Latest / The Chief Customer Officer Podcast / EP022: The Real AI Work: Agent Command Centers, AI Slop, and Building Systems That Save Time

Transcript

EP022: The Real AI Work: Agent Command Centers, AI Slop, and Building Systems That Save Time · The Chief Customer Officer Podcast

• Jay Nathan: talk about more of this later, but I just don't like the placement of that post button. It's just in a weird spot. It doesn't so maybe there's a different place we could put it. ⁓ maybe it's over in the left hand navigation, like above feed even. Like and make it say like start a post or something like that. And then I'd like what you did about simplifying the the the feed but I I think it's like almost too sparse now. Jeff Breunsbach: ⁓ yeah. Jay Nathan: So, I mean even Hacker News has like the website, URL, the the name of the poster, but it's all just really small and compressed. So ⁓ d I don't know, go take a look at Hacker News and see what you think. I also noticed ⁓ like the algorithm for top. We might wanna we might have to play with that a little bit. ⁓ and I had a ranking algor algorithm built for the last thing that we did that Jeff Breunsbach: Yes. Jay Nathan: And we can incorporate that if we want to. But there's a time based ⁓ function there too. Jeff Breunsbach: Thought we did. When I asked when I asked Claude, I was like, did we bring the algorithm over? It like it like spit back how like there's a time decay algorithm in the feed. but I can ask it to go like optimize it more. Jay Nathan: Mm. ⁓ good. Okay. Perfect. Well we just don't have that Mickey Post, so maybe that's exactly what's happening. And it's that's fine. Jeff Breunsbach: ⁓ I think that's probably true. ⁓ actually yeah. What I'll maybe what I'll do is go ask Claude to basically like at least for right now, can it just go auto generate some posts just to show us like how some stuff can move up and down move up and down the board more? ⁓ so I can do that. ⁓ Jay Nathan: Yeah. Yeah. And I think we could talk about ⁓ how we've set up like I've set up nanoclaw to monitor this now and every new post that comes in I can respond right from Slack to this community. So you could actually show people like how to use cowork to do the same thing or whatever, you know. Jeff Breunsbach: So ⁓ that's something I did too, is I adjusted Jack's onboarding flow and like I put MCP and CLI as basically like steps in the onboarding flow, like right from the like that's like essentially like they have to acknowledge it before they move forward. ⁓ is the way that I tried to do that to try and like like you said, like encourage that type of thing. ⁓ it's a good use case for nanoclaw too. ⁓ and I guess that's like a have to ask Claude, I guess, maybe about like coming up with a multi pronged like notification. Jay Nathan: Yeah. Jeff Breunsbach: Or like, you know, like do we do notifications in app? Like in so like if I'm logged into the community, is there like a little, you know, I don't know, like is there like a little notifications button that I have to click and see all the stuff? ⁓ but I guess like trying to figure out like how do you notify across multiple things and like so that it's also not I don't know, you feel like it's not overwhelming or something. But ⁓ Okay, cool. what was the other thing I was just gonna mention real quick? Jay Nathan: Yeah. Yeah, yeah. Sweet. Jeff Breunsbach: ⁓ I also added a members tab in the top right if you saw. ⁓ and I also asked Claude to make sure that it's like locked down so people can't access it via the API or the MCP. So you can't actually like download the roster of people. And then what I'm actually doing is ⁓ instead of somebody coming in and being like, Hey, I'm new here, like I'm Jay, like as a post, I'm trying to actually like push all that stuff actually to the profile. And I'm trying to like like we should actually make profiles like The way to say hello, essentially. And so I don't know, I I think you might have good ideas about this, but I'm trying to think like what should go on the profile. Like right now, I think I have this little tech stack thing. So all the tools that we add to that tech stack tab, that little matrix that I'm building of like what are the tools that you could be using, that essentially is now generated on your profile and you can go click which tools you're using. Jay Nathan: Yeah. Jeff Breunsbach: And then anytime I go log into another profile, I can go see like, ⁓ we're using the same tools or we're not using the same tools. I can basically see your tech stack versus my own ⁓ as like one thing. And then I think I've asked it right now, like what's your what's the top problem you're trying to solve? But I think I guess like I I'm trying to avoid because I feel like in in Gengar Ottain, it became like a nefarious like, Hey, I'm Jay. You know, I have twenty years of experience, here's what I'm doing. And everyone's like, Hi Jay, hi Jay. Yeah. So I'm like, so I'm so I'm trying to figure out like it Jay Nathan: Yeah. That's not a post. Yeah, agreed. Jeff Breunsbach: Can we make the profile, I guess, like rich enough that it's actually like, hey, this is the way you kind of introduce each other. ⁓ so yeah, think about I guess think about other things. Jay Nathan: Yeah. Yeah, yeah. Sure. And you can even put you can even put new joint like new people in the feed so that it like a like a like a new post. Like you make the post for them or put their profile in the feed as an item. Jeff Breunsbach: ⁓ interesting. ⁓ or I wonder yeah, I wonder if it becomes a card or something in the feed and then like you click it and maybe it just expands and