welcome back man long time hey man thank you it's great to see you and it is good to be back so you you got into like saving time because you had crones and you were able to go through a recovery process but you scaled your business to you know seven figures doing what an hour of work a day and then not even yeah not even there you go so so now ai is here where do we start dude Where do we start?

3:19It's so, you know, I've been doing the productivity stuff for probably about 15 years now. I think this is just the wildest time of I've never experienced anything like this. It is so exciting to be able to build the things that we can build now and say yes to the things that I can say yes to now that, you know, there's always there's so much that productivity people teach about, you know, being able to say no. It's so much more. it's true but like it's actually so much harder to say no to things now because you really can take on so much more with ai and yeah it's it's wild um i'm using uh ai agents i think i've got about 32 right now running up for me across six different organizations so it's amazing so where do people start if they want to get into ai agents like and what are your tools like where like talk where do you stop man yeah so this is a I think one of the most interesting things in terms of how you think about this stuff I'm going to get very detailed with you on technical aspects and point put people towards all the resource all the things they need I'm not nothing will be held back but I do need to take a minute to sort of explain a mindset element here that's so important so one of the big focuses of what I've been teaching founders and entrepreneurs for the last decade and a half is about delegation.

4:43And I want to talk about that for a second, because it's extremely relevant. The skill that people need to develop is not getting better at AI or certainly not coding or things like that. And it is absolutely, and I'm sorry if you guys have a course about this, but it is not about prompting. Anytime you see like, this is the 25 word prompt that's going to 10 extra business, like it is touch utter bullshit. And it's just social media fodder because if you don't have a solid structure or foundation in place none of that stuff matters and if you do have a solid structure in place prompting isn't prompting it's just delegation so the skill that everybody needs to get really good at is delegating because if you set it up right that's ultimately what you're doing with the ai agents is you are delegating more effectively so the best place to start where a lot of people start with delegation is with a VA, basically, right?

5:38So I have worked personally with over 300 virtual assistants over my career, and I owned a virtual assistant company for a while. So it seemed to make sense to me that the first kind of agent I was going to build was going to be an AI chief of staff, basically. And so I took everything that I've ever learned and all the protocols and all the methodology and the frameworks and everything, And I put that into a architecture for what I think is like the ultimate AI chief of staff. And it's, it's like, I think it's a 17 page PDF. It's freely available on my website. People can just grab that and either read it or load it into the Asian platform of their choice.

6:19And so for me, that platform is tasklet.ai. So T-A-S-K-L-E-T.ai. There's a bunch, right? There's perplexity. There's Hermes. There's Claudebot. there's cloud code all those things i use all of the different tools but for agents specifically taskfoot is my favorite platform and part of the reason that it is such is because it's just so easy to connect to your systems which is one of the biggest things so you can do you can take cloud tro and you can say like connect to my gmail or connect to my troll and you can do that but there's always a little bit of sort of tinkering around there and then the other thing is that it'll do what you tell it to do but where it starts to get tripped up is continuing to do things and figuring things out on its own overcoming sort of obstacles so i have two agents on tasklet that have been running non-stop for over 160 days doing stuff on their like they're both working on very very complex tasks which i can get into but i've been working non-stop and I have developed 18 foundational rules for every agent that I use and those are in that the whole chief of staff document in terms of things that keep it from sort of going off the rails or hallucinating and being able to do things in a really practical way and all of it is basically the same instructions that I would give to a person that I was training to do some of this stuff and then once you have that one then you can use that to spin off more specialized agents such as a cfo agent a marketing agent customer service uh legal like you name it and they all kind of spin off without that base one but you start with that and again like my framework you can put on any platform you want but tasklet is just it's amazing and i don't have any stake and Tesla.

8:14So when it comes to, I guess, you know, an e-com founder, you work with some DTC brands, you're saying offline, and you're doing some crazy stuff, what kind of things and tasks are you delegating or roles? And talk us through like that level of autonomy, right? Because that's, that's what's really interesting to me. Like I'm playing around with Hermes, but like, that i haven't got them running autonomously yet because that's scary to me well and so and it's funny because i would consider myself fairly advanced but i have not found success with her with uh hermes i think it's probably hermes i think it's more the god of war than the the handbag brand and yeah there's a lot of times where it's like can you do this thing and it works on it for a long time and it's like yes and it's like but what happened oh let me check on that Yeah, yeah, yeah, yeah, yeah, yeah.

