How I Trained an AI Agent to Do My Cold Outreach

Sean Weisbrot: Cold outreach is broken, open rates are dropping, reps are burning out. But what if your sales team never slept and never missed a lead? I interview Jim Weldon, the CEO of Prospect Desk. When we talk about the rise of smart agents in sales prospecting, how they're automating 80% of the grind, enriching your data, sinking with your CM CRMs and pre-qualifying leads before you even lift a finger. But this isn't just another AI hype fest. We'll get into the real challenges. From personalization and tech complexity to why humans still matter in a world of agents. If you've been wondering how to scale your outreach without burning your team or your brand, this is the episode for you. Why should people be thinking about using agents to do their prospecting?

Jim Weldon: You know, I think there's, there's three things that, that come to mind on that as we look at it internally. Um, number one, the amount of repetitive tasks. Sales teams and prospecting, marketing teams, SDRs are required to do that. Don't involve actually talking to a customer. I think the last study was 80% of 'em. Um, of the tasks they do are something that could be automated. So I think that's the, the first reason you, you wanna look at agents or smart agents. Um, the second thing is the learning. It can see things in the data. You can't, uh, it can help you identify ideal customer profiles more quickly. The channels they come in from, uh, we have confirmation bias as humans. So it's like, oh yeah, it's coming in from search, it's coming in from paid campaigns. And then when you really look at the data, you find out that's not really the case. The, we like some of the brands coming in, but they're not part of our ICP and I. I'd say it's a little more ruthless in a good way on how to analyze this data. Um, and the third thing is, is. It's like anything you, you have to figure out how to execute and implement and deal with the change. So instead of getting rid of the SDRs, what we're doing is, is we're creating a digital twin org chart, which is how do we use these agents and name them. So if there's somebody's, uh, we're doing in-market audiences that. Process is called Im a, so Im a works with Vinny, who's a real person in SDR and he has five teammates that are not human, that do discreet tasks from four to 15 things in the background that he normally would do. And he gets to kick off. So he actually runs a team of agents underneath him. So I think we're in that period of, uh, exploration, which is how do we bring them together, um, which is. What's the human's responsibility? What's the agent's responsibility? And I would to wrap that up, I would say there's, you know, the people that aren't testing, exploring this are gonna miss out on a huge opportunity for leverage. That's why I would implement a agentic bottom line. Leverage.

Sean Weisbrot: Do these agents also have the capacity to have conversations with the human that they're working with? Kind of like to build a relationship and rapport with them? Or are they literally like, uh, chat bt that's just sitting there waiting to have you tell it to do something or it is just constantly doing the things that it's programmed to do and, and you just go, Hey, by the way, also do this thing too. Like how does the actual communication work? Yeah, that's how it started.

Jim Weldon: It was, uh, passive and waiting is where we started. Then we started looking at, Hey, can I initiate a process or conversation via Slack? So that becomes our interface so it can slack the actual. Digital teammates to get them to do something, which is pretty cool. But no, we haven't activated the voice side of it, which is where it's going. So eventually you'll be able to, instead of saying, Hey Siri, you'll be able to say, Hey, Pam, 'cause Pam is our master controller, the MCP of all of our agents. Then we know if Pam, what are all the agents active that are with this SDR? You're actually gonna be, voice is gonna be the last interface. That's gonna be the ultimate interface.

Sean Weisbrot: But I mean. Like in, in the current state, even if you can have, you can send it messages through Slack, for example. Will you be getting messages like, Hey, did you have lunch? Hey, are you know, are you done for the day? Can I talk to you about something like, are they acting like humans and, and trying to build an emotional connection with the human counterparts, or is that just not necessary?

Jim Weldon: Not yet. I think that's where it's going. Um, I think the. I, I'll call it the humanizing and intimacy component is definitely on that path, but it starts with rudimentary things, which is, you know, request response. It starts there and we're still fundamentally figuring that out. Then there's gonna be anticipation and prediction of recommendation. Hey, I think I should talk to Jim because these tasks aren't done, or Is there anything he's working on outside the system I haven't seen that I can help with? Those aren't here yet, but we're still in the request response cycles on what's going on with AI and smart agents.

Sean Weisbrot: Do you think the humans enjoy working more with the agents? Because they don't talk back. They don't have feelings. They don't have times that they clock out. They can do things whenever you need them to.

