Kevin Williams: the idea that raw process and compute is going to overwhelm specificity almost every time. Hey Matt, how's it going? Matt: Good. Nice intro, man. Kevin Williams: Yeah. So the people who are listening can't see it, but Gabriel, our video editor and I have been working on a new intro. And that like hits Uncanny Valley for me. The last one, like I was like the silver fox. This one's pretty close to what I look like. But the wild thing is, I I actually took a bunch of photos from the the Liber Novo people with the chair. Thank you, sponsor. And of like my actual office. And if you look at the details of that video, it's completely wild. Like the sort of like sort of random tchotch keys and things like that that it's picking up on my desk. there's a map of the world. It even has like my monitor configuration that I I really use. So it's got like Slack on the left and it has it had Claude up on the right. I mean it it doesn't actually look anything like my real office because my real office is tiny. But just the fidelity and the detail, my gosh, we are so close to video being a solved problem, dare I say? Matt: Yeah, it's crazy. So my, I quick tangent taking pictures. We've talked about being able to take pictures of objects and 3D print them. and my friend had a was telling me he had an he was trying to fix something in his house and it was broken. So he just took a picture of it because he was gonna print a a replacement for it. It actually saw in the little tiny corner the ID number for the the object and it went and looked it up and found the precise dimensions. And he didn't even see the ID on the on the when he was taking a picture. He's like, I'll just take a picture. So not only is video getting good, 3D modeling's getting good, video gaming production is getting good. it's and then of course the ability just to note these little tiny pieces of information in any kind of picture or video and then use that to go look things up, like in a way that a human wasn't even thinking about. It's crazy. Kevin Williams: So, yeah, I'm I'm a 3D printing geek and I'm a member of a bunch of these forums. And it's been interesting to watch, even just over the last couple of weeks, this morphing from just abject horror. How dare you use AI to do this? To wow, I just solved this problem in 15 minutes and I didn't have to use CAD at all. And Matt: Yeah. Kevin Williams: you're seeing the tide turning a little bit, like the artisans are still losing their mind. Matt: Yep. Kevin Williams: why didn't she just go into Tinkercad and do that? Well, I didn't want to go into Tinkercad to do that. I just wanted to get to this to the solution as fast as I possibly can, because that's what's exciting to me. So this whole this is happening everywhere. And I think Matt: Yep. Yep. Kevin Williams: just you know, can look at a vertical like you know, 3D printing, CAD design, or whatever, and it's interesting to watch it just progress over time. But what are those other artisanal type activities? And, video production, video editing, you need to have taste. I'm not saying we're gonna get away from Matt: Mm-hmm. Kevin Williams: human taste, human input, whatever. But the editing itself is dang near a solved problem for like commercial activities. Film, you know, whatever. It's a it's it's a much bigger deal. But ask yourself audience, do you sit in this artisanal seat? Are you doing something Matt: Mm-hmm. Kevin Williams: that's this hard fought skill that is what differentiates you from the rest of the market? If so, you need to be thinking through how you can amplify that skill to be able to apply these these technologies. And if if that's not something you're keen for, okay. But is there the possibility of an artisanal label in whatever you're doing? artisanal Matt: Yeah. Kevin Williams: cad. Is that really a thing? I don't think so. Matt: Yeah, yeah. The the I like the way Sequoia splits this up. It had a great matrix. And then one dimension it's, knowledge information is on one side of the spectrum. And then judgment or experience or artistic ability is on the other side of the spectrum. And so something like CAD is like very scientific, very precise. Those industry any kind of job that has that kind of dimension is very in trouble. a job that like let's say recruiting. Takes a lot of experience and judgment to be able to judge judge someone quickly within a one-hour interview. that kind of thing is different from just pure information and knowledge. And so that will stick around. Of course, you'll get superpowers with AI to be able to do those types of jobs better and faster. But if you don't have a lot of judgment in your in your role or when you're thinking about other industries or roles, Those are the ones that are toast. And I've been watching these videos too, Kevin, of these th CAD. you know, the developers. We even I even had a tutor on Upwork for my son because I was like, my son loves 3D printing. Maybe he should learn to do some modeling. And now I'm like, Well, that was a giant waste of time. because, Kevin Williams: Ha ha ha. Matt: you know, I don't may remember last year he had his 3D printing business for the holidays, printing ornaments, and we're doing V two of his business this time of year. And now he's not even like going he's just letting the AI come up with its own models. He's like, I want some cool action figures. I want XYZ and it's just generating it all and printing it all. It's like magical. It's crazy. Kevin Williams: Wow, the future is the future is now. So Matt: Yeah. Kevin Williams: speaking of the future, you know, definitely not a politics show, but it's been a relatively big week in in politics as far as future change of control. and from our perspective, I'm curious on your take on what a change in legislative leadership is going to mean for AI. Like is this something that that's the Democrats are going to lean into and try and put the kibash on? Are