9:10So TASLIT doesn't seem to do that. So TASLIT seems to actually be able to see some stuff through. So for e-coms specifically, yeah, so I'm working, one of the companies I'm working with is a DDC consumer package company in the wellness space to sell supplements and things. And there's a number of things that I've done with it, but one of the biggest ones recently is I replaced all of their customer service with an AI agent that I built. And so I'll tell you about that journey because it's been fascinating. So when I joined the company, well, so I've been coaching for all this time. And for some reason, in the last six months, several of my long-term coaching clients have asked me to come into the business as an employee.

9:54I don't know what that changed, but it's been an interesting experience. And so when I joined this company, one of the first things that the founder said to me, he's like, we've never had 24-7 customer support, and I want to have 24-7 customer support. Didn't have anywhere close to it. So the first thing I did was outsource it to a company called Influx, which has been around forever. Yeah, they're a long time, man. Yeah. Long time. Yeah. And I've used some other companies. They're great, very reliable. And we're using Intercom as our ticketing system, basically, right? So it's taking in emails and Facebook Messenger and live chat and text and all those things into one place.

10:35So we bring in Influx. They gave us, I think, eight people in four time zones. We have 24-7 coverage. Great. So my goals were I wanted CSAT scores that were over four. I wanted first response time under 10 minutes or 15 minutes. And I wanted our time to resolution to be under 180 minutes. and we were way, I mean, we were like way, way, way off of that when we started. So I bring them in and we're doing much better, but we're not hitting those numbers. It was like, I think our first response time was averaging like 25 minutes and our time to resolution could be up to two days. And the human providers were out, were escalating a lot of the, the, like almost 50 % of the test tickets were getting escalated to the one supervisor that we had.

11:25And a lot of it was things that just shouldn't be escalated. And so then it's like, okay, well, let's build that into the knowledge base. Let's do this or all these things we're trying to do. And I had an AI agent that was reporting on all this for months, but not really doing much. It was just saying like, oh, this is where we're failing. This is where this issue is coming out. So I got fed up one day or like two months ago. And I said to the AI agent, I was like, can you, can you actually deal with these escalations? And it's like, well, let me check. So it was a really interesting switch. And again, this is not like an AI thing per se, but so it looked at 175 escalations and it said, I can clear out 171 of them now.

12:06And it did it. And the reason was it has access to WooCommerce. It has access to the Shopify plugin. It has access to ShipHero. It has access to, you know, WordPress in the backend. So whereas the human front end person was like, oh, this person has a broken coupon code. I don't know what to do. my ai can go in and say oh yeah well they're not a first-time customer and this is a first-time customer coupon but they've been with us for a while so let's issue them a new coupon for 10 give that and do it so no matter how nice your customer service is at some point people just want resolution i don't care how nice you are so and this was so cool is so i'm like all right so you so fix all these escalations okay great so then i said to it i was like could you handle the front end too and talk to people in intercom and it's like sure i can so it starts doing that and i have this whole document which is also in the ai chief of staff framework on human writing guidelines and not sounding like ai and so it starts writing these messages that are so human and so great and for the first time ever in the history of the company people are responding with like five-star reviews and how the customers are so great and so for a day i'm like i'm like It was like I was on cocaine, like watching this.

13:20It was like, oh my God, this is so cool. And so then I'm like, all right, let's build in this process and let's do that. And then all of a sudden the response time started to go down. So I'm like, what's going on? So I ask it, what's happening? And this is so great because again, like a human, it said to me, well, you're having me build all these really interesting reporting things while at the same time I'm trying to deal with the front end stuff and I'm getting distracted. So I was like, totally fair. I was like, so why don't we split you up so you can be the supervisor agent and let's spin off a CSR agent that just handles front end stuff.

13:58So we do that. It builds it out. And that CSR is called Caesar now. And Caesar just handles the front end stuff, doesn't deal with any of the other back end, the coupons, all that stuff. And it can escalate things to the supervisor. And so within three days, I've got first response time averaging 44 seconds and time to resolution averaging two and a half minutes. I mean, it's battery customer service and I've studied almost any company I've ever seen like Interactive with ever. And then the best part is we dropped influx, which was 16 grand a month. and then we don't need 14 seats in intercom anymore.