shows you like, hey, here's new people. Because I I did another thing that you can basically hover on anyone's profile and it'll pull up like a little modal that shows like it's Jay, he's the CEO of Balboa, like like some other details. But I like that idea. Like tr maybe trying to insert it into the feed so it's not ⁓ I guess it's more so that like they would have to and ⁓ Yeah. Okay. That's a cool idea. Jay Nathan: ⁓ cool. Yeah. ⁓ the other thing I was thinking is maybe when you're filling in your profile, a good way to make sure we get everybody's LinkedIn is say, you know, summarize your LinkedIn profile and just give us your your link to LinkedIn and then we'll we'll just go rip it and and create a summary for you. Jeff Breunsbach: a good idea. ⁓ Jay Nathan: Or if they log in with it with LinkedIn, you've already got it. Jeff Breunsbach: Yeah. ⁓ 'cause I think one of the things I did is I added a connect button. Like you can basically go click I think it's on profile. ⁓ yeah, I think we have LinkedIn as a as a field. And then if you fill it out, I think when I go look at your profile, I think there's like a connect button or something that'll take you directly to somebody's profile or something like that. So but yeah, I think it's a good idea. Jay Nathan: Sorry. Yeah. Yep. ⁓ one other th well, ⁓ one other thing I just saw here is we should put other in the tech stack. Jeff Breunsbach: Yes. So we can capture more. Yep. Yeah. Cause ideally I in my mind, like what I wanted to happen is like that tech stack matrix that I'm building that shows like, hey, here's the market and here's where all these tools are, like, should be almost like never ending. Like in the sense that like I would hope new people join the community and basically like like you said, like there's a a way basically you can add a new tool and then it gets populated to that tech stack diagram and then it shows on the like the profile. ⁓ Jay Nathan: Selections there. Yeah. Yeah. Yeah. Jeff Breunsbach: And then over time, what I was trying to do is on that tech stack is if I go click like Fathom, I'd want to see like here's like 20 members who use Fathom if you want to talk to them about using this tool, you know, like that would be like the like the thing in the future. And then like also like if Fathom's ever mentioned in a discussion thread, can I can actually be just auto generate that into that Fathom thing where I can see like, ⁓ somebody's already talking about this tool and like what they're doing with it. And it and that's just from the community itself. So like it Jay Nathan: Yeah. Yes. Jeff Breunsbach: Basically that tech stack, I don't really love the name tech stack, by the way. I I don't know if you have ideas on that too, but ⁓ but I think that should also be a way to generate discussion that I think is gonna be pretty valuable because I think that's how people are thinking right now, like the tools that I have and how are they being leveraged, like what are people doing with the tools that they have kind of a thing. Jay Nathan: Yeah. Cool. Jeff Breunsbach: ⁓ all right, sweet. ⁓ I'm gonna play around with a bunch of this right after we I've got a couple hours this morning. So ⁓ okay. Welcome back to another episode of Chief Customer Officer dot IO. We're gonna jump in here. ⁓ yeah, I just recorded the first seven minutes. ⁓ I could cut it again. All right, welcome back to another episode of Chief CustomerOfficer.io. It is this is gonna be coming out on on ⁓ Thursday, June eighteenth. Jay, how we doing? Jay Nathan: Awesome. That's great. All right. Are we recording? ⁓ cool. Awesome. Sorry. Doing good, but the question is how are you doing? Got a new baby in the house? That's awesome? Getting some sleep or no? Jeff Breunsbach: We do. Yeah. ⁓ you know, it's it's sparse. ⁓ so now we've got ⁓ you know, I've got two older boys and ⁓ we've my my wife just delivered ⁓ baby girl. ⁓ it's been exciting around the house. I think everybody's just in a I think we're all learning new rituals and ⁓ we're all learning, you know, new habits and whatnot. So ⁓ it's changing but it's been fun. ⁓ especially the two older boys right now. At least they're in their stage where they they're very sweet. ⁓ you know, the baby is very fragile, so ⁓ they're like Soft hands, soft touches, they're kissing, ⁓ kind of blankets and feet and foreheads and tops of heads. ⁓ so it's been it's been fun. yeah. ⁓ but I feel like you're on the other end of the spectrum. It sounds like, you know, you're it sounds like you almost you you've got a ⁓ a summer free for the first time in many years. Jay Nathan: Aw. Well, I can't wait to meet her. That's awesome. Yep, yep. Mine are getting older. Second one graduated. Third one's driving now. So it's just ⁓ my wife and ⁓ Beth and I. ⁓ most most evenings, which is pretty sweet actually. So ⁓ yeah, we've been getting used to that. So definitely ⁓ in the opposite end of the spectrum from you right now. But but Jeff Breunsbach: Ha ha ha Yeah. ⁓ man. Well, I have not stopped playing around with AI and actually before we even get to our auto generated ⁓ or our generated ⁓ script from Mav, our podcast producer, ⁓ have you heard of kickbacks.ai? This is something I just wanted to throw like wanted to throw off to you that I don't think you've heard of. Okay. So there's a guy who ⁓ I Jay Nathan: No, no, no. I have not. Jeff Breunsbach: I'm gonna butcher it. I forget. He he basically is in hardware. He's building some he's building he's w he works at a company building something in like space tech hardware or something, I believe. All of a sudden he gets the idea where he's like, Hey, you