we going to see slowing down in data centers? what's going to play out here? Matt: I I don't know. I'm I've heard a couple interesting political commentary. I think one thing is they're gonna use like AI is so politically charged emotionally for people. I think the data centers is just a co corollary to how like people feel about AI in general. And so like politicians always seize that. Like if it's something about identity or race or whatever, like politicians always seize any like negative feeling and use that to their advent advantage, either either party. So I think they, we're going to continue to s you and I are talking about all day. AI is getting better and better. It's displacing jobs. It's causing hardship for people in the labor market to some degree. And I think that will just continue to compound. And whatever party can make the most out of that fear or hate or disgust or all those negative feelings, they'll do so. So I think they'll continue to play with it. I don't think I don't know if slowing down will win. I think Trump also's like loves making deals, or supposedly loves making deals. So like I think he he says like we don't want to slow down, but who knows if the deal is good enough, maybe he'll like agree to some legislation. so I I think you're right. I think the Democratic Party could maybe not slow down, but maybe some sort of we'll see more legislation being proposed and possibly get passed in the realm of AI now, now that the both houses are controlled solidly. Kevin Williams: So I'm gonna take the positive view, which is we have needed some federal legislation around AI since all of this started. And Matt: Mm-hmm. Kevin Williams: individual states are doing some really interesting things around privacy, around child protection, et cetera, that really should be administered from a federal level. And those conversations are going to start happening, hopefully in a serious manner, as opposed to it being all focused around stopping AI, et cetera. Because I don't I don't think that's really an option as far as yes, we need more data centers. Yes, we need more power in order to power the curve. But there's still plenty of activity that's going on out there that it isn't necessarily going to slow things down, even if you stopped if you stopped data center production, what's going happen tomorrow? Data center projects are going to be springing up in the Middle East, in Latin America, in Southeast Asia, and all of these other places. Like bits and bytes don't generally respect borders. So I don't really expect anything to slow down. you know the the the I would say the flyer is that there something bad could happen. And we don't know what it is yet, but something you know, you saw the the Binance hack last week Matt: Yeah. Kevin Williams: of six point two billion dollars. Matt: biggest place in history. Kevin Williams: by the way, that was that was social engineering stuff that was AI augmented, but it wasn't like bots hacking into them. It was like good old fashioned, corporate espionage type stuff. Matt: Yeah, compromise a computer. Kevin Williams: Yeah, that yeah, they got they got like an engineer to like accidentally download a file, and that was that. But you know, something bad could happen and that could bring it to the national conversation, and then things might be able to change. But I'm not overly concerned about it. as far as stopping, but it is going to be interesting to see the types of conversations. AI is enormously unpopular in this country. So you know, the farther you are down on the development curve, the more excited you are about AI. Like 95% of Africans are absolutely delighted with AI. And I don't know what the numbers are in the US, but it's sub thirties, about maybe even lower than that. It might be in the teens as far as people who are excited about it. So I don't know, you're right. Politicians are gonna do the politician thing. I just think from a practical perspective, not much is gonna change. Matt: Yeah, I agree. I d I don't think we'll slow down. I think there'll still be states that happily have their data centers built. I think hopefully we get some some positive things, like you said, out of the the privacy and data controls. But yeah, well let's see what happens, Kevin. Yeah. Kevin Williams: So as far as state of the state, one of the things that you and I are noting is the the actual death of email, which we both sort of came up with independently. I think this has been rumored for a long time. Anyone listening who does email marketing knows that results have just fallen off a cliff in the last few years. a good campaign in 2021. Would have had something like a 3% response. that would be a pretty good campaign. Now it's like 0.3%. So the amount of AI slop that's filling all of our listeners' inboxes is just completely absurd. But what's happened now with the rise of more highly developed orchestration agents like Muse or Dot or Duet is email is basically Gone. It's like the first thing that people do is set up an email triage system such that they only have to pay attention to the things that they want to pay attention to. I'm a three or four hundred email a day guy, and I haven't been in my email client other than for like kind of special hunting around stuff in weeks. I don't even see the email. Matt: Yeah, I'm with you. So I'm I haven't been checking my email for a couple of years now. so I maybe I'm a little bit biased here, but I know I I was just at the Celtics game last night and it was like a kind of private equity bunch of investors kind of thing pairing with other founders and one of my friends was actually there and he runs an email, a very, very cool email business, and it's all outcome-based pricing. So like he gets paid when they when they succeed with their cold campaigns. And he's like, yeah, it's coming. He's he's a he's at that event because he's thinking about the future of this and what it