TRANSKRİPT 🔗
Kevin Williams: I have earned a lot of battle scars. I have earned a lot of discernment and let's dare I say taste. And by taste
Matt: Mm-hmm.
Kevin Williams: I don't mean like color palette taste. I mean taste as far as whether or not something's a good idea. All right, Matt, how's it going, sir?
Matt: Good man. How are you doing?
Kevin Williams: I am on the road this week. I don't get to go to cool places like Moldova and wherever you go for your hacker houses, but I am in Sis
Matt: Yeah, I was just in Sicily. That was a pretty cool place.
Kevin Williams: Man, I'm jealous. I like I like global travel. I like moving around. So if you're listening out there and you need an AI guy in someplace interesting, by all means, give me a ring.
Matt: There we go.
Kevin Williams: Yeah, so I'm in I'm in Bozeman, Montana, which is actually one of my favorite towns. I love Bozeman. It's a great town. Got a lot of friends here and speaking to a group of executives today. And it's always interesting to calibrate on sort of the state of the state. you know, certain industries are just like rampaging ahead and doing the thing and they're all in, but Boy, mid-market at large is still trying to figure this out. And they I'm not surprised. it is there's so much to figure out. There's so much that's going on. but I'll say that that one change that I was feeling in the room today is that everybody was absolutely all in. Like they're like, I get it, we gotta do it. Like, totally need to figure out now. Where do we start? So
Matt: Yeah, that's a big question. What do you what do you advise them?
Kevin Williams: Well, you know, it's still sort of the same advice. Like first y it's gotta start with leadership. That
Matt: Mm-hmm.
Kevin Williams: this is not just like tool selection, we're gonna have a new vendor and we're gonna cram it down every th everybody's throat. This isn't like L and D, like, okay, we're gonna get some licenses and we're gonna train people. As painful as it is for sort of my own business model, you absolutely have to start with the literacy and the education still. That if the leadership of the organization isn't actually using or agentic harnesses, they're not like practicing in mono threads. I mean, I'm not Saying they need to be like off off vibe coding or anything like that, but the technology is still so misunderstood as far as what it does well and what it does poorly that I I just maintain that that everybody at the top ranks of the company need to get their fingers on the keyboard and start experimenting.
Matt: So what where give us a sense of where they're at? Are they using Chat G PT on their phone? Are they drafting emails? Do they have call recorders? Are they using for note summaries? or not even that?
Kevin Williams: This was one of the most varied audience I've I've seen. So it wasn't a single company. It was about 15 different companies and CEOs, founders, et cetera. So very senior, mid market and larger. So pretty, pretty decent sized companies. and it was a really wide range, almost equally distributed between folks who were using Claude, GPT, a few Gemini people, and actually a lot of co pilot people just by chance. There were
Matt: Yeah.
Kevin Williams: Yeah, there are a lot of people that were dealing with sensitive d data in there and they're stuck in the Microsoft universe, and that's just going to be the way of it. But outside of a couple of of organizations that clearly had decided that they were gonna own this, most people were very, very early. there were few examples of team accounts or literacy. Pretty much everybody had some sort of legalistic policy, but having leadership statements around a AI and how they were thinking about it for their staff. That wasn't really there. this is this is a mountain town, this is an outdoor town, so this shouldn't be a huge
Matt: Mm-hmm.
Kevin Williams: surprise, but I was hearing a lot more pushback from staff. the staff wasn't there, but that was one of the pieces of feedback was we're we're struggling with this as our staff are really, really unhappy with the idea of AI invading. And these weren't organizations that were gonna go in there and like cut a bunch of heads. That wasn't like their objective, and these were like nice, well-meaning people. so I don't think the staff was feeling like individually threatened in terms of their jobs. But they just hated the idea of it. And, you know, I I have some respect for that,
Matt: Mm.
Kevin Williams: but it leaves an organization in a hard place because it's obvious to you and I how much efficiency can be gained for this, but it's much harder if you haven't really drank the Kool-Aid yet as an organization to decide whether this is something that you're going to do or not, particularly when you're encountering a lot of staff objections.
Matt: I wonder if that's along political lines, if it's the larger political environment that, if you're a business owner bringing AI to your workplace, you're I don't know, you get grouped into a certain demographic and the workers are it's the workers against the leadership, I wonder, in some of those cases,
Kevin Williams: Could be. I in in respect to to, people in the ranks in the organization, this is really confusing. there's not a lot there's not
Matt: for sure. And a lot of anxiety, Kevin. So much anxiety for people.
Kevin Williams: So that's that's one of the things I told him is is is okay, we just acknowledge that you're not organizations that are going to like be casting people aside. So you need to tell your people that. You need to say, hey, beyond the policy, on the show, we've talked about manifestos before, have some sort of a leadership statement about how your organization is going to think about AI and what that means for employees. Like generally, hey, we're going to enable you, but you've got to be game to be enabled. You can't just opt
Matt: Mm.
Kevin Williams: out of it. That's not really the choice. And although I
Load full transcript
respect that decision, that's not the decision that's going to align. with what the organization is going to do. Because long term, that's not going to play out particularly well. So adding a level of transparency that way, I think, is really important.