know what the most underutilized marketing asset is? Is the clawed line when you type something and it's just starts saying like percolating or you know, whatever, like the little clawed symbol. I don't I don't know what you call it, the byline is what I'm calling it. ⁓ so he c he created a script. Jay Nathan: Yep. Yep, yep. Yep. Okay. Jeff Breunsbach: That changes that into a marketable asset. And now he is, if you go so go to kickbacks.ai right now, and you'll actually see if you scroll down to the bottom. So he pays people out. So like myself, I downloaded this. So he's giving me 50% of the cut that he's getting from companies that are trying to now market themselves on this tool. ⁓ and then he's have it. Jay Nathan: Okay. Jeff Breunsbach: And then he's having them similar similar to like a Google AdWords, he's having them bid on these campaigns. And then he has like an open basically at the bottom, it looks almost like an open source where he's like showing you how much people are bidding on a thousand impressions of this. And Jay Nathan: But how does he okay, I wanna know how it works. Like how is he getting this into so just for context, this is okay, this is ⁓ a VS code extension. So you have to be in Claude you have to be coding in VS Code, right? Is that what I'm what I'm reading here? Jeff Breunsbach: Correct. But but you can so I here, I'll show you my screen right now. Cause I ha basically you can basically hack it together. I downloaded VS Code. I like did this. So yes, it's kind of it it only works in some in certain tools that he has. But I've gotten it to work in my terminal because I have VS Code live on my screen. So let me show you real quick. ⁓ I just thought this was like so fascinating and like he literally like spun this up in ⁓ like it seems like a weekend. Do you see this down here in the bottom corner where it says add numerous? So like if I start typing like ⁓ give me all of our linear tickets. So you'll see it now changes how to get hired when you build today, no resume. ⁓ and so this is like an ad for somebody. Whoever wrote it didn't write who their company name in the in the line. ⁓ but but if you like if I hold control, I can click on I can click on this and it's a link that like links me out to their website. Jay Nathan: Yep, I do. Jeff Breunsbach: ⁓ and so it's you know, ⁓ promptbook.gg, whoever that is, ⁓ is the advertiser. And then if I like this is my VS Code. And so like if I go to my VS Code right now, I've earned one cent today. ⁓ if they look at this little kickbacks thing. So I've earned one cent. Jay Nathan: Okay. Okay. But how did you earn anything? How did you by looking at an ad you look you earned money? Jeff Breunsbach: It's yeah, that's serving an impression. Yeah. Like that's they're counting that as an impression of me like looking at that while I'm while I'm working. And so you know, whoever this company is, pitch book or I forget who it was already, you know, good ⁓ good advertising. ⁓ promptbook.gg essentially went and paid, you know, a hundred bucks for a thousand impressions, and now he's delivering the thousand impressions, and then he's giving me a cut of what he's making on each of those impressions. So he's giving the user Jay Nathan: So he he's actually Jeff Breunsbach: So I've I've connected up my Stripe account and so he's like passing me a kickback that says like, Hey, you you watched this ad, thanks for participating, basically. And then also he's taking a cut to run the business. ⁓ and then this company is paying for the ad spot. Jay Nathan: Very interesting. I could see all kinds of problems with this, but I'm not gonna poo-poo, I'm not gonna crap on it. Jeff Breunsbach: So yeah, I I don't know I mean I don't know how, you know, who who I don't think this is like I don't think this is all I think like for for starters, right? Like I imagine ⁓ Claude can make one code change and shut this down you know within within seconds, I would imagine. So I don't think this is a viable business, but I think he's like treating it yeah. Well if you go look, I I again I forget his name, but if you go look on Jay Nathan: ⁓ probably. Yeah. Yeah. It's very clever though. It's very clever. Yeah. Jeff Breunsbach: on Twitter and X right now. He's just writing tweet he's like open sourcing tweeting this like hey just had this idea. Hey just spun this up and like, you know, like people kind of eat that stuff up. And so he's like just live coding like he's all of a sudden like, I he's like this thing took off. I couldn't he actually didn't have like stripe verified. So he couldn't pay people out. So then people were like turning against him and he's like, no no I promise. Look like I can only pay out five hundred dollars a day. And he's like showing screenshots of all the stuff. Like he's like I the money's coming. I promise like it's coming. Like Jay Nathan: Yeah. Yeah, yeah. Wow. Jeff Breunsbach: ⁓ so anyways, it was it's I thought that was pretty fun. I I don't know if didn't know if you'd heard about it, so I thought it was a good starter as like a you know, you spun that up. Jay Nathan: So interesting. No, that's really cool. That's really cool. And I mean the the problem with all this stuff is as soon as you put it out there in public, it's you could replicate this so easily, right? But he's got a little gravity already. He's got a following. He's doing it the right way because the moat is the attention. It's not the actual tool or the technology, because the technology is so democratized to all of us, right? We can't go do it. But like now he's already got something there. Jeff Breunsbach: Yeah. Yeah. Jay