means for his business. but I think the other interesting thing is you don't have to be Kevin and Matt. To see this happening. all the email clients are just gonna naturally start getting better at filtering out what emails are about and if you know the person and if this is relevant for you. So if you use Gmail, for example, you have your four labels or tabs at the top, like primary, news updates, social, and it the inboxes are just naturally gonna get better. at filtering and everything. So regardless of if you don't if you use Muse or not, or if you have agents in your personal life or your work life, just the the basic level level of layer of email clients are going to filter this stuff out for you. So I'm with you, Kevin. I think for a lot of people they're not even really I I even have a few friends. They literally don't even have an email address. Like they're I mean they they use it for like two factor and stuff, maybe, but that's about it. They're like, dude, never ever email. I will never see it. so Kevin Williams: Yeah. Matt: that that's like a younger generation. This guy's maybe like twenty-five. so I think you have the younger generation who doesn't understand why you would use email, and then you have the older generation or just regular generation and we're all gonna be moving away from it as well. So I'm I'm definitely willing to put the nail in the coffin here, stake and see if this video ages well. Kevin Williams: So, I think it's gonna age well. I think we're we're not going to escape email. Like there are too many like notifications Matt: Yeah, yeah. Kevin Williams: that that are that are necessary, but you you can escape the drudgery of sorting through it. Now some people actually enjoy that. I do find that there are all these, I agree. But I do find Matt: I know. Yeah, we talked about this with muse people love the experience of wait, I Kevin Williams: They are Matt: wanna complain about this or w I wanna do this other thing. Kevin Williams: But just think about that your me your a meeting finishes early and what do you do? You pivot over to email and you kind of graze on email for a bit. What I'm finding is it is a tangible amount of time that I'm freeing up. So you've been email free for a long time, so you've already realized this, but it has gotta be an hour a day that I am no longer ferreting around in email, and that's an hour a day that I can go do something else Matt: Mm-hmm. Kevin Williams: as far as launching other agents or whatever. So from a practical lens, anybody who's in G Suite, you can experiment with this today if your account is enabled. just go into Gmail and look for the AI inbox. And this still requires you to go to Gmail, but sort of the the crawl stage of this is just to turn on their beta feature and you're going to see like basically a summary of what's actually important in your inbox. And it's going to give you simple triggers for replying to reviewing, whatever. to be honest, this is what my email has looked like for two years using tools like Shortwave that have been doing this for me. The more advanced version is to just connect your inbox to your bot of choice, Grokbot, Instinct, Hermes, you know, Muse, Dot, Duet, whatever, you know, connect and just set up a routine such that It goes through and it vets your emails. And yeah, it does take a little tweaking as far as you don't really want to miss some things that come through. So calibrating on what's transactional that you don't need to see, what's personal, like you know, Matt: Mm-hmm. Kevin Williams: et cetera, takes a bit of time, but it's so cool. And then I what I like is the proactivity of of the bots, like so and so later is canceling the meeting. But it just pops up as, you know, Matt's canceled the meeting. He wants to reschedule for another time. Okay, cool. Let's schedule it for another time. And there's like Matt: So many of that. Yeah. Kevin Williams: no friction because I'm also just talking. So, you know, great, dot. Go on, reschedule with Matt. You talk to his bot, figure this stuff out. Maybe that's okay. That's next. I want to get excited about this, like this next evolution of the agents, which is. Basically collaboration agents. And I know that you have built a lot of this stuff internally, that, but it's important for people to understand that that's next that my bot, my bot is named Kermit. My Kermit bot can be connected to Matt. Matt, does your dot have a name? Do you do Matt: No, I have AI Matt and AI Graham, and they're different baby pictures of me. I don't know if we talked about that last time. Yeah. Yeah. Kevin Williams: Okay. But no, we didn't, but that's hilarious. So Kermit can talk to AI Graham and they can basically start collaborating. And the same is true internally. And that is kind of the next form factor that I think is going to start surfacing are like agent to age agent type communications. Matt: So yeah, this is I totally agree with you. And what this was something I've been excited about for a year, but we've had it's been challenging because when it comes to governance and security, right? So I have certain bots that no one else can command it except me. So it's in Slack, and so people are asking me questions, and then I have to s write a message asking my AI bot to go ahead and execute this or get the information for them that they need. and that's because that bot is just hooked up to so many systems. it, if anything went wrong, it could really grab something inappropriate and reveal some information that's private to the team or to the wrong department, you know, like maybe somebody's salary to the wrong person. so yes, there's waste and this is a big topic we've talked about, writing deterministic rules, you know, and having these governor agents to basically, you know, babysit the other agents and make sure they don't do anything that's against the rule set. But it's hard to program in everything. You know what I