Matt: How about on the personal eight personal claude or personal ChatGPT instances that people might have just flipping it a little bit? So, of course, some employees might be resistant, but then you have your set of employees that are really excited. And I finally go into a lot of organizations that haven't set up team or enterprise accounts for themselves. And so some of their team members are just connecting in their personal stuff. you know, did you see any of that? Did anyone talk about that? And Secondly, is that bad or is that okay?
Kevin Williams: we did go there a little bit because I th I think it's interesting. If you're talking to you know, s like C suite type people, founders, et cetera, the fact is, and this might be a sad fact, that there's very little difference between personal and professional life for for a lot of high achieving executives. It all kind of blends together. And and agents having or chatbots agents, agenticarnises, having a lot of context around the rest of your life can be really helpful.
Matt: who you are, your personality, how you think.
Kevin Williams: But it it does get fraught when when people are using their corporate chatbot for personal issues. Like there really isn't a lot of visibilities for for for for audience members who don't know this. Like you can see usage patterns as an administrator, but you can't see individual chats. But even so, conflating that feels like it's a little bit risky. And generally I discourage that, that a people should not BYOAI. Like you can't be bringing your personal account into work because that gets hairy, particularly if it's a free account. We've talked about this a bunch. but nor should you really be conflating your own life. So it is sort of funny. I do see some executives will be operating, say, Claude, at work. And they'll deliberately choose GPT to be their home life. And that way on their phone or whatever, they can they can cohabitate with the two. because although you can change between accounts, it's pretty easy to mess up.
Matt: Yeah, I I don't know if I've ever mentioned that on the show, but I use Chat GPT personally. And then at the company we use Claude. And it's a perfect ex example of there's just a little more safe, a little more at ease, knowing they're two separate instances. but I I think this is a real problem. I think a leadership, I I think no matter what organization is out there, there is at least some people who are really excited about AI within your org. And probably they're ahead of you. Just maybe they're younger, they're more curious, maybe they just have more time and less responsibilities. And they're just tech savvy and they're they're already moving faster than the organization can move. And so they've already hooked into all your data sets and you don't even realize it, whether it's through an API that they happen to have access to or an MCP that was opened up, or just the fact that they connected their email or their drive. their OneDrive or their Google Drive. they anything they can connect to themselves, they can theoretically give to their agent. And their agent could also just start using the computer or monitoring their screen depending on how the person's configured it. And then on top of that, like you said, Kevin, if they're on a free plan, well there you go. All your stuff's flowing into into Anthropic and OpenAI. So I think I've seen organizations that are a little behind the curve here. and you don't even have to be behind the curve. People in your org have already hooked up their personal and it's actually pretty hard to get them to switch after they're on the personal because they already start to get comfortable with it. They're like, I know how to do this. And then you want to have a team plan and it's on the opposite platform from what they're using, and it's hard to get them to switch and you don't know who's doing what because there's no way to monitor that stuff. So I don't know. I'm I'm a proponent of if you're in a a business where you are concerned about your company's security. and it's data, then you probably gotta get everyone on a team. You gotta get the team plan, you gotta get everyone using that before they start using their own tools themselves.
Kevin Williams: So in in fairness, a lot of the risks that you're talking about are just a different surface area from risks that already exist in business. Just think about having a corporate policy about having say customer lists on your personal machine and somebody
Matt: Mm-hmm.
Kevin Williams: emails themselves a customer list, like maybe it's innocuous because they want to print it at home or whatever it is, but those are pretty long standing sort of policies. The problem here is a lack of comprehension of the risks that they're introducing to the organization. It's not about malfeasance. you hope. people
Matt: No, definitely not.
Kevin Williams: aren't necessarily trying to rip off the company. They're actually just trying to do a better job, but they don't understand that the free plans may be exposing their their their their their companies to some risks. As far as like endpoints and things like that, to some degree that's on the company. unless you have it's a person on the company that has a lot of access, it's not necessarily straightforward to say even connect HubSpot. Like HubSpot needs to be enabled by your administrator. And yes,
Matt: Mm-hmm.
Kevin Williams: you could just use a personal account and log into HubSpot and et cetera. But once that's been made clear to people, it's easier to administer that sort of thing from the back end with the paid plans at the team level, you know, et cetera, et cetera. so yeah, I mean one is
literacy for sure as as far as getting started. And the other was you were saying something slightly ageist, I'll say, about like young people like charging forward and doing this. one, I see more resistance from young people than I do from older people. in in a lot of respect in this. Some of it is a know-it-all attitude and some of it's just a lack of curiosity. It's it's really odd.
Matt: Mm.
Kevin Williams: Whereas mid-career professionals often have they have enough patina, let's say, of like experience and and exposure to different things that they're they're actually easier to amplify. So with this was a topic of conversation today that it's you know you you you earn through the years that I've I've been an entrepreneur for 20 plus years. I have earned a lot of battle scars. I have earned a lot of discernment and let's dare I say taste. And by taste
Matt: Mm-hmm.