14:42We need two basically. So intercom went from 4 ,000 a month to 300 and I'm spending about$3 ,600 a month in tokens to do this. Crazy. Okay. So you said you have 30, how many agents you got running right now? 32. 32, all through Tasklet. one agent is the equivalent of a role effectively right right so you've got that's how I shoot it yeah yeah okay so what are all these other agents doing and it's all through tasklet yeah so I have a like I have um six tasklet accounts so I have them from one for each company basically that's working with yep so for the cpg company i've got uh one two three four five six seven eight eight agents running currently right now so uh the i have a chief of staff all them every one of the accounts has a chief of staff agent that's sort of the jack of all trades that one can communicate with the other agents if it needs to but for the most part that one's dealing with scheduling and one of my favorite sort of general AI hacks is do you use Granola at all?

16:00No, I've seen it though. Okay, so I think Granola is the best it's not a call recorder it's a call note taker and it's interesting because in America there's a lot of laws around what you can record and not when you're on calls but it's not a call recorder, it doesn't record the audio it just transcribes and then it does AI notes from it so you and it doesn't show up on the screen or anything so you can use it in almost all sorts of scenarios so i use it all the time i'm using it right now and every time i use it the chief of staff looks at the call notes and it makes trello cards for me of things that i need to take care of or it actually just starts working on things that it knows it can work on so uh it's Anyway, so I have a chief of staff.

16:49Then I've got this customer service supervisor. Then I've got the customer service ramp. I have a legal agent. So this particular company deals in a product that is regulated in some states. And so it's constantly looking at legislation that's upcoming, that's changing, and where we can ship to. So if a state bans this product, the legal agent goes into WooCommerce and blocks that zip code and all those areas automatically, immediately, so then we can't ship to it. I've got a marketing agent, so it's looking at social media, sorry, doing social media, newsletter, all of our e-commerce, like DTC emails.

17:37It's doing that, generating the coupon codes, like keeping the cadence, tracking that stuff. It self-optimizes. So basically, it is set to every single week. It looks at the traffic. It looks at the click-throughs. It looks at what happened in the newsletters. And it suggests, hey, this week we should try these four things. Let's try this headline because of this. Let's move the sign-up button halfway down the page. It's stuff that I don't know this stuff. and it really works um on my own personal not personal on my my lessdoing.com web page i have it do that as well and in two weeks it went from an average viewing time of a minute and a half to seven and a half minutes so that so almost all the agents are designed to self-optimize um additionally there is a uh uh where is a sales agent so we have a sales team of six people and so it's doing a couple things one of which is it monitors every sales call that's made so there's every call they do they use air call and it's recorded the agent looks at the transcript identifies coaching opportunities so it's like gong or chorus you know which has been around forever.

18:54So it's identifying sales improvement, which it posts into a sales channel in Slack for everybody to see. But then each week it analyzes across the sales team. So it'll say like, hey, this competitor was mentioned to three salespeople this week. Like we should look into this. And it starts to look for those patterns. It also has a voice AI agent that I built on retail, retail.ai, sorry retailai.com so it can actually do like appointment setting and stuff with that and again every every ai call that it makes it's then looking at the transcript and being like where are the knowledge gaps what's something that the person asked that we didn't have an answer to let's fix and like so every single call gets better that's crazy all right so so how do we how do you build this like i reckon for anyone listening probably the most interesting especially for founders is the chief of staff one but then also the marketing one right like the marketing one that's self-learning so you use task let talk me through the stack talk me through how you get it to work autonomously what level of approvals do we have like talk me through that how do you actually do that um again there's all these foundational rules i mean i can go through some of them but i mean so because in terms of like trust and authorization all that kind of stuff so one of the rules that i have in place for all these agents is and this some of these are going to sound very basic but again people don't a lot of founders and people don't do this when they're dealing with people so they certainly don't think to do it with with ai and you really need to so like one is i use trello i've used trello for years as you know from my courses i love trello i've used all sorts of calls but i always i always come back to trello every agent has a Trello card and there's one board I have with this and it's the nightly knowledge card.

20:46So every night at 11 o 'clock, every single agent updates that card with a comment. And it's like, what did it accomplish today? What automations ran? What blockers did it hit? What's the plan for tomorrow? Um, and, uh, in and update that card. And so every day it is then supposed to refer to that card to see where it's at. So it basically, it makes it very hard for it to sort of like hallucinate or kind of go off the rails because it always has context. And it's deep context, right? Because like the agent I've had running for 160 days, there's 160 comments. So if I decide tomorrow, I don't want to use Tassel anymore.