mean? It's it's hard to program in all these things that would be you and I know you're not supposed to share the salary of another employee or the hourly rate or how many hours they worked or other things like this. but maybe to an agent it's not that obvious. so as soon as you allow agents to interact with other agents, then you have this problem. And it's a you know, it depends on who in the org has it. how senior they are, what is the work they're doing, you know, it it the rules get pretty complicated, it's hard to control. Kevin Williams: Okay. So I I I but I this is this is important and this is a the the the one of the topics that I wasn't sure we were gonna get to, but I'm gonna use the opportunity to get to it. Matt: Yeah. Kevin Williams: which it's this idea of the bitter lesson. And listeners, if you if you don't know what the bitter lesson is in AI, you definitely need to do some research, but to to try and put it in a nutshell. It's the idea that raw process and compute is going to overwhelm specificity almost every time. So, Google, for instance, a couple of years ago put out these healthcare type specific models. And by the time that they were done, the frontier models with all of their additional neural connections and nodes and whatever simply outperformed them because more information is just better for the LLMs to do this. So when you roll back the clock, and I think it it's interesting because of the way that I know that your how your organization evolved over time. From going from no code into AI assisted development and then autonomous AI development, et cetera. you think back to like the early days of N8N, and longtime listeners of this show know that I, you know, I I had NADN flows going all over the place and like these deterministic workflows and like connecting this and connecting that. And then it was like a switch flipped in January of this year when Matt: Open claw. Kevin Williams: The when well open claw, yes, but but really clawed code becoming so much more useful. And all of those complicated things that we were doing, like basically connecting the dots and soldering things together and whatever, just got wiped away with the ability to create edge functions or cloud workers in in Cloudflare that were doing all of the same things. So process overwhelmed. The the former approach. And then you go forward to where we are now, and it's happened again with these orchestration agents where they're able to bring so much process to bear they can solve problems. Why is this important to what you were just saying? It's important because, yes, you you at this point in time need to have a stack of deterministic rules to make sure that you don't end up in trouble, but that is going to be just a point in time. And what you're going to see next are basically, I don't know, angel on the shoulder agents or something like that that have their prime directive of how they interpret the information that MatSpot is supposed to release that can act as Matt: Mm-hmm. Kevin Williams: a proxy on your behalf, as opposed to setting up all those reels, those rules. Because inherently they're going to be better at it. Like they're just going to be able to get that you're Matt, you're the founder, you're the CEO. These are the types of things that of course you're going to need to keep private. And they're going to start regulating the other agents. I think that's going to be fascinating, but it's going to be ragged to get there. Matt: I love everything you're saying, Kevin. I agree heavily that it's been really fascinating that we haven't seen more models come out that were really differentiated on a single category, like law or you know, health x-ray stuff, x-rays for cancer or something. So it's been interesting to see like all this stuff is really accelerating with just generalized intelligence. So that was really interesting. But the thing that I'm saying is. Yes, you're right. We can write deterministic rules. That's a moment in time. That will move on where they just understand Matt's the founder and they understand the rules of a CEO, what he can convey and what he can't convey. But Matt is a pretty unique guy into terms like, there might be a call recording that it might not think is appropriate to share, but Matt tends to just like share everything and not worry too much about it. just like, you know, I give my bots like a lot of credit cards and just, yeah, you know, I'm pretty sure it's gonna be fine. and I s just for the record, Kevin Williams: YOLO Matt: I still have not, yeah, YOLO, and I still have not had any issues. My bots over the last two months I've been hyped about this, guys. I've even been on this. If you're listening to this show, you know how excited I am. I still have not had a moment where I regretted giving my bots like full access to my entire computer, control everything on my desktop, and control my credit cards in my bank. Still not regretting anything yet. Knock on wood, okay? Don't jinx me. But My main point, Kevin, is I think some people are much more reluctant to share. They're a little more like holding their cards to the chest kind of thing. And I think it will be difficult for a bot to really understand somebody's personality of like, okay, Matt had a call and Matt generally doesn't care if the call gets shared with somebody else. but like s you know, I a lot of people that wouldn't like that type of sharing or that amount of sharing or transparency at their company. so I'm with you. There's like the average CEO, founder. But then, how do you deal with this huge widespread of use cases of do we share the call or do we not share the call? Was there something private said? Was there not something private said? there was something said. Is that bad or not bad? Kevin Williams: So I think it's interesting because I I actually think you sort of split generations. you know, you're you're a touch younger than me, but like the you're missing