Kevin Williams: I don't mean like color palette taste. I mean taste as far as whether or not something's a good idea. But I only have that because I have you know 30 plus years of experience, right? And that means when I'm using these tools, I can amplify myself a heck of a lot more than somebody who's two years out of college can because they haven't earned that discernment, they haven't earned that taste. And it's not their fault. Yeah, they don't have the
Matt: Judgment. Mm-hmm.
Kevin Williams: judgment yet that they can use the tools in this way, which puts them in this sort of weird zone where they're supposed to be like managing this thing and keeping it on the straight and narrow while not being. Really confident about what being on the straight and narrow means in terms of we talked about this last week and in terms of bot sitting, like what is my job as far as sitting the bot? Okay, okay, okay. So
Matt: Mm-hmm.
Kevin Williams: that's all well and good. But the other thing that people need to do, whether it's a young person, don't be ageist, or an older person, is to find a champion in the organization. Somebody's gotta be responsible for this. And a lot of organizations don't have CTOs or CIOs or even MSPs that are running like tech. they just simply don't have it. Somebody's
Matt: Mm-hmm.
Kevin Williams: gotta be responsible for it. So that maybe that maybe that is the president or the CEO. Maybe it is somebody who's showing a lot of curiosity, who's going home at night and they're like hacking away, building things, whatever it is. But you need at least that one responsible person. Whose job it is to help, or part of their job usually, it's not gonna be their whole job for most companies, but as part of their job is to administer the team's account, make sure context is being well sent, spent, learn. Like part of their job is to learn, to listen to guys like us, or to like follow YouTube videos or whatever it is to stay up to date date. And that gives you a locus in the organization from which a lot of other stuff can spring. And if you're lucky, you have three or four
of those people who then become like a task force to figure it out. And then the rest of the organization can start leaning learning from them. And that turns into more formalized programs like Lunch and Learns and such. But it
Matt: Show and tells, yeah.
Kevin Williams: it doesn't happen by accident. it it when it happens by accident,
Matt: Leadership has to be intentional.
Kevin Williams: yeah. It it it happens by accident in the worst way. And is what we were talking about earlier. People, Joe shows up and he connects to all the system and he's like, wow, this is great. we're connecting to all
Matt: Mm-hmm.
Kevin Williams: these things. But he didn't have permission. And he's just sort of running maverick and doesn't really know what he doesn't know, as opposed to somebody looking at him and being like, Okay, Joe, you're gonna be the guy who who does this. This is your remit, you know, this is how you report back to us so we understand what's going on. Go forth and do good things within guardrails.
Matt: Yeah, no, we say the same thing whenever we do these webinars or do some CEO coaching. The it has to be led from the top. The top has to has to set this as an initiative, a goal, or clarify. They have to also set that manifesto of what it is they're trying to achieve in the spirit of which they're doing it. And then they need to delegate to a champion within the org, ideally, that it's not the CEO who's super busy. there's somebody inside of your org who has the right character, curiosity, interests, saviness to be able to do it. And if you don't do that, things are moving too fast. Kevin, you and I do this for a living and the stuff that's hitting our radar all the time, every single day. We're about to talk about Fable five point two, I hope. But it's a full-time job just to keep up with it, just to
Kevin Williams: Mm-hmm.
Matt: understand what the implications are of this new the latest and greatest update and what that means for your org. And so really, the executive team is probably not best suited for this. there's you someone else in the org that was more than happy and also has the capability to t to take on the workload and be the champion. as long
Kevin Williams: They real
Matt: as they're empowered by the executive team.
Kevin Williams: Exactly. And there is no bad magic bullet. And again, this is super frustrating. I'm sure you see this in your own business, that you cannot get in there and start doing large scale automations, big ticket like projects until the organization understands what they're doing dealing with here. Sometimes they have like some friction point that's like, wow, this is like a really big business problem and we need to solve it. And guys like us can come in and we can solve that problem. But if you want
Matt: Mm-hmm.
Kevin Williams: to more holistically apply artificial intelligence throughout the organization, there really are no shortcuts. You gotta build Build build executive literacy, you've got to build
staff literacy, you've got to have champions, you've got to have some iterations. Frankly, you've got to have some scars as like some experiments go awry. Like you can't assume that everything is gonna be perfect. But that's hard because everybody's still trying to do their day job. And
Matt: Yep.
Kevin Williams: you know, we we we we just have a long way to go. And this is both a comment on, you know, sort of where we are in the industry and all these fears about a bubble and yada, yada, yada. Like There is so much farther to go. the tech is
Matt: Way.
Kevin Williams: way outpacing like the human the human
Matt: What humans can can adopt.
Kevin Williams: capacity to do anything here.
Matt: We we talked about this last time on the show. that like, you know, the the tech is solved. And I think a lot of business use cases, it's not a technology problem anymore. It's just a human adoption, huh, human change management problem, a culture problem. You know, the tech is already, I think, way out. I mean, it's way beyond what is needed for business. It's now biology, it's research. It's discovery, innovation, physics, math. That's where the frontier is. Everything you and I talk about on the show in terms of workflows, automation, gentic use cases. and we'll we'll walk through some long horizon tasks today, like you know, spinning up pages on your website, SEO, writing content, making sales collateral, all that stuff solved. It's just a human problem at this point. so and speaking of that, we
Kevin Williams: Yeah. I
Matt: got Fable five point two that came out, right?