Nathan: So there there's a lot of value to being the first to market with something right now, I think. Something novel. Jeff Breunsbach: Well, I think the the point that you're making too is the one I've I'm imagining you've seen that chart by now. I forget who it was, but somebody put out the chart basically number of apps has like exploded by five X, but they but the actual engagement on those apps has actually gone down or like flatlined. And so it's you know, like you said, it's like the you know, you can build anything in the world. It's you know, a tree falls in the woods, who who hears it, you know? ⁓ and I think that's the people are realizing that like ⁓ solving for audience Jay Nathan: Skyrocketed. Jeff Breunsbach: ⁓ getting attention is actually the hardest thing to do. It's not it's not the ideas, it's not building the product anymore, right? It's actually ⁓ garnering enough attention and weight in the market that you could actually build a viable business around it. Jay Nathan: Yep. That's right. All right. So what have you been what have you been building over the past week since we talked? Jeff Breunsbach: ⁓ well I've gone on paternity leave, so I'm out of I've I'm out for the next four weeks ⁓ for I guess from my my day job. ⁓ but like I guess on the way out, like you know, like some of the things that I was ⁓ I was using Claude Cowork quite often because it's connected to all of my tools ⁓ to try and close up loose ends before I was leaving for paternity leave. So like I I kept saying, Hey, I'm leaving in two days. What are the things that are still, you know, where are loopholes? What are things that are open? And so like Jay Nathan: You you built a baby. Jeff Breunsbach: I guess having that versus like having to remember myself or you know, it's like across linear Slack, ⁓ Gmail, ⁓ and not only like internal stuff, but external stuff with customers, like it was just a nice way basically to start ⁓ I almost treated Claude Cowork as my like to do list because I'd like wake up every morning, w I'd like log in and be like, Okay, ⁓ I I'd have a scheduled task run at about seven AM and it would go across all these tools, it would pull in stuff and say, Hey, here's loose ends, here's things that you need to close and then it would like remind me like you have two days left, you have one day left, you have zero days left, like getting ⁓ to like leave. So I think like that to me was actually like super helpful ⁓ just to be able to tie that up. And then again, all that knowledge, like I pushed that into a notion doc for my team and like basically sh almost showed like here's all the things that were outstanding. Here's where they here's where they are right now. And that was just like an easy way to to hand off things. ⁓ I met with my CEO. He like, you know, was super appreciative of like, ⁓ yeah, this is literally like I could click into any of these links and it links me directly to the conversation Slack or directly to the linear ticket. Like But it has context that I'm adding. I don't know. It was just a good way to to do it. ⁓ so that's one, you know, that's one thing that's super simple that ⁓ was was helpful ⁓ for me before I went out. Jay Nathan: Cool. Can I can I tell you something that's starting to bother me a little bit? I have ⁓ Say this carefully. I am starting to receive a lot of big documents from people on my team and externally that I work with, my partners, and ⁓ I don't like big documents, and I cannot lie. Like it is it's it's getting to the point. I read a really great article ⁓ this week that said something like. Y if you really want people to pay attention to what you're sending them, then craft it as if you meant it for them. And I'm receiving a lot of documents that are large documents that are just generated by Claude and the first draft of something. And I think this is something that I'm gonna have to deal with. You and what you said about having a document that you've shared around with your team just made made me think of it. I don't know. I'm sure yours is better formatted than than maybe some of the what I'm talking about here. But I think we're really gonna have to start enforcing this with our teams, which is it's not okay to send me version one of what you get from Claude, right? In my position, you know, I'm trying to run a team and build a company. And that's not, you know, anything bigger than what most of our listeners are trying to do. They're trying to run a team, you know, contribute to their companies, maybe build a company. So we should I d I don't know. This is like an emerging thing. Maybe it's just now happening to us and and ⁓ and I'm experiencing it. Jeff Breunsbach: No. I Jay Nathan: for the first time and I just don't like it. But anyway, ⁓ yeah. Sounds like you're seeing similar Jeff Breunsbach: It yeah, yeah. No, I so I think about two things. I agree so I agree. ⁓ I've come across this often and I'll tell you like a quick anecdote ⁓ where I've or where I come across this is I send so a part of my hiring for CSMs right now, I send them an exercise and I tell them it's gonna take you less than an hour to do. I also want you to use AI. I like explicitly say it, like I know you're gonna use it. I want you to use it, feel comfortable using it. If you don't, if you actually if you don't use it, I'd be more concerned if you didn't use AI. The exercise though is actually not a QBR, because I think that's what typically people have done is like, hey, a mock QBR is part of the CSM hiring process. I actually list out I actually list out like five or six scenarios of customers. Here's here's customers that