Jim Weldon: Some of the people we work with, yes, because they're introverted and they'd like to minimize the interaction. I would say, um, we've formalized our amount of interactions on the team so we don't get people completely disconnected from other humans. So I would say there's a fine line to watch. You don't wanna do that. Totally. 'cause then you'll have these people in their own little isolated worlds. I don't think that's healthy. So

Sean Weisbrot: let's say for example, you were talking about Vinny and Vinny's got five agencies working with Yep. Let's assume he's managing them. Who is responsible for recognizing the need for X number of agents and then building that team and then deploying that team for this human to then be a

Jim Weldon: part of. So it's our, um, we have a digital execution manager that sits over Pam, which is actually an MCP controller. It's a fancy way of saying it's the orchestration layer for the agents. So there's a actual person that looks at the discrete role. Vinny's, SDR role, sales development role helps 'em define all the processes, then goes into our, our pool of employees, digital smart agents, and says, Hey Vinny, these are the four you wanna work with. So there's a, a manager that sits at top that really helps design the playbook, and that's the way this would work instead of it being, um, like a battle card or a tear sales sheet. Instead of it being, you know, they're having to decide themselves as an SDR, it's more of a, Hey, what do we wanna get done? What's the playbook? And then Vinny coordinates with his manager. The manager coordinates with Pam, who's digital, Pam coordinates. That's our prospect activation machine. That's why it's called Pam, that MCP controller. Fancy way of saying orchestration layer. And then we build a playbook of five agents for Vinny, and then it goes and works with Vinny. What we're working on now is a dashboard that looks at the velocity and throughput. What was Vinnie's baseline of activity and interactions and success compared to with the agent team? So do we see a two or three or four X leverage in throughput and success? So that's who decides it's a coordination between his manager and himself.

Sean Weisbrot: So let's say someone like myself has a business that's just starting out. I've figured out my offer, I've figured out who my ICP is. And I want to have access to more potential leads, but there's so much time and energy that has to go into prospecting for myself because I'm trying to do a million other things. I don't have a a Vinny, I don't have a manager. I don't have any of those things. So how can I do this for myself? How can I build my own

Jim Weldon: team of agents like this? You, you wouldn't want it. You'd wanna partner with a company like VMG we work with, they offload. Those activities because as a solopreneur or one or two person firm, super hard to implement on your own. You don't wanna do it yourself. You wanna leverage a service from a third party, we power those third party partners. 'cause to bring all this stuff together, it's, it's daunting because you have to understand what a demand side platform is. If you wanna re run retargeting, you have to understand how to put the meta tag in a single impression ad. To activate a, a social audience, you have to understand. How to use press Master to get content out? You have to understand how to tie your TAG infrastructure into, um, pat, right, which is our prospecting activation tag. Pat's responsible for all enrichment and de anonymization. That's just no way you do it As an entrepreneur, you would come in and say, Hey, I wanna pay 500 bucks a month for an SDR agent. Here's my CRM, here's my email system, here's my tag fire for my website. Call it a day. Two logins, a tag on the website, you're done. That's 15 minutes of setup. That's where it's gonna go for the small business. The problem is, do you have enough volume of, uh, visitors to make it worthwhile or is there enough search volume for, IM a in market audiences to go and market to them. My experience with small business, they're gonna get left behind in a lot of this, the, the SE and UP are really gonna take advantage

Sean Weisbrot: of this. You were mentioning. Search volume and all of that. So it's not like, uh, they're looking on LinkedIn and trying to find out who the ICP is from, from the ICP information that you provide. They're not like trying to go on LinkedIn and find potential clients for you and then reaching out to them and doing cold dms and or trying to get them to book a call and then the agent takes the call and then gets them to pay and then you just provide this like, sounds like there's an email component there as well. What component

Jim Weldon: called email or something. Um, so that was just one activation channel. Everything you just described is Lynn, our LinkedIn, uh, navigator. That's Lynn, everything you described on LinkedIn, it can do so it can go out and profile, uh, the company you're part of. It can determine if you're enter ICP. It can determine if you have the right title. It can then do a direct message out to you. It can then send you a link that says, let's set up a meeting. It can answer a voice call from you and say, Hey, Jim's not available. I'd love to set up a, a meeting for you. That's our Alex. Alex is our voice person that handles our inbound phone calls. Uh, and it's pretty fricking good too. It's crazy. Um, so yeah, it's all the activation channels are available through the smart agents. There isn't one we can't do. It can slack with you, it can do a chat bot with you, it can email with you. We just have learned in business to business or in e-commerce, especially those two. Email is still the killer communication channel. Full stop. I'll still read an email in business if it's crafted properly. And when you abandon cart or you go to a website, you're trying to buy something, you leave. If I can capture and identify your email, that's just still a fantastic way to drive business. So if you look at how big e-commerce is and you look how big B2B sales are doing one off B2C, it's, you know, okay. It's hard 'cause people are so finicky, but when you're in an e-commerce cart, you are way down the path. Something happened to, to drive that abandoned cart. So you're, you were right at the precipice of a sale. So,