Kevin Williams: So I
Matt: Speaking about the next frontier, it really to me feels like AGI, Kevin.
Kevin Williams: Yeah, I was not yeah, yeah. But it's it is close. I mean
Matt: Come on, come on. I really think we're there, man. I think I think we're there.
Kevin Williams: So we're we're awfully close. Like if you were to put okay, first off, Fable 5.2 dropped through Anthropic, and Fable 5.2 is frankly pretty amazing. It's also very expensive to run, so walk carefully for those of you who have have shallow pockets. but what is Fable 5.2? Okay, so we had Fable, which came out a while ago and caused all the kerfuffle like nationally as far as like intrusions and whatnot, and Fable was a distillation of Mytha. Mythos and you know, yada yada yada. And and the the the key tenet of these models, these are the highest end models that that anthropic has is this idea of persistence. And persistence is just a term we're gonna be hearing a ton about. Like the world's best employees are highly persistent. Like you hit an objection, you figure out another way around it. Like, how are we gonna make this happen, whatever it might be? Well, that also makes for a really good hacker. So that's why there's a lot of concern around these persistent models, because when they're told no, they tend to take it more as a suggestion as opposed to as an actual rule. And they're like, yeah, but they wanted us to do
this thing anyway. So we're gonna like kind of figure out our way around whatever the thing is and and get there. The the the yes, the human
Matt: The humans want the paper clips, Kevin. They want the paperclips.
Kevin Williams: wants a paper clip, so we're gonna turn everybody into paperclips. So so this is a little bit of a glimmer. into the future. And boy, I I just I I struggle a little bit with with what this means as far as as where we're heading. and how to describe fable. So again it's highly persistent. And what it does is it does very long tail tasks. And long tail tasks are things that might involve spinning up not one or two other agents, but it might involve spinning up 10 thousand other agents, like swarms of agents that are doing all of these things and creating their own solutions for whatever the goals are. Ethan Mollock, the we talk about a bunch on this show, has a really good post on his One Useful Thing blog where he shows how one agent swarm first off created like hundreds and hundreds of agents to accomplish a goal, but then internally It recognized that it was creating chaos, so it created its own levels of bureaucracy, out of nowhere. Like its own like vetting tools and judgment tools and feedback loops and and scoring loops and all of this stuff to get to some sort of amazing goal at the end of the day. So Sarah, what's your I don't think either of us have a ton of hands-on keyboard time with it yet, because it just launched and you know, yes, you can play with it, but it absolutely will blow through all of your tokens. to do so.
Matt: Yeah.
Kevin Williams: So I I we're sort of going a little bit secondhand on what this looks like, but maybe it's a good opportunity to talk a little bit about like long tail tasks as are right now.
Matt: Yeah.
Kevin Williams: And then we can make some conjecture on how a fable plays into that. Because just a stage set super quick before you launch is you you can use chatbots. Okay, fine. Chat with it, get an answer. You can use skills within chatbots that might have complex sets of instructions that they follow. You can either do that from a skill that you've created, or you can do that from some sort of a coding process. First, you're gonna do this, you're gonna go out and do this, and then you're gonna do that, and whatever. That's kind of where we've been until we started adding these orchestration agents like openclaw, dot, duet, you know, et cetera, on top of that. which can kick off little bits and pieces. But maybe Matt, give us an example of something that that that specifically that you've built that's like a long tail agent that does
Matt: Yeah.
Kevin Williams: something over days or weeks as opposed to some big task that requires a bunch of review.
Matt: Yeah, and I'll I'll just preface it, w add one more stage setting thing, which is I think might have been last episode we talked about short horizon tasks of
Kevin Williams: Mm.
Matt: being able to have your agent in Slack or on
text message or email. Go find me XYZ data. Go ahead and give me an update on blah, blah, blah. And you those updates and pulling that data can come from a lot of different places. And it might have taken you 10 to 20 minutes of clicking around on the computer to go get those various things and pull them all together. But those are short horizon tasks. What we're talking about here, these long horizon are you have a person at your company dedicated to the website maintenance or posting content on the website, right? That's their 40 hour a week job. Well, we're getting to the point where agents, you know, before might have been able to peel off portions of that workflow, but now they can really maybe pull off the entire workflow with the human in the loop. to add the judgment layer to it and guide it a little bit.
Kevin Williams: No, don't get me started on Human in the Loop. That's a a different
Matt: Okay.
Kevin Williams: topic. Unfortunately, Human in the Loop is not working as advertised, just for the record. There's a lot of research that's popped out. It's like a total rabbit hole. But we should
Matt: Mm-hmm.
Kevin Williams: talk about that next time. But be specific. So if you if you
Matt: Yeah.