we're working with. Here's they're all in different situations right now. They're all going through different versions of things. One's got an upsell, one's got a renewal, one's got a a delayed onboarding, one you need executive sponsorship from. And ⁓ I basically ask them to go through each of those scenarios and kind of gameplay like how would you handle the situation? What are the steps that you would take? And then how would you prioritize these things throughout a day? You're working on all these things in the same day, prioritize them. And so more so like, hey, how are you gonna handle the day today? And I'm looking for, you know, certain signals and whatnot. And so I again I tell them to use AI. Jay Nathan: Yep. Jeff Breunsbach: And y I'll tell you what, the number one thing I look for is if they put their hands on that document before before I see it. And I'll tell you there's a handful already that I, you know, we start the forty-five, sixty minute interview and I'll read something or I'll read through it and I'll just explicitly ask, like, did you did you put your hands on like did you, you know, ⁓ adjust this after you used AI? And a shocking amount of people will try to say no, like will try to say yes, I I did. And like, ⁓ it really looks like it, you know, ⁓ I I'm noticing these trends, blah, blah, blah. Jay Nathan: Yep. Jeff Breunsbach: ⁓ but there's a couple people that have actually come out and said like, no, I didn't. And ⁓ I actually like thought there was like much more respect for that of like telling me, you know, up front that you didn't ⁓ adjust it. But that's like one area. Yeah. But but that's like one area that I've noticed it where I just like I already know you're gonna use AI, but please just like go put your your own fingerprints on it like after you you do it. ⁓ so then there's two things I do with my team right now that I don't know if if you think that they will be valuable or not, but one is is we talk about that often. ⁓ Jay Nathan: Yeah, it's like, well, we can all save the next half hour then. Jeff Breunsbach: I I I try to showcase that as well. Like I'll generate things a lot with with Claude Cowork or Claude ⁓ Claude, but then I'm I'm in documents, like adjusting things, changing it. Hey, like this sounds better. This is what we need to be doing. and then I'm like coaching my team on that too. Like every time you use it, you've got to put your own fingerprints on it before it gets into a customer's hands. So, like just trying to train that as like a behavior that like every single time you use something, like you've got to put your fingerprints on it. You can't trust the answer. The other thing I've tried to do is put data in front of them that shows like how inaccurate these systems actually are. Like if you just auto-generate something and just put it in front of a customer, it doesn't, it's not 100% accurate. You don't even know that. The last thing I'm doing, I'm sending you this in our little chat. ⁓ but have you heard of the Minto Pyramid? This concept. So I ⁓ so so Minto Pyramid is a communication skill. ⁓ it's effectively a way Jay Nathan: Yes, you've you yes. Yep. Say it though. Talk about it. Jeff Breunsbach: To I don't know where it came from. I learned it from a site called untools.co. They have a couple of cool frameworks on there. I don't know who runs that thing, but I've used it bunch. ⁓ But the idea is to make your communication more efficient and clear. And so there's a pyramid, and there's the tip of the pyramid, which is the conclusion, and then there's key arguments, which is the next layer down, and then there's detailed information. And so I think what I've taught my teams is that too often we start with detailed information. Then we go into our arguments and then we get to the conclusion. Right. That's like generally almost like how you're taught to ⁓ like write in school. Like more often is like you need to basically get all the details out. ⁓ but this mental pyramid is basically inverting that and saying, like, tell me the conclusion first, give me the key reasons why you got to that conclusion. And if I need more convincing after that, then give me the like the detailed information. And so I've created a Claude code co-work skill for my teams based off of this. Jay Nathan: Right. Jeff Breunsbach: And so then like every time we're putting together a proposal, ⁓ a response to a customer or something, I'm asking them to run it through that first. ⁓ and actually where it's become most effective for me is I'm actually telling them to do that before they send requests to my CEO or to our like head of product. I'm always like, go use the Mental Pyramid because like if you tell them the conclusion first, like they're actually then going to be like, ⁓ intrigue, like, ⁓ okay, you're telling me that this customer, they want to do X, Y, and Z the product. ⁓ super interesting. Tell me why. Like, what are the key arguments? ⁓ Jay Nathan: Yeah. Jeff Breunsbach: There's two or three compelling reasons. Great. Let me figure out if I need to know more or not. Like anyway, so I've ⁓ created a cloud cowork skill that I think ⁓ I guess is a another way that I'm trying to help our teams basically get out of this idea of like creating too much information. Like I'm actually trying to condense it down. Like I want and when I ask for things, I'm always trying to say, give me one page or two page, like give me something that I can read and digest. Like ⁓ our proposal documents to customers, I basically have cut off like three pages is the max that we're gonna send something, and it's got Jay Nathan: Yes, yes. Jeff