Sean Weisbrot: yeah, I've thought recently I'm, I'm quite new to, uh, Ana, you know, Google analytics and tagging and all of this stuff, and pixels, and I'm used to other people doing those things, but I've been trying to learn for myself. And it, it is very complicated. As you were mentioning before, trying to figure all these things out and trying to add these little pieces in and then understanding all the details, it, it's very difficult. I have come across a lot of people that have YouTube channels and whatnot, and they say, oh, I can create an agent team for you and 20 minutes using N eight N. What's the difference between what those guys are doing and what you're doing?

Jim Weldon: Uh, we use n eight N uh, NAN is just a, a low code. It means you can drag an object and link it to something else to create a process flow. There's like five or six of those platforms like that that are. Really went well, funded and work well. The issue is, once again, it takes a lot of cycles to figure this out, to get something rudimentary that demonstrates a, a quick, rapid prototype, super simple to make it work functionally. Yeah, good luck. Uh, if you haven't done it, it, it takes, I don't wanna say weeks, but it takes us four to five days from idea to finished Alpha takes another week or two to get to beta, which is something that we can use in the wild. Alpha we use internally for ourselves and we use, every single agent starts with us. Then if we really like its approach, we take it to, to a partner. Then if the partner goes, this is cool, and we give it to them to use, then we say, let's go find five customers. That's when it goes to the wild. 'cause the partners don't count. They're closely aligned with us. So if it fumbles, you're okay. But to get it, so it's customer facing and a beta, it's, it takes some discipline and, and, and some learning cycles. It probably took us. Two solid months to really get our hands around how to get it to a proper beta. Just the overall architecture of our system. And by the way, you have N eight N, you don't have the data we do. So now you activated the system that's only relying upon a form fill or you know, grabbing your general, you know, Google Analytics data. You don't have any real data to drive ICP scoring, lookalike audiences, activating retargeting. Sure. Outside of that, you could do it yourself. Um, people are well under gunned. They don't, they don't know what they're, they don't know what they don't know at the end of the day, and that's why it's like anything else, you know? Buyer beware here. These YouTuber, or mostly Instagram. Actually Instagram is amazing for these things. They're just promoting their stuff to sell their program, to get you, to make you feel like it's simple to do. To build an alpha, super simple, make it work in your business more complicated.

Sean Weisbrot: I'm gonna go back to the LinkedIn and uh, to Lynn and Alex, so. I know that I can talk with Chacha Bt about different things that build up over time with memory. So it understands, you know, what my service is, what is the pricing structure, what are the tiers, you know, where people are coming from to find me. Like I can, I can build up this knowledge with it. It can't go and do any of those things on its own. I can't find people for me and I can't sell them. How do you get these bots, the these, these agents to understand what the DM strategy should be? If you don't know what it is, or maybe the AI goes, okay, I know what you're selling. I think this should be the strategy. Is it a, you tell it everything and then it just applies it? Or does it say, Hey, I noticed that if I try this 5% people convert more than when you told me what to do. Is it figuring out how to do these things better than you can? Like how does all of that work?

Jim Weldon: I would say for the most part, for most people, no. I would say the, the prompting and the strategy piece is probably the weakest thing people drive. I think a couple things. One is understanding how to find who it's working with in the market, whether if it's a LinkedIn strategy, there's lots of people who have been successful that have shared their learnings. So first thing I would do is I would have to do a competitive analysis. And the proven programs that have worked in approaches, how does that compare to mine and what do you think and recommend? It can tactically recommend some good things, but strategically it's, it can't decide very often. Here's the customers I want, and what it can do is look at and say, Hey, here's the customers that have closed that you have. Here's more like it, but it can't blue ocean and say, Hey, go make a left here and go take your sailboat that direction. Not as good on that yet. It's getting there, but not as good. So I'd say strategy is still something that as the leaders of the company, you wanna set. And then I always compare it to what's going on in the marketplace. So what's Oglevy saying? Right, because Ogilvy has an amazing amount of data. Presentations a corpus of knowledge in the market. What is IAB saying? Uh, what's the digital Marketing Association saying? What are their best practices and approaches? Compare what we're doing to them and, and tell me, give me a gap analysis. That stuff, oh yeah, that's real. That can get you to a very good tactical extra, uh, execution program to test then you just a b it Like anything else, it doesn't matter what everybody thinks. It matters what converts. So the way to do that a