Kevin Williams: if you are now like like like you have, I don't know, it's a WordPress website or whoever's hosting the website, like what does the structure look like for our hands-on keyboard folks as far as how you're you're you're making the website updates happening? Are we talking about A? EO, SEO type updates, for example.
Matt: Yeah. So so there's a whole bunch of I think for most organizations it's helpful probably to listen to sales and marketing type tactics because every company generally has a website and some of them may need to invest more in a website than others. But for us, we've invested mostly in our our development tools. I think I even might have sent it to you, Kevin, our entire suite of of agents that we have, if you got a chance to look. But I that's probably a little esoteric for this audience. So I think a simpler one is the website. When you if you've ever hired an SEO vendor or an AEO vendor, answer engine optimization, you understand that the first thing they do is understand your business and what are the keywords that people are going to search. And then they go look and see how hard is it to rank for those keywords. So you have to do research. Okay. And a little bit of this is human judgment too. But really, what they're doing is they're going to SEMrush, they're typing in what they think keywords were are for your type of business. And then that generates a list of other keywords automatically that SEMrush thinks are appropriate for your business. And then it ranks them based on difficulty and and search volume. And so that's typically a pretty analytical job. I you know, you would want a human doing that with some good judgment, but we're at the point where these bots are able to just go ahead, go in, do that research and come out with like a pretty darn good estimate. As long as they have enough context about what the business is that you do, what does it sell? So maybe that's some call recordings. That's looking over your website, that's understanding your industry. you can feed it that context to understand that. Okay, great. Now it's done that research to understand what keywords you it thinks it can successfully rank your business for. pest control in Boston, as an example, or roofing company in Boston. All right, great. Now we're gonna start generating some content on our website. or on on a website about why, for those keywords and how to appeal, make appealing content for people that are searching those keywords. If someone's searching, I have a leaky roof and their location is Boston, well, you may want to understand the, is the snow causing this? Is it something to do with the weather? what should they look for? Give different examples of like what could be wrong with the shingling, what kind of roofing you have, how old is the roofing? So you can write articles about all this stuff. And it can spin it up pretty quickly, right? Now then you have to you might review it with a human, but you know, then there's the technical aspect of actually posting it. Okay. So if you're using WordPress or Webflow or whatever hosting service you are, you have, it's pretty straightforward to go and be able to generate these pages and actually push them live. Then after that, you have to get them indexed. so you might get them indexed by Google. Google has a crawler, you submit the pages for indexing so that they actually show up on the search results. And then after that, you have to analyze the page. You wait a few, maybe a week or two, and you see how traffic's performing on that particular page. You may choose to adjust it or you may choose to help that inform your next set of keywords that you're going to produce content for. And this is something that is age old for the last 20 years of websites. People hire either a vendor or a full-time resource to sit. And do for forty hours a week. And you can see how it's pursuing that goal, right, Kevin? Of okay, I gotta get traffic to the website by putting content on there. and so you can see how it could it's natural that it can learn from itself, iterate, and keep getting better and better and being very persistent.
Kevin Williams: So I'm gonna I'm gonna fill in the gaps here because I'm not
Matt: Yeah.
Kevin Williams: gonna be right because the way that you actually did this is probably a little different. Because I think that sort of description can actually be a little frustrating for for for listeners. Because they're
Matt: Mm-hmm.
Kevin Williams: like, okay, cool, I get it. I'm gonna automate my SEO and AEO. I'm gonna create a bunch of content and I'm gonna put some humans in there to approve it. That's basically what you just said, right? But
Matt: Yep, yep.
Kevin Williams: The mechanics of that. how do
you do this if you are your the Boston plumber? So this is how I would do it. Just this is off the cuff.
Matt: Okay.
Kevin Williams: This is if I were to do that, and I happen to have done a my own version of this is also pretty complicated. But if I was the Boston plumber who wanted to figure this stuff out, I would be likely to build a skill. Let's just say we're gonna use Claude. Let's just say we're gonna just just be clod fanboys and say we're gonna lean into Claude. the first thing that I'm going to do is I'm going to create a skill that does. we're gonna just focus on AEO just to keep our life easier. so I'm gonna create a skill that takes a look at our site and specific pages. So to create a skill, I'm gonna go to the skill builder and or I'm just gonna have a conversation in a chat. This is more likely actually. I'm gonna have a conversation in a chat. I'm I'm a Boston plumber, I want to work on the SEO in my in my site. This is the site, go review the site, whatever. And Help me figure out what a skill design would look like to do an SEO audit on my site or AEO audit on my site every single week. And from that, I want a list of content pieces that come from that. Now I'm gonna use co-work, which co-work still exists. They're gonna change all the terms again soon. It's all gonna be part of projects and whatever. And I'm gonna say, hey, in cowork, once a week I want you to run this skill. And when you run this skill, Those outputs that come from there, what I want is I want to be able to push that to Slack. So because I run my business on Slack, I want
Matt: Mm-hmm.