Breunsbach: And even even at that point, like I'm I'm like pasting in like a a pricing sheet, you know, like there's other things that are going in there. It's not just like pros. But ⁓ so anyways, those are a couple of things I guess I've tried to do, but I I agree. Like I think like the bloat is real in terms of just like receiving a bunch of AI slop and like somebody trying to pass it off, like, ⁓ it it's you know, the idea almost of like, ⁓ my ⁓ you know, back in the day, my white paper is a hundred pages. And you're like, I don't I don't know why we're saying that as like a good thing. Like You're asking your audience to read a hundred pages before they get to the actual conclusion of like what you're saying? Like that doesn't seem like a good thing. But sure that means I guess that means thoroughness, or at least it means a a version of thoroughness, but ⁓ anyways, I'm I'm rambling at this point. Jay Nathan: Yeah. But it's that that used to be a differentiator when we knew that everything was handcrafted to some degree. And now that we know it's generated, it's like, did you even read this yourself? Like, is this your best work? That's another way to do it, right? Is this your best work? No, it's Claude's work. So you're you asking the question is not good work. Jeff Breunsbach: Mm-hmm. Yeah. Yeah. Well I think the other interesting part that you mentioned though is like this idea of personalization. Like, you know, I think that's going to continue to go to the nth and nth degree, right? Because ⁓ the more that I know, this actually I think gets back to your idea of c ⁓ I g I continue to refer back to your idea that your your company has code. Like, if I knew Jay, if I if you had a file that was Jay Nathan and it was basically a particulars about how you work, how you ⁓ process information about your, you know, ⁓ Jay Nathan: Mm-hmm. Yeah. Jeff Breunsbach: I don't know, the types of pros that you enjoy, like whatever else. Wouldn't it be great if I like built something and then was able to go to your GitHub and pull your J Nathan skill file and basically be like, hey, go write it as if this is for Jay and then be able to send it as a personalization? Like, you know, we I I think we thought about personalization in the past, and really personalization I think was just segmentation at scale, right? It was basically just like, hey, this group of people likes this, this group of people likes that. ⁓ and now we're I think we're starting to like see that, ⁓ you could get to true personalization, like, ⁓ Jay. Jay Nathan: Yep. Jeff Breunsbach: Yeah, Jay likes documents three pages or less. Jay really enjoy you know, Jay wants like writing that is direct. He wants the c conclusions up front. He, you know, wants like whatever arguments put in this way. And like, ⁓ wouldn't that be cool if you received a document that said the same thing as my as like I receiving a document, but it was in a way that I interpreted it or like that I could ⁓ I guess reason with better than in better than your version. Jay Nathan: Yeah, exactly. And some people want the detail underneath too. But I think some of this, like the Mento principle that that you've, you know, instilled in your team, it's no different than like an old journalism, you know, trope, which is like don't bury the lead, like the headline should tell you the story and then the details are in the article. ⁓ I even have coached my team and we've created a skill for presentation review before it comes to me or any of the other leaders on my team, which is Go do an executive readout, like run your presentation through the executive readout skill. And it has a very specific instruction in there that is your the headline of every slide should read like a headline, not a noun, right? It should be a statement about what it is you're gonna show in detail below. ⁓ and you should not bury that lead, right? Like in so what now what? What are you actually saying on that slide? So, but I do like your idea of everybody having like a preferences, although I think. Jeff Breunsbach: Yeah. Yeah. Jay Nathan: You know, in some ways that's just good communication. It's also the smart brevity format, you know, which is, you know, start start with the the end and then unpack it as you go. I think it it's it's always been a deficiency that some people have in terms of ⁓ how they communicate. Like some people just think in detail and that's fine, but it's a communication. technique and strategy for good communication to communicate this way no matter who you are, right? Because our brains weren't meant to we need to understand the big picture before we drill down and understand the details, even if you know we do need both. ⁓ so part of it is just courtesy though, right? And just good professionalism. ⁓ and I think the what this is teaching me, and by the way, I've had to turn this around on myself too and say, okay, like have I been doing this? I probably have. Am I setting the wrong ex wrong example here at times? Probably yes. And I need to be more careful because I'm if I'm doing it, everybody in my org is going to do it. Right. So ⁓ anyway, yeah, I didn't mean to go on to a whole, you know, rabbit hole about that, but that was just just just an interesting thing ⁓ that I started to run across this week. Jeff Breunsbach: Yeah. Yeah. One other ⁓ one other quick thing I was ⁓ I've been building ⁓ that I'm curious if you would have a take on is ⁓ so I right now at work I use linear as like a capture system for tasks for myself. Like I just have my own kind of board and you know, use it with sprints and like the idea of like me trying to get