21:29I'm going to use Hermes. I can say, look at this card. This is you now. so that's one thing um the other one super basic sounding confirm that the thing you said you did is something that actually happened so a lot of people hermes is a great example of this do something for me hermes like boom boom yeah i did it where is it i don't you know what do you mean so every every action with my agents if you if you say that you sent a slack message next step is read the slack channel and see if the message matches as the one that you sent that you say you sent you sent an email look in the sent inbox sandbox read the email does it match the one that you were supposed to send uh you know make a trello card do this it and it and of course that uses a few more credits so it's not token inefficient.

22:25But it, again, it's like, it makes it really hard to mess up that way. So there's all sorts of ways you can do that. Now, the other one is that almost every agent I create starts in what's called shadow mode. So for the first two weeks, it doesn't do anything externally without approval. Meaning if part of its job is to send emails for those first two weeks, it's drafting them and showing them to me first. Now, sometimes it's like two days and it's like, wow, you've got it. You can get out of shadow mode, go for it. But you start that way. So it starts with that automatic URL that like nothing is going to go external facing until it's really been allowed to do that on mass.

23:08So you put all those things in place basically. And then again, there's a whole bunch of others, which I could list out, but it's easier if people just get the PDF. And then in terms of the marketing stuff, so I am not a marketing expert. I've been around a lot of it. I've been around some really, really great people. I've taken my founder courses, you know, but I'm not an expert. So it's, what I found really interesting about that is that there's so much that you can learn from the data that I'd never even realized before. So it's not even like, I need to know how to write the best copy ever.

23:46Sometimes it's literally like, here's two random tests this one did better and in the past they were like well maybe it did better because it was talking about this thing now the ai can have a pretty confident scoring of why one did better than the other and then sort of pull that threat and we can really rapidly test the other thing which has been so interesting too is in terms of the iterability of the ai so uh i built this voice ai agent for doing sales and to test it i sent it to the eight sales people i was like just pop your uh pop your phone number in the slack channel and it'll call you and after it calls you it's going to leave the transcript but also can you put feedback into slack so first person immediately is like call me it calls them and that guy gave feedback and then it immediately built that feedback into the next version.

24:44And so the next person that got a call was getting v2. So instead of asking six people, give me feedback on version one, one person gives feedback on version one, next person gets version two, next person gets version three. So you're at version seven in 10 minutes. Now, in terms of like the marketing stuff specifically, so there's a few different tools that have been really helpful. So one of them is called Zernio, Z-E-R-N-I-O. So Xernio is basically an API that lets you connect to Instagram, Facebook, Google ads, now Chatshubity ads. And it's not like a shadow one is going to get banned by any of those platforms.

25:23They've got actual like legit connections to all of them. So with one platform, my agent can build a social media post and then post it to all different platforms and then actually deal with the comments and the inbox and all of them. So he can do all that stuff now. and then very recently i started playing around with a uh it's not a it's not a repo it's like an api that connects to like 2600 marketing tools and it is called i'm blanking on what it was called i just started i think it's called trink i started using it today yeah so t-r-e-g so it lets you connect to like search console and uh google tag manager and again everything is sort of through an api and so you know you mentioned zapier before which i haven't used zapier in months now but what's so fun to think about for me is for years the litmus test for me of whether or not i would try a new tool or a company was do you connect to zapier because it's like i think it shows like the proper sort of open-mindedness about things but it's also like that's where I live, you know, and that's where I want to connect to.

26:34So if you don't connect to Zapier, I'm going to default to the one that does. Now the question is always, how good is the API?

26:45And, and to that point, one of the things that's also so great about this whole experience for me is that the agent becomes the interface. So like Notion is really an example. I hate Notion. I think the UI of Notion is just terrible. But I use Notion every single day because the agent, it's really great for LLM stuff. And it's really good for like an agent to read things and to know things. I just never touch it because I don't like it, but I use it all the time. So a lot of questions I'd like to unpack here. So tasklet, obviously we follow your rules and principles and we start to you know build that agent as a role in isolation but how do you keep an eye on what it's doing and know what it's doing it's a great question and how can you see it visually because there's this kind of black box for me anyways like you have like you said you have 32 agents how do you go and visually view them because in Claude Code or like you know I speak to mine through Slack, but yeah, it's a bit tricky to know what they're working on, what they're doing visually and seeing it all.