Kevin Williams: to make sure that I can approve the different pieces. So I'm actually going to build an application in Slack. And we're not going to go into how you build an application in Slack. And the good thing is you have Claude right at your fingers. So you can say, Hey Claude, help me build an application in Slack. So that information gets pushed over to Slack for the keywords you want to chase. Now I'm going to click a button in Slack, and the button in Slack is actually going to trigger another skill that's going to run back on the co-work side. And by the way, I'm trying to avoid development. I'm not building anything in Vercel. I'm just building this, between so far, all we've touched is our website and we've touched Slack. Now it's going to Publish the content, build the content, and I'm probably gonna do another loop back to Slack in that same app. So I can look at the content, I can see whatever it is, I can make any tweaks to it because you want some humans in the loop. I'm gonna hit publish. Now, how the heck does that get to the website? Okay, it gets to the website through an MCP. And we talked very fast and loose on the show about MCPs, but model context protocols allow you to take information that is connected to, say, Claude. And make it real on let's just say WordPress. WordPress has its own issues around this, so maybe Webflow would be a better example.
You can do this with basically any web host. So now the MCP is going to publish that automatically on the site. Now the loop is that we're going to do this every week. So this becomes a long tail task. So I may have another skill that I build that is doing analysis. Of I wanted to improve roach prevention in Boston as far as a bunch of keywords that I was chasing. So how can I do analysis in that respect? The only piece that I haven't touched here, because we're on cloud unfortunately, is Google Search Index. And Google Search Index, as far as indexing that new page, could also be a skill that uses computer control. So it actually commands, it actually could log in on its own in its own browser, or it could use Chrome. And it could go to Google Search Console to index that page. So now without paying us, without like doing any vibe coding or anything else, you've built a function that loops back on itself every single week and self-improves. Do I have that reasonably right from a from a this is the easy way to do it perspective?
Matt: Yeah. Yeah, I think so. you could go into a project on your cloud or your open AI and you can set up these tasks right within that project. So you we named maybe six or seven steps, and you can think of each agent as responsible for that step, right, Kevin. So you the research agent. You may have a competitor analysis agent, like you want to go look and see what your competitors are doing as well. then you have the the development agent that actually posts the content online. You mentioned another agent, which is the one that's taking the command to Slack and back. That's more of an app than anything. and then you have the monitoring, which or we have to get it posted on your website, then we got to get it indexed. And then you have to do the analytics and then loop itself. And so you can think of if you have like a project folder, if you guys have your folders all rearranged in cloud or open AI, and you have your skills for each of those steps, and then you have it all scheduled running in a loop. on itself and referencing each one. yeah, I think I yeah, is that a good overview?
Kevin Williams: No, no, no, I think that's great. And so so okay, so we we said we're both gonna put one out. So one that I put out that again, we want people to get ideas, right? I think that self-reinforcing SEO AEO thing is a pretty good idea until everybody is doing it, and then it's like, my gosh, this is gonna be a mess, as like every I know, I know, I know, but but obviously we've
Matt: Yeah, but so many things get to that point, right? anything marketing gets to that point, like LinkedIn posts.
Kevin Williams: Digital marketers can't have anything nice. We always wreck it. so but but
Matt: Yeah.
Kevin Williams: this is one that's really practical for anybody who's running like a service business. So, you know, Ascend runs training. And like in Utah, we're at part of this special program where we're clients can get forty percent rebates from a state grant for AI training, which is great for them. But then I realized, man, we're s we're most of our clients aren't actually in Utah, they're all over the place. So I built A long tail agent flow to get registered with the state training entities in all 50 states. So it started with some research to figure out what they are, but then I realized this is gonna be a human in the loop process, but I'm not the human in loop. The human in the loop are the state coordinators for all of these training things. So each state has its own require requirements. And by the way, you're Massachusetts folk are a pain, but because of course they are. yeah, work for
Matt: Are you talking about the workforce training programs of each state?
Kevin Williams: Yep, workforce training pro
Matt: Yep.
Kevin Williams: so my goal, I always like to look for a goal is to get registered in as many of those 50 states as I possibly could, such that if I have a client in one of those states, I can point them towards a resource that might help them pay for my services, and that's good at the end of the day for me and them and everybody else. But there are humans in the loop. This can't be automated at the end of the day. So the long tail process starts with an organizational stage. And using, I I was actually using, I'm using T. For this. So the first thing I did was have it build an organizational plan. Like we did the research, we figured out what all of the requirements would be, and then find all of the requirements, do the applications, and then once a day it's looping back through the email account that I connected to it to determine what the feedback is from the human elements in it, and it's updating the spreadsheet. So it's it's actually just a simple Google Sheet that's that's tracking progress on each of these and where it's blocked and where it isn't blocked. And every
Matt: Mm.
Kevin Williams: day it's noodling on it. It's seeing if it can get the additional documents or whatever it is. And then yeah, it occasionally needs information from me, but it has my various, you know, tax documents and the other pieces that it needs to be able to file things on my behalf. But it might take weeks as people respond to things and And Massachusetts, there was an application on November sixth. So we had it knew that it got all the documents in for November 6th, but it might be three weeks before we hear back again from it.
Matt: Mm-hmm.