into this like flow of work and whatnot. ⁓ so what I've cr what I've tried to get Claude Code to do. Is set up a way that I could essentially create an ingest system. So it'll go through my emails and my slacks. And then basically a modal pops up and is basically like, hey, here's information from these slacks and emails. Like, are these things tasks that you feel like you need to do? and then I can push a button and it goes and creates an autologs, a linear ticket. And then after that intake process, it basically goes and categorizes my day and looks at my calendar and tries to assess like, okay, what gaps do I have in terms of like working on things? What is highest priority? ⁓ so I don't know, it just ⁓ in my mind too, like ⁓ it's been fun to mess around with some of these tools to even just get more out of like my I guess like my day to day like stuff and and feeling like I've got more control over it. Again, like I f I feel like too often I feel, you know, especially we've got slacks and we have we have customers in Slack, so pretty quickly I feel like my day can just get like wrapped up and like reactive, somebody's pinging me, whatever. And so I've been trying to be much more deliberate, like, okay, here's my two to three things I gotta get done for the day. You know, here's the realistic things. Jay Nathan: Yeah. Jeff Breunsbach: ⁓ also just trying to capture, you know, I'm part of a leadership and executive team, right? And I think like some customer stuff, I don't know. I get sometimes myopic on customer stuff when like, hey, we've got quarterly planning coming up next week. Like, have I prepped for that? Have I given some feedback? Have I done some stuff? And so, like, I don't know. So I've been trying to create, I guess, a more flexible project management or like I task management. I feel like these all have taken the same shape. Like Asana, Monday, whatever, they've all just like just throw your stuff in versus I guess like trying to help. Jay Nathan: Yeah. Jeff Breunsbach: ⁓ create a system of capture. I don't know if that makes sense or not, but like I guess like so I've always just thought of like these tasks like basically a a project management or task management system is only as good as me inputting the stuff into it. Right. Like I have to put the tasks in to begin with. And that's the hard part I think for anybody. And that's where I I've been trying to use cloud code like hey, can you actually just go out, surface all these things and then like figure out if if these are the right things to add ⁓ with like a level of detail that's nice. Jay Nathan: Right. Well that that's right. And and that's like the gap that an agent fills that none of the tools ever did by themselves. Jeff Breunsbach: Yeah, that's true. ⁓ so that's another one. But ⁓ what have you been building? What have you been working on? I know you've always got big brain going. Jay Nathan: So All right. So we we've been working on ⁓ what we call Agent Command Center for Balboa. And ⁓ I I think I I've talked about this a lot, so I I I won't go too deep into it. But what so I started playing with so every let me back up. Every cloud provider, Azure, Google, ⁓ who am I leaving out? Amazon, they've all got agent studios. Okay. And they're all part of like their their ⁓ their cloud platforms. Like Microsoft Azure is Microsoft's, you know, cloud, yeah, platform as a service components, AWS, similar. and so these agent studios, they are really fully featured. They're gonna let you go build agents, connect data sources. ⁓ capture memory and history, but they're they're still very techy, right? You still are probably gonna be an engineer building agents here. ⁓ and so what we've set out to do, and again, I don't think this is really unique. I think there are a lot of platforms out there that that do this. They're charging, right? They're they're they're chargeable, expensive platforms, enterprise grade. But what we have set out to do is build an agent studio for our own team internally. And that agent studio allows us to essentially create an agent on the fly. So I made a post about this last week. We had a call last Friday with the team just going over what this thing could do. And we started talking about some of the requirements of our delivery team, like what our head of delivery, some of the things that she needs to be able to basically automate parts of her job. Like she talked about coaching for our staff. ⁓ based on call transcripts and emails and slacks that we have access to. she talked about ⁓ how we automate basically time entries. We spent a lot of time on that because we are a consulting firm. ⁓ we talked about building project summaries on a weekly basis, which we have to, you know, we have to post for our partners and for our customers, but all that's just like busy, busy work, right? So ⁓ so ⁓ We we started getting really excited. We literally took one of the things that she was talking about, our our staff coaching need requirement, which, you know, we're probably doing today, but a little more reactionary and a little bit more, you know, one one off as things come up. And we built an agent right there on the fly by chatting with this agent command center. So what we've what we've done is we've basically opened up a chat interface of our own. It's not in Claude. It uses Jeff Breunsbach: That's cool. Jay Nathan: Anthropics models behind the scenes, but it could also use open AIs, but it it can optimize models too, right? That's the other sort of backdrop here. ⁓ but what it allows you to do is basically chat with this thing and create an agent. And then if you want to