Sean Weisbrot: b test, it. So same thing with like Alex, you would say, Hey, this is my sales script. That's, that closes people. I want you to, to just use this and it'll,

Jim Weldon: we don't do out outbound on a, a Alex is for inbound. So how do I give that, um, human feel of an operator, we'll call it more so than an SDR rep. Um, we have found that people don't enjoy those unless, unless they're super well trained. It's really hard to train those right now. So that they can answer the, so here's one, ask the same question two or three times on any of those agents, and it starts to sound like a robot. 'cause it gives you the same exact cadence, tone and response. So it's got some weaknesses in, in the, uh, um, if you're doing it for outbound, uh, inbound, it's fantastic. It, it gets you four outta five conversations are handled by Alex. Then there's that odd one where it's like, oh, they asked a question that he couldn't handle, and uh, then you go back in and listen to it and you're like, oh, I understand why he didn't So. I would say the auto dial, our robots are tough.

Sean Weisbrot: Well, I wasn't thinking of it in terms of that. I was thinking it in terms of if someone, let's say my strategy is LinkedIn called dms, and let's say Lynn is doing all of his prospecting for me, and then someone says, Hey, I wanna have a call. And then Alex ends up taking the call. Alex can take the call and

Jim Weldon: set up a meeting for you. And could we, could we have it answer some questions? Sure. But. The potential questions are hundreds, not thousands, and we don't have the, as a smaller company, you know, it's somebody like IBM would have a fantastic time with that because their internal memory and knowledge base and corpus of data that they could feed into it would be incredible.

Sean Weisbrot: Hmm.

Jim Weldon: But as a smaller company, we have a few hundred artifacts, documents, tear sheets, white papers. That's actually not a lot of data to drive. Um, some of these response and, you know, question response models.

Sean Weisbrot: So Alex can just pre-qualify people who would be prospects for me. He can't actually close a deal and get them to pay.

Jim Weldon: No, I think you said it perfectly there. I think it does a good job of that screening and it's like, Hey, can I ask you a couple of quick questions? And then it'll say, you'll say, sure. Um, I just wanna make sure I've got this right. So when I talk to Jim, I can get the information over to him. It could ask a, B, C. We don't do that with him yet. That's where we're going with ours. So you get to two or three questions. Um, you know, if you're under 10 employees, loser. For us, if you're over 10 employees, it's somebody we wanna start. You have somebody on your team responsible for performance marketing or lead gen. If they don't, not gonna work for us. If they do, we want to talk to them. You have an agency you work with today? Yes, we do. Fantastic. We should probably invite them to the call then, because they'll be the first ones to bury us that they don't wanna work with us. It's our number one, uh, knee capper is an agency. Feels like we're gonna take from them instead of supplementing. Then when they hear what we, they we do, they're like, oh shoot, this is cool. We get to test it for free and if it works, we get credit and we can do more with, um, you know, more juice from the squeeze. Yes. More top of funnel prospects to work with. So that's where three simple questions we know very quickly if it's in our ICP, we don't care about revenue 'cause we can infer it from employee account. I, I don't care about industry because, you know, that just gets sussed out when we do the data enrichment. Um, th those are things we do in the background. I don't need to ask those upfront. And we capture email. Hey, I can Jim email you, can I capture email? Then it repeats it back to you, so we got it right And we always say business email. We parse the business email and then we can find the URL. We, we scrape the website on the URL. Then we can go and look at the data. Then we figure out very quickly if it's in our ICP

Sean Weisbrot: and you have a, an agent that's specifically doing that enrichment. I forgot the name of it, sorry.

Jim Weldon: Yeah, it,

Sean Weisbrot: uh, almost

Jim Weldon: all of 'em do. Pat is the one that I'll, I'll do some of the stuff with Lynn can do it. So with Lynn, it'll parse it. It'll go back and look at your profile, and it'll write a custom message with chat, GPT, turn around and do the dm. The thing we ran into is it tries to write too much and we hate those.