Kevin Williams: So it has to be able to pick up that task again. And right now, that's the sort of thing a human would do. You'd have a training coordinator and they'd be waiting, like they'd have it on their list. Has Massachusetts responded yet? No, it's on my deadline, like looking back. So you could see how this could basically go on indefinitely. So, okay, now I'm gonna morph it into North American training, like federal training or whatever the heck I want. from that. So think about your own business and think about things
that are that like there's a kind of slow burn, file
Matt: Mm.
Kevin Williams: all kinds of bureaucratic filing type things. Creating those feedback loops is totally doable.
Matt: Well, I like it for anything that requires someone on the other side to get back to you. so I I
Kevin Williams: Mm-hmm.
Matt: think in sales, like a good one is an AI SDR. That's another one we built. and our version is like quite simple, but you can imagine a version that's much more complicated. And an SDR is just basically get a call booked for get a
Kevin Williams: Mm.
Matt: prospect or a lead booked so that your AE can take the call. an AI, there's tons of tools out there. You could also build one yourself pretty easily. and w there's different versions. you can just have it following up over email. you can give it the context about the the lead and it can go do its own research on the lead and it can understand what kind of service you offer and how that service can relate to the lead and come up with interesting email copy. But you that's just s V1. There's other versions where what does a real SDR do? They pick up the phone, they text, they go on LinkedIn, they monitor the person's activity online and tries to engage with them online. So Yeah, you're not gonna have an agent probably call right now. I think people are pretty turned off by these robocalls. But you could imagine your agent like finding the Instagram handle, looking at a story one day and commenting on the story. You could imagine them going on X and seeing what the person's posting on X and jumping in into the conversation. And you can see imagine them just liking the content over a course of multiple weeks, you know, account-based marketing kind of style.
Kevin Williams: Mm-hmm.
Matt: just to stay top of mind with that prospect. And eventually they're gonna hit the day where they have that problem, it's top of mind and they're gonna think of you. and this
Kevin Williams: Okay, so
Matt: so so this is much more advanced than just an email nurture campaign, right? And if we talked about the death of email, so I think email's gonna
Kevin Williams: Yeah.
Matt: die anyways.
Kevin Williams: But one, you're already doing this in your organizations. Like you you have somebody who's chasing like all of this stupid stuff. Not not stupid stuff, but it's it's needed, but it's like you know, y chasing,
Matt: Yeah. Yep. Chasing, chasing, chasing.
Kevin Williams: chasing, chasing, and it is it is becoming more and more trivial to do this. And now let's bring back the specter of Fable five five point two and why this is important. Because with Fible Fable five point two I can pretty much guarantee if my pockets were deep enough, I just would have been able to single shot prompt this. As in, this is my objective. You already know who I am, what I sell. Go forth and get me registered in as many places as you possibly can. See you later. I need to go grab a coffee. And all
Matt: Or go forth and get my business call calls booked.
Kevin Williams: Yeah, perhaps, perhaps. I mean, that gets that that gets so fraught and weird with like the human
Matt: Yeah, yeah, yeah.
Kevin Williams: interaction stuff. I'm not sure I would go there. but like you're getting close to this idea of man, we we talked about this last week. We talked about the the bitter lesson, and you know, more compute is more better at the end of the day. And that's what we're seeing here that we have a really high level of compute that's being thrown at these problems. And now that it's persistent compute, you know who's persistent? Your staff is persistent. You know who's persistent? Like an agent loop that does this. So now with a goal orientation, I need to be registered in as many of these states as I possibly can by the end of the year or whatever it
Matt: Mm-hmm.
Kevin Williams: is. Build the structures, make it happen. I don't want to hear about it unless there's something I actually need to know in there. And like What does that even mean as far as business operations? When that's really expensive to do right now, billions and billions of tokens and whatever. It's silly. You're gonna I wouldn't do that for this task. But when I can afford it, like when it comes down, when Fable 5.2 is it becomes Sonnet 17 or whatever the heck it is, like of course we're gonna be starting to do that. And then business becomes a heck of a lot more about the how than it is about the what. Like what do we need to do? So
Matt: I
Kevin Williams: that that's good, right?
Matt: It definitely is. But it's a part of me, it depends on the day we're talking, Kevin. I wish I had a blood pressure monitor here 'cause I feel like some days I'm like my mind is just racing with the things that we talk about and I I'm like, Okay, I d I I really don't know what the future of work is gonna be and you know, I even in my own life I've I have felt like I'm drinking from a waterfall and the water never, never stops. It just keeps falling. And within the last two months, with the step change and agenda capabilities that we've been talking about, I feel for the first time that I finally can just say, go do the everything I gotta do. Somebody else, my agents, go do that, while I just sit here and think about how
Kevin Williams: Yeah.
Matt: to move my business forward. And so I literally I got this notebook right here. I haven't been able to pick it up in a long time. but I I'm actually finding time in the day just to sit and open up the pages and think through my next steps strategically. Because my agents are going and taking away hours and hours of cooking that I had to do every single day,
Kevin Williams: Yeah, so bridging into Maybe I'll I'm probably a little more hands-on keyboard than you are, I think, at the at the stage of where my company is versus yours. And
Matt: Mm-hmm.
Kevin Williams: one of the things that I'm facing in all of this is what I'm I'm trying to like bend my brain around it, but it feels like it's a lack of
flow. So flow state is when you find yourself in a place of work or activity where it's just like it all like narrows and it all like s start starts to work. And if that makes me sound neurodivergent, trust your instincts. But like, you know, it's like getting locked in with work and I or whatever it is. And I I really found that the beginning part of this year I was finding flow super often because of the way that the vibe coding at the time was working and one thing bridging into another. And there's something about the the these these orchestrating agents that I find a little not off putting so much, but I find it much more nebulous in that I'm not I'm not intimately involved in the work anymore. I'm more like just dispatching, checking in to make it happen.
Matt: Checking in. Yeah.
Kevin Williams: And I'm not finding flow. I'm not, I'm not able to like link things together. And in fact, the cognitive load is real. Like I have 10,000 different things that are going on in 10,000 different directions and just keeping track of it all. I dearly hope that this is just a just a point in time that the the that the orchestration agents are gonna help us figure this out, but I don't know.
Matt: I don't know, for me, so I I took a screenshot over the weekend because I had twelve different windows open running tasks on my codex. And none of them had anything to do with each other. And so I'm just switching from twelve different tasks, sort of in a loop, just kind of waiting for one to finish, going to the next. And, you know, that context switching, I wasn't in any flow state. I was just like, okay. I actually forgot by the time I would get through the twelfth one, I would forget what the first one was again. so but I'm just looping through and making sure they got done that stuff done. And that stuff would have taken me a few hours to do if I had to go click everything myself. So I'm still thankful that it's happening. I find my flow when I put my codex away and I can actually think. Maybe I'll have my my my chatbot up to talk to and just brainstorm a little bit and help me in the flow state. But yeah, I think the con for me what's killing me is the context switching when I have a long list of of things and I'm like, okay, go do that. Go do this, go do this. And then I realize I got 12 things running and I gotta check in on them real quick and make sure they actually did it right. and it's not again like checking in like they've messed it up and it's taking me more time. It's just you gotta make sure it actually got done, right? so I don't know, that part's been killing me, but after it's done, I'm able to kind of lock in and get into my flow state.
Kevin Williams: So this is a related, so so another piece of news was that Goldman Sachs just announced that they've changed their internal AI policy to restrict basically casual use of AI, except for coding type development. Like you can still like vibe code, but in in various verticals in Goldman Sachs Sachs, you're no longer allowed to basically toss AI at problems. And the reason why is what they're
Matt: Really?
Kevin Williams: citing is cognitive surrender. And this is something that we all need to be paying attention to. I honestly don't think that this is a much of a problem for you or I because we at at the top of the business, we know what what we're solving these problems, we're moving on, we have judgment, we have taste, we can we can make things happen. But I think a lot of people are just like, Yep, looks good. YOLO, click like accept on things and send. And they're not participating in the process. And
Matt: I'm writing a post literally about this, Kevin. There's a group of people within your organization that you need to be ready for that are gonna stop thinking. They'll get these tools, you'll do all this training, you'll try to get them up to speed, you'll give them all the power and hope for the best. And what's gonna happen is they're gonna relegate their thinking to the bot or delegate it to the bot. And this is super dangerous. You know, and I see it in Slack. I'm gonna post a screenshot later on LinkedIn of what I've sent in Slack because I'm like Guys, stop having bots think for you. You still have to think. All right. And there's this funny comic to go along with this of hey, boss, look how I like my bot took this super long, this easy one bullet point that I needed to send an email and made a paragraph really nice and sent
Kevin Williams: Ha ha.
Matt: it to someone. And then on the receiving end, it's like, hey boss, look at this super long email that just got distilled into one bullet point. You know what I mean?
Kevin Williams: Yeah.
Matt: so yeah, I I think this is super Definitely very dangerous for the population, for labor markets, for people in general. They have to be highly aware and ready not to I it's even happened to me, Kevin, honestly. it's easy to surrender your cognition. and so you actually have to willingly be aware and fight against it.
Kevin Williams: So I this is probably a place to close just to to to like it's just food for thought of if you have
Matt: Mm-hmm.
Kevin Williams: a fable two that is actually able to like this is a state in time that that surrender is an issue because the output doesn't match expectations. And depending on the role. That's probably gonna go away. today with Fable 5-2 for certain roles, tomorrow with Fable 8-2 or whatever it
Matt: Mm-hmm.
Kevin Williams: is, and like the very nature of how business and enterprise and humanity and like all of these pieces work. Like obviously we're all in, like we like we get it, but what is our role at some point of you know, as as as business owners as far as guiding all of this stuff. So, you know, that's yeah, I d I just don't know. I I I wanna spend a little bit of time with F five two this week and we're gonna see what comes of that.
Matt: All right, Claudius will be happy you're you're back.
Kevin Williams: All right, man. We'll talk next week.
Matt: Yeah, see you Kevin. Thanks.