build out skills, you chat with it to create skills. ⁓ and then as you sort of interact with the chat interface, you can refine those skills and you can commit them to the skills repository. in the agent command center. You can connect tools like Hubspot or whatever has an MCP, right? You can connect a direct API to this thing now. ⁓ because not everything needs to go through MCP, which we can talk about later. ⁓ you can connect data sets to it, which are essentially if you have you know a customer data set that sits somewhere, it's been scrubbed and cleaned and normalized, you can go connect your Snowflake instance to it or your Databricks or, you know, whatever, wherever your cloud Data warehouse, Redshift, whatever is running, you can connect it directly to this now. So it can any agent that we build, you chat with it and say, hey, use this data set when you're coming up with or when you're doing the work that I've assigned you to do. Then there are a couple of ways that you can trigger an agent to run. You can either run it on a schedule, sort of like you would do in cloud co-work, which is all that you're doing in cloud co-work, right? Is you're setting up an agent to run on a routine basis. ⁓ or You can have it respond to an external trigger, which might be a webhook. So a webhook, just for those who don't know that terminology, is you know, if you create a new opportunity record in HubSpot or Salesforce, they can just basically send out a signal that says, Hey, I got a new, I got a new opportunity created. And we can catch that in the agent command center and then do the do the appropriate thing with it. Kick off the right agents to basically run and go do what they're supposed to do with that information. We recorded a new call. Okay, well, let's go update our knowledge base about the customer that it was with. Let's go, you know, create the action items from that call. Some of that stuff is hand handled in our fathom, but you know, beside the point. Basically, what we're creating is a command center that is not married to one set of models. It's not technical in nature, right? So the the vision behind this is that my team can chat with it and create agents, right? I want them to create agents. ⁓ not necessarily. you know, have to do it for them. There will be some technical tasks, right? Which is why you're always going to need an ops person or a Ford deployed engineer to help, you know, connect data sets, make sure the APIs are interacting appropriately. But that's, you know, sort of trivial work relative to the hard work of getting the process out of somebody's head and into ⁓ into the environment. So anyway, that that's what we're working on. I know I've talked about it before. But ⁓ we've made a lot of progress on it. We've got some secret sauce in there that is really cool in terms of how it remembers and stores the things that it learns from you as you interact with it as well, which is what makes it really powerful. ⁓ so yeah. Jeff Breunsbach: So a couple of questions. Yeah, this is cool. No, and I think this is like ⁓ I think we should like spend time on this because I feel like this is what people want to hear is like the like the real stuff and how so I guess a couple questions. Like one, what what made the decision for you want to want to build it versus like you said, obviously I think you could go I guess like I'm looking at it as like three layers and you can tell me if I'm wrong, but I you could almost go down to the tool level where like I could build agents inside of a tool, right? But that seems too myopic or that seems like okay, I'm I'm almost like Jay Nathan: Mm-hmm. Jeff Breunsbach: ⁓ you know, then I'm locked into a specific platform. I have to connect probably tools to that platform or data sources when maybe it's already connected to other things. So like it feels that feels like very myopic. Then the next level up seems like ⁓ going to a cloud cowork or somewhere that, you know, is more of a a a layer across multiple systems and tools. And then it seems like this would be maybe like the highest layer, which is like, okay, we're we're almost tapping into the server, ⁓ that we're storing a bunch of information and and so that's really sitting on top of almost all information that we're pulling in. I don't know if those are the right layers, but that's the the way I think about it. So ⁓ I guess is that in in your minds, are you almost like getting to the the highest source that you can so that you can control it again at the end of the day? Like I guess it's it's more about the control ⁓ and the and the process of like almost like owning your own destiny versus being locked into these pre built or defined specs. Jay Nathan: Yeah. Yeah, yeah. That's it. I mean, it's it's a it's a lock in thing. ⁓ any vendor, right? And you know, it there's nothing wrong with with with choosing a vendor to do this, right? Because there are like I said, I think there are plenty of tools out there. I'm actually thinking one of the things we're considering is whether or not we just open source this, right? Because I mean, yes, it's it's got some pretty magical stuff in it, but it's not something that any good engineers probably not thinking of, right? Just to be real fair. ⁓ but ⁓ yeah so so vendor lock in is is one of those things. Now I think you have to sort of counterbalance that against you have to counterbalance that against what it takes to build and maintain something like this. Right. We have a vested interest in doing that because we're also building a data and AI practice. Like we want to know this stuff inside and out. Not everybody's got