Sean Weisbrot: Yeah. I, I specifically tell people, do not use a chat gt created email template. Like, just don't. I'll ignore you. So

Jim Weldon: obvious from a mile away. So what I did is I sent it about a hundred of my founder, what I call founder prescribed emails. And they're, they're completely different. They're bottom line, they're a sentence or two. It asks the question, it gets to the point, and um, it's usually something a little provocative. Hey, listen, we're trying to partner with folks in your space. We're also looking at acquisitions. I don't know what your appetite is. Love to have a 15 minute call, Sean, and see if you have some time. Yeah. Then it might be a ps Hey, I see you went to Georgetown. You know, uh, so did I. Something maybe just a little personal. Um, not very often though, 'cause we usually don't have that much in common. I hate those ones where they over familiarize themselves, like they're my buddy. They annoy the heck outta me. They're not my buddy. So I, I hate, I hate the buddy email approach.

Sean Weisbrot: Fair enough. So I, I've spoken with a guy, I, I interviewed a guy recently whose business is very heavy on LinkedIn outreach and he said that what he found was when you do a voice message as the first thing, that people are much more likely to respond because they are not used to people sending voice messages on LinkedIn. So they're like, oh, what is this? And they want to hear it. And he said that that gets him seven times more responses. Oof.

Jim Weldon: Did not know that. I will literally be slacking Vinny on my team to find out a prerecorded message on that one. That's genius if it works that way.

Sean Weisbrot: So he actually, he, he decided to turn this into a product and he's getting ready to launch it. But yeah, I, I find it very interesting. And I was talking to Chad, GPT, not, no, I was talking to Gemini 2.5 Pro yesterday. I was like, Hey, what should my cold outreach strategy be? 'cause I, I am changing my business model. I have changed my business model. I'm in the process of putting out all the pieces together to make it work. And one of the things was, Hey, what should my LinkedIn strategy be for dm? And it was like, well, your first message could be a voice message. Your second message could be a text message that has a link to like an article like your, your foundational LinkedIn post that talks about the problem. And then the third thing could be a video. And the video is where you're inviting them to a call. And I'm like. If an AI agent could do all that for me, then that would be excellent if it per, if it personalizes their name. 'cause I've seen some software products that you say the foundational message once, so like a video or a voice and it'll then store that and then when it has a prospect come through, it'll make it with that person's name in it as if you did it for them right then and there. And so I think if you can make that automated with an agent, I think that would be killer.

Jim Weldon: Yeah, that, that doesn't seem very hard.

Sean Weisbrot: I don't know how

Jim Weldon: to do it. Oh, we, I mean, I'm just, we've built 13 of these agents. It doesn't seem very hard. The issue is how you scale it from one to many, and that's what we have found is the challenge one-to-one NA then, and those tools work fantastic. One to many. We're still working through how do you do your one to many? That's, that's why it.

Sean Weisbrot: One to many, meaning a single instance of an, of an agent supporting multiple clients through multiple needs,

Jim Weldon: through multiple customizations, multiple logins, multiple delivery of data to multiple systems. So imagine you're an agency is the model we think of, and you have 10 clients the way those systems are set up. It's NAN is one configuration into the agency and then it can't handle the 10 custom deliveries after or the 10 customizations of the message.

Sean Weisbrot: And doing 1 8, 8 n for each of the clients is not scalable.

Jim Weldon: It's, it's messy. It can be done if you wanna do a dozen, but if you know I have 25,000 in clients through my dozen partnerships, that would be messy. So instead, what we decided to do was we can do so many cool customizations and get them the copy and content with the agency and the client's. Id deliver it to our partner who then parses it in their database and triggers it from their systems. So they use their email system. They're a CRM to help their clients generate leads, and when it responds, it puts it into the client's end system. We figured out a workaround to be able to deal with one to many, which is we make the agency do the push and the sequencing of the outreach, and then the responses can go to forms. And, uh, things that are the client side. So the, the, the agency has its own CRM and its own marketing automation that kicks it off. It gets a blind copy of everything, so it has a record, but it also, then the response pushes everything to the client's CRM, which is really, you know, that's what you're trying to do at the end of the day is. Fill the top of their funnel unless it's a managed service. Then if it's a managed service, it comes back into the agency. They run it through and use somebody like Master Inbox or something like that where you can handle 50 different emails for that client. You can handle the inboxes with one inbox, and then you're mastering the, the top of funnel to middle of funnel. That's the, this is, it's, it's doable. It's just not elegant yet.

Sean Weisbrot: So what I'm hearing is you don't have your own CRM. You'd prefer clients to work with the CRMs that they would normally use and to, to enrich data and all of that.

Jim Weldon: You, you have to be able to work with 20 different CRMs. You can't push anything on. A client, they're not gonna change. So that's, it's, the agency has one and their, their clients have 20 different ones. We're used to dealing in those environments. The issue was how do you do N eight n and agentic flows within their, it was a little stumble fumble for a while.

Sean Weisbrot: Random question as well. Something that, uh, I'm just curious about because I hate Slack. You said that you can, you, you can have conversations with your agents in Slack. Does Slack then count them as seats and force you to pay $5 a month, whatever, to not lose your history with the agents? Like how does that work?

Jim Weldon: Yeah, that's fine with us.

Sean Weisbrot: Okay.

Jim Weldon: Because what we do is we put it as at Ima and that, that's like a, an action call inside of Slack for us. 'cause you know, you put at, you know Sean and it comes up with your name. So there's only one unique instance of that name in the system. It's a three letter acronym, so there's no. A LX is in our company. There's no imas, there's no, you know, we don't have Enemy Macs Mac, that's our, you know, multi activation channel for our demand side platform for retargeting. Um, it, it's just, it's like a person comes up and the system and it knows to communicate with them. What it's really doing is it's sending a response to another channel and it, we're polling that, watching for that, um, at sign in the name, and then when we see at Mac, it's like, oh, what does it want us to do? Okay, here's the instructions and it's smart enough to know and kick off something for us. So it becomes the, just like the interaction window, if you're, you know, on chat GPT, and if you put a prompt in, it's just another way to prompt without leaving the system.

Sean Weisbrot: Where do you think the next five years of this is going?

Jim Weldon: God, I mean, I'm listening to everybody from, you know, Jensen and Nvidia passing 4 trillion in market cap yesterday to Satya saying, you know, cursor and software is going to eat SaaS applications to look at what Oracle. Oracle's been a quiet sleeper in this whole space. Um, where do I think it's gonna go? I think two things. One is nobody knows. I think there, there will be more process And agent automation is one generic theme, but how does that manifest? No idea. I think the, the chip wars and the model wars have just started. I mean, if you go look at the, the amount of large language models that are accessible, it's hundreds now. We only know of the top seven or eight because they're so big. But people are gonna take their corpus of data and put it up in a private marketplace. And I'll give you one example. Let's say, um, Watson, just yucks, IBM decides it's gonna put its entire Watson approach to how it does. AI and learning and puts it in the general domain, well, sorry, puts it in a private, um, marketplace and allows you to access all of their research data in a large language model. Holy crap. That would be incredible for deep research. Like I think that's some natural places. So large language models, I think large action models, which is where gents going. Um, I think we're gonna stumble and fumble with poor returns on investment for the next year or two until the use cases get really well defined. And then there'll be a great marrying between the human and the agentic flows for leverage. And if we sell it as leverage to that teammate, not BS or or AI washing it. But if we really do a good job of. And this is what we're doing with Vinny Vinny's our, our case study internally. He didn't know anything about AG Agenty, so we put him on a two week learning course, which was designed by chat GPT. Um, you know, and then it was like, if you were gonna start from a beginner and this was your job and you wanna get to this proficiency level, what would you need to learn? What videos would you need to watch? How would you recommend he tests his knowledge? It was pretty cool. Um, I think. We're in the stumble fumble, phage phase in a good way because we're in full hype cycle right now. It's gonna solve everything for everyone. Bullshit. But there's some smart people out there like Benioff and Allison and Satya at Microsoft, and you know, Sanjay at Google, they are putting Sam Altman at, you know, open ai. They put half a trillion dollars the venture and those seven big seven companies did in the last 24 months. What do I think is gonna happen? I think it's gonna be the greatest technological change ever. 'cause we don't have a physical limitation. The industrial evolution was about physical transformation, so, you know, horse to car, right? You know, before that it was steam powered engine to, you know, uh, uh, electricity. Now imagine what's gonna happen when you can move an electron. That's all we're moving. We're moving electrons at the speed of light. And guess what? I think change is gonna start, it's gonna be four or five orders of magnitude faster than anything we've seen. Moore's law is 18 months. What's the law now? They're releasing a new LLM model every two days. Some new L LM model somewhere every two days is being released. That's insane. Not versions of the big seven, just new LLM models that are of material value. What the heck's gonna happen. So anyways, long-winded answer to say, um, you wanna get in the middle of this if you're an entrepreneur or you wanna be relevant.