27:58Trello. So a lot of the agents are in Slack and I do have like a channel for each one. Well, I mean, unless there's a marketing channel, but I've got a couple of agents that are very specific. They're not like departmental. But so I basically have them post everything that they're doing in Slack. And I usually, I almost will never read those things, but they're there if I want to, if I want to check them, if I want to search them, they're there. But Trello, using Trello has been the real game changer. So I'd say it was about a month ago, actually, not even that long ago, I started having it manage Trello for me.

28:36And so anytime I give it a task, any of the agents, anytime I give them a task before they even start really working on it, as soon as they understand that it's a task, they will make a Trello card. in the to-do list with all the details. And I have a whole methodology around that, which you probably didn't expose to with the courses in terms of it's got to have a due date, it's got to have a label, it's got to have all these things. So it makes the card. And then if it's working on the moves that you're doing, and then there's a waiting on RE list, so it can move it there. And then there's a done list.

29:09Now I can go into Trello now, and I can say, oh, there's five cards waiting on me. So instead of me going to Slack or back to Tab Slit or anything, I just go into each card and I make a comment just like I'm talking to a team member. So it's incredibly easy for me to manage and keep organized with. And then it keeps that history. It's almost like a threaded reply basically. But at any given moment, I can jump into Trello and see. So instead of me going to task and be like, what are you working on right now? What are you waiting on me for? I just go into Trello and I can see in real time like, okay, it's got these three things going.

29:43Yeah, that's not important anymore. Let's forget that move that's done. and it's waiting on me for this document go ahead do that it's great so now it's like there is um the movie bruce almighty when he's like gets emails for all the prayers i don't know if you remember or if you've ever seen that but anyway it's like everything's just coming into the channel that i want now and i have to poop poop poop it's just i can push 10 buttons and make 10 things happen in like 10 seconds basically but it but it allows you to stay really really organized. And one of the things I've always liked about that is that Trello allows you to context switch in a way that is not overwhelming.

30:21All right. And it doesn't have to be Trello. You can do this in really any tool, but I just happen to think that the Kanban layout is really the best one. Yeah. Yeah. That's literally what I was about to say. As long as we do Kanban, visually, Asana, Monday, doesn't really matter. Notion, you can see it all. Yeah. Yeah. Yeah. Okay. I love that. That's really clever. So you're basically having your agents running it for a project management tool. And then how do you make sure that you don't smash your tokens that you're spending too much? So like for this company, this D2C company, customer service, marketing, all this kind of stuff, what are you spending on tokens?

31:03It's a really, it's a great question because I'll answer you. But before I do that, let me just turn that around you because you've been around, you know, social media and ad buying and stuff for such a long time, right? So if somebody was to say like, hey, every dollar I spend on ads, I get a dollar five back, right? How much should I spend? As much as you can possibly spend, right? At that point. Of course. Yeah. Depending on your gross, depending on your gross margin, but let's say you make a hundred percent margin and it's a digital product and I'm making 50 cents uh for every dollar I spent then yeah I would spend as much as physically possible exactly and that's how I feel about passlet and about AI agents if you're setting up like I truly believe that every dollar I'm putting in I'm getting like 15 out of right now so um I'm probably spending across the six accounts about seven or eight thousand dollars a month no that's not too bad there's a cost of like an employee yeah it's a call but i'm getting 32 for that first of all yeah and i mean with that one company alone saving twenty thousand dollars a month not having influx like it's it's a pretty easy sort of thing for me so i'm i think i'm i'm doing i guess it's three or four million tokens a month with tasla and they're like three hundredths of a cent for each token something like that but uh so there's a few things a few different guardrails so with tasla at least you can set daily limits for every single agent.

32:25So like my chief of staff agent always has the most token use. Well, no, actually it's not true. It doesn't have the highest. The customer service one has the highest limit right now because that's sort of just constantly going with stuff. We get about 200 tickets a day with that company. But you can set individual limits. So that's the first thing. The second thing is that again, with Tasselit, they're all kind of similar in this way, but Tasselik can use pretty much any of the Anthropic OpenAI or Gemini models. And also now we can use Kimi K3, which is a whole new ballpark. It's fantastic. And then Nemetron, I don't know if you've used Nemetron at all.