the tick of it is okay for you to need to interact with something else to clarify. This doesn't all need to be completely automated, completely away. That Matt: Gavin, come on man. Kevin Williams: but but they can be quick conversations. Like Matt: Yeah. Kevin Williams: a lot of my clients in in bigger companies in particular have. Awesome executive assistants, or they have chiefs of staff. Matt: Mm-hmm. Kevin Williams: And those conversations that they're having about, you know, this priority, that priority, this focus, what should be communicated, what shouldn't be going, those are happening all day long. They're just happening with a person. And now I'm seeing those EAs and chiefs of staff get amplified by using these tools and be able to Matt: Mm-hmm. Kevin Williams: just make all of the things happen. So a well AI amplified, EA can be like the voice of the CEO. But in a smaller organization, you are your own voice. So Matt: Mm-hmm. Kevin Williams: having your bot push back on certain things of like what should be redacted, what shouldn't, without getting granular about it. Like, hey Matt, this conversation looked like it included some private details about Kevin's life. Like, is that something you want to be included? no, just strip that part out. Okay, cool, done. Matt: Mm-hmm. Kevin Williams: And it it just moves on. You still have the obligation to kind of interact with it a little bit. yeah, I don't mean to call you like a a millennial or anything like that, but like, you know, it's Matt: Yeah. No. No. No, all good man. Kevin Williams: like you can't automate everything away as much as I would like to. Matt: That's what I'm going for though. Yeah, because I mean I when I look at like I think the interesting thing from using these over the last two months so extensively with this new upgraded version is you just see how much of your day is just answering things that a bot could have done if it knew that it had the permission to. You know, I've already seen so much gains over the last two months. My Slack is so much easier to navigate, my to do list is so much easier to navigate, but I can see. If I gave it the permission to actually just reply on my behalf for certain types of things, I would really make a big dent. I just don't trust the guardrails yet. and like you said, it's a moment in time. Let's see. You know, in one year, it could be drastically, drastically different. Kevin Williams: Yeah. Matt: but I agree. Like there for certain things, there's always gonna be some human in the loop for permissioning and com and approvals, definitely. I just think, especially in my org, my team is so Tech savvy, that they're really good at breaking the rules. and we all try to like, I I don't know if I ever if I sent this to you, but my teammate sent a screenshot of my bot conveying a bunch of private information to them on a DM. And they they and I was like, How'd you do this? Like, this is crazy. I thought my bot was like a hundred percent secure. There's no way you could have hacked it. He's like, dude, I'm a genius. I'm your best developer, man. And he actually just generated a screenshot pretending that it had that he had had. Kevin Williams: that's too funny. Matt: So like my heart was like, my gosh, how'd you do this? I thought I had it slow locked down. So, anyways, it's funny, but when you have a highly technical team too, you know, they're pretty good at like finding ways around things. Kevin Williams: Okay, so in interesting point. exploits are exploits, bad bad actors, internal, external, are gonna do bad actor things. the the you know, the Binance thing that we were talking about earlier was just social engineering. Matt: Mm-hmm. Kevin Williams: what does social engineering against a bot look like? So for a while it was super fun to jailbreak things, like people were were like breaking the Delta Airlines chatbot and you know, other things like code like prompt injection and Matt: Yeah. Kevin Williams: and whatnot. But it's going to turn into yet another angle on penetration testing. And you know, I know you have things going on in penetration testing. So I guess I'd be curious about thoughts that you have as far as how do you figure out how to harden your systems when they are bot driven? Like here we are telling the world how bot-driven we are. It's like, are we painting a target on ourselves that you Matt: Mm-hmm. Kevin Williams: should start shooting like incursion bots at us to try and convince our bots to give information and, you know, we know all know about Matt's YOLO with his credit cards, so Matt: Yeah, well actually I have a a t a task set up pretty regularly to to check like certain systems. So first like check all your your security systems on your workspace and on the major platforms you use like Slack. And then I I've also recently, I don't know if you've done this, Kevin, like when you start installing things on your machine, I have it go through the code. Like, you know, if it if it's installing a repository from GitHub, just like, hey, review this and go into, you know, like your higher end aster mode or fable mode and re review this code thoroughly before we install it. so just like really careful on what I'm putting on my machine and my device anymore. but yeah unfortunately security is just becoming a game of bots against bots with you know humans just thinking of clever trying to think of clever workarounds. But I don't know if you've been keeping if you've watched some of the other insane like Linux had a major, major hack the PA I think we talked about Black October last week, but it's very quickly leaving the realm of what humans are even able to process, of all these different dimensions and work well clever little hacks that the bots are able to find. Kevin Williams: So I I can't I I can't screen share it, but a friend of mine sent me a a text exchange that was like a classic like scam, like card scam. But it was so good that he actually went down to he was in London and he went down to like the corner market to go buy like scratch off gift cards and like photo like the the person on the other end was like, Okay, now just just just scratch off that number for me. type stuff. And it was, but it was so well tailored to his boss's voice that he was like, eventually he smelled a rat. But he Matt: Yeah. Kevin Williams: got as far, and this is like a smart guy. Like he got as far as being at the counter and he dropped a couple hundred pounds on on these gift cards. So this is my opportunity for the PSAs. The PSAs, your security checklist are Number one, you need to have two FA, two factor authentication. As annoying as it is, you need to have it on dang near everything. If you don't, that is a problem for you. Two is you need to have special purpose credit cards that I know you use these that have their own limits so that if something gets in trouble, like there's only five hundred dollars or whatever it is that's actually at stake. Three is you need to have similar limits on your APIs. So if you're using particularly LLM type APIs, you have capped spend on them. In most cases, it's better to have your systems go down than have some hacker like charge $500,000 on your Amex Platinum through your API. the fourth, I would say, is consider analog controls. So in terms of financial controls internally, reintroduce like code words and things like that. And this is pretty basic stuff, but people still aren't doing it. They aren't getting that this is going to be a real problem. And it's very easy to get entire bank accounts like wiped out. So four thoughts. Anything do you have any other additional easy security? Matt: Yep. I I'll add one more. I think real quick for those four are fantastic. And if you're not even sure how to check all that, just ask your Claud or your OpenAI to go ahead and do a sweep of your most important systems and make sure 2FA is enabled for all of them as an example, right? Or all your API keys consolidate all the ones that have been generated. You should be an admin to these systems. It can literally just take over your browser and find them for you. as the admin and figure out you you don't need to know what the key is, but you should know all the ones that are open right now and s check and see if limits. So again, if you're not technical and you're not in these systems, just ask your your codex or your your Claude code to go ahead, take over and find it and make a list and either fix it or you you can pass it off to your security manager. So that's one thing. The second thing I'll say Kevin is we also if you do payments in crypto, which I highly doubt many of our listeners do. Or if you do do large wires or just significantly big, big payments to people, as more of your team gets agents, I think that increases the security risk. So we just had a conversation internally, literally this week, like three days ago, about how we want to add some multiple layers of confirmation for payments over a certain size. So yes, like all payments are generally on a a virtual credit card that have a limit, like Kevin said. That's like great. But of course, you're going to be paying multiple hundreds of thousands out for payroll and for other types of systems. and especially if you use crypto at all, that's you have to have like multiple approval methods to make sure if my agent someone my agent goes rogue or someone hacks my computer and can take over my agent, they I'm not even allowed to send money out of the company by myself past a certain limit. Kevin Williams: So we just saved you, an hour a day on email and gave you fifteen minutes a day of additional security. So like I it is there are things that are worth it. Like Matt: We save them a bunch of time on email and we save them the risk of a hundred thousand or a million dollars leaving their bank account accidentally. So we're creating immense value here, Kevin. Kevin Williams: You're right. Our listeners should be our sponsors. All right. So moving on. I think the other the other topic that that I wanted to talk about was AI adoption. It's easy for us to be super enthusiastic and our clients to be super enthusiastic because you know, once you've experimented and you sort of have a taste of the the possibilities, it's just like hard not to think about all the ways you can execute. But I'm seeing the rubber hit the road in different organizations in different ways. And I'm finding it really interesting as far as the particularly younger workers and how they're perceiving AI versus their role in the organization. And frankly, I think it's quite dangerous for a lot of them. But I'm I'm curious about sort of things that you might be seeing out there. Matt: Well, I think I think we we try to tell people like I think there's a lot of fear. Anytime we're in an organization and we're talking about AI, there's a lot of fear. And so we actually even have like some steps to help your organization embrace it, employees embrace it and not be fearful. Like giving them more training, the type of language you use around it, helping them like with office hours, giving them like as many resources as possible to learn this stuff. And like as a leader, framing this all as in this is helping us do our job faster and better. So that's the main thing is I think certain organizations, people within them are very resistant and it comes from a place of fear, you know, because you know, they may just, you know, take one quick example and say this bot's stupid. It messed up. It didn't even do it right. So they don't even give it a chance. And they just want to kind of like sweep it under a rug or crush it. So I see it mostly with just people that are not tech as tech savvy. They're a little bit further along in their career. but I think you were saying, Kevin, you've seen it even with the younger generation too, which I was surprised to hear. Kevin Williams: Mm-hmm. Yeah, just the anecdote that keeps coming to mind is is a young marketing manager at at a pretty well known brand. And you know, she was Matt: And when you say young sub twenty five or sub thirty? Kevin Williams: I've never actually met her in person, so I'm making some assumptions, but I would guess around twenty-four to twenty-six, somewhere in there. you know, been a been around long enough to be a professional, been around long enough to have responsibilities, and she owns email marketing for this organization. And she's under a pressure from the organization to to enhance output, to get more emails out that are better, that sell more, etc. And she was so disinterested in leaning into the technology. And I don't necessarily think it was fear. I think to a degree it was just comfort with her current process. Like she was happy Matt: Mm. Kevin Williams: like dealing with Clavio and like, you know, all of the little knobs and dials. She enjoyed the artisanal aspect of it. Matt: Yeah. Kevin Williams: But for the leadership who was also on the call, like they're looking for the results. And the way that they intersected was interesting. sh basically ca casting shade on the whole process. but it was obvious it was it wasn't gonna play out particularly well. Like the organization needs more emails, and you know, either you're going to have to work a heck of a lot more hours in order to make that happen, or you're going to need in need to lean into the automations that that will will cut down the the task. And you know, I was trying to make the case to her that she was hired to be. a a highly creative, highly visual type person. And if she could spend her time focused on that creative output and tweaking the creative output like a creative director, and spend less time building templates, going back and forth with Adobe Illustrator, you know, whatever, that should be her happy place. but I I'm positive that it didn't really resonate. And, you know, I like that's gonna be a problem at some point. are they or what are their choices? the the the brand has to be hitting certain targets. either she needs to double the amount that she works, which isn't great. They're not going to hire another role to do it, or they're going to need to lean into the automation to make it happen. Matt: So do you think this is similar to what we talked about a few episodes ago when we had Muse and we talked a lot about the consumer aspect of AI and how there was a big backlash of no, I wanna complain about this, or no, I wanna click these buttons and take the time to do this random task. I don't want AI to do it for me. Do you think there was like a little bit of that, Kevin, in her reaction? and this is just sort of a segment different portion of personalities that just don't like this sort of thing. Kevin Williams: And she was happy. again, I'm not nothing against like against her. Super creative, very visual, et cetera. But she was happy to have the friction. The friction Matt: Yeah. Kevin Williams: was giving her purpose in a way that probably the organization didn't understand. That she feels good all day because she's doing the things. She's turning the dials, Matt: Yeah. Kevin Williams: she's pulling the levers. Matt: Yeah, I think this is this is I think we'll see this in the consumer realm too. People just doing the things, checking the things off the box manually themselves. I haven't encountered as much with young people and within organizations. Generally, I feel like most of the interactions I've had has more been about fear-based and and if we can convince them this is more convenient, they're like, Okay. But in general, it's a good point. Like a lot of we talk a lot on the show, like if you want to get the maximum benefit out of this stuff, you really just You don't automate what you're doing, you just reinvent what you're doing and do it a completely different way. And that's, I think, super uncomfortable for a lot of people, just the unknown and the needing to relearn a whole new process that may not seem intuitive because you've been doing it one way, the old way, for so long, Kevin Williams: On the flip side, I walked into one of my longtime clients on last Friday. And it it was wild as I was walking through sort of the cubes on the way to the CEO's office. Every single person had dot up. It was a GPT organization. And dot's really obvious because it's really colorful. So you'll see like green, mine's green because it's Kermit. like you might see purple, you might see orange, like based. So you can really see it at a glance. And In this organization, they had just introduced it like a week ago, and everybody was using it in a way Matt: Right. Kevin Williams: that I haven't seen adoption at this organization with just GPT. So in a fell swoop, dot allowed these people to go from that sidecar where you're like pivoting in your chair and you're using GPT to like answer questions to going to like mainstreaming it. So you know, you and I call that like a mono thread or whatever. But there's just Matt: Mm-hmm. Kevin Williams: something about that user interaction with these agents that is just easier for people to understand, or at least certain people to understand. And it's wild to see. And I I don't know, I truly don't know what that means as far as what does this organization look like in three or four months? Like now everybody's going to be, there's going to start to get it. And their dot is going to say, I'd love to do that for you, but I need a connection to this. So it's telling them what it needs, as opposed to them trying to have to like figure it out within a large bureaucratic Matt: yeah. Kevin Williams: organization. So now they can go to the AI team or the IT team and be like, hey, I have a use case for this, or have their dot like reach out to them to justify that it needs to they need a connection to the CRM or the ERP or like whatever Matt: Yeah. Kevin Williams: it is. And they're gonna start. Being able to solve their problems in a way that is going to be a departure from these deterministic workflows that we've talked about before. So behind the scenes, this organization is doing all kinds of really neat, very complicated automations, but they don't have the time or the space to like help individuals figure this stuff out. So now the Matt: Mm-hmm. Kevin Williams: individuals through orchestration agents and skills and connectors, those are kind of the three, I think, key parts here. are really going to be able to start solving their own problems if if they're told how. But this organization suggests they don't even need to be told how. They're just figuring it out. Matt: Yeah, I I think we just had a show and tell right before this podcast and I reminded people, because you do see I mean, you do see start to see, I don't know if classes is the right word, but huge discrepancies between people that can like really just understand this quick and adopt and adopt it quickly and the people that are laggers that you know, those discrepancies used to be very much smaller or hard to d hard to decipher or or see, but now with AI you can see massive, massive discrepancies. But my point was in the in the show and tell, I reminded people if you're stuck, you literally just ask your codex or your chat GPT to take over your computer and show you what to click. You know, like when I had to configure something in our Google workspace the other day, usually I'd ask one of my teammates to do it, but I wanted to get it done quickly. And it wasn't like that high risk. So I just asked my codex to literally just pull up the screen and show me the configured area. and you know, like those kind of things can get pretty complicated pretty quick. but you just got to remind people, guys, you don't even need a coach anymore. Like literally just have it take over your screen and tell you what to do, you know, or show you what to Kevin Williams: So Matt: do. Kevin Williams: I want to push on that. I think this is a really good takeaway for our audience as far as sort of the last thing we want to leave you with is these show and tells. So I I had this conversation too this morning with my partner. And I was like, we need to be doing more show and tells because we're, you know, we're a virtual organization. People are all over the place. And I know that Sonia over there is doing some interesting stuff that I haven't considered. And I know that there's interesting stuff that I do that other people aren't. necessarily doing. So my question for you is when you go into those, how do you structure a good show and tell such that it's productive and it's not just you pontificating at people? Because I'd be a worr I'd Matt: Yeah, yeah. Kevin Williams: be a little worried that that's what I would do. Ooh, look at all the cool stuff that Kevin's doing. Matt: Yeah, you gotta be careful of that. we keep it short because people don't want to commit to a long meeting. So it's like twenty minutes, one presenter a week, and we do it weekly. So it's meant to be like a quick, low, low burden meeting. Go and it's meant to be very practical. Like today, our finance I'm I'm super impressed with our finance team. Like they're doing the work of like a 20 person because we own six different companies. So there's actually a lot of bookkeeping that needs to get done. And this young woman is doing absolutely incredible with it. And she just showed some quick agents that she has. And something as simple as hey, don't use Gemini in Google Sheets. Use your Claude in Google Sheets. Something just a little quick and easy like that. Cause there's a lot of teammates that have to use spreadsheets and they're still using Gemini. And Gemini is really not that that great for certain types of things, especially finance, where you need a lot of complex formulas. And she just showed them real quick for 10 minutes how she does that. And then she showed something else in QuickBooks. but it's meant to be like highly practical. Like, do you guys use spreadsheets? Probably. Here's how I use my coat, my Claude with my spreadsheet. Not a mind blowing thing, just like a little tip and trick. And this is what I can do. So we can try to get short. Kevin Williams: And you do these weekly? Matt: Yep. Weekly, yeah. Kevin Williams: I like it. So it's it's around Robin. Somebody is either volunteers or is volunteers, somebody's Matt: One one person a week. Yeah. Kevin Williams: volunteers or is volunteered told to come out and do it. They keep it c succinct. I like that. I think that's that's what I would like to adopt myself. yeah, but you all should be doing it. Like we talk about building champions and whatever, but anybody listening to this, if you're working with more than one person, Particularly if you're not sitting shoulder to shoulder with them, like everybody is learning this at their own pace, they're approaching it from different directions. It's Matt: Mm-hmm. Kevin Williams: wild that you know, Matt might be solving a problem in a completely different way than I would, than Sonia, my team member might be doing. And you all get to the right answer, but that journey of how to think about it is is super interesting. Matt: And one real quick tip is like you can ask your teammate also to share a prompt to help other people just replicate what they did. here's a quick prompt of how to just copy and paste this into your cloud and it will replicate everything I'm doing so that you can try it too in the future. So yeah. Kevin Williams: So cool. All right, Matt. We gotta cut it short a little bit today as I need Matt: good. Kevin Williams: to run and get on a plane. But great to see you as always. And I think there was a lot of practical stuff in this episode. So by all means, folks, Matt: Yeah. Good stuff, Kevin. Have a safe flight. Kevin Williams: grab the transcript. Matt: Yeah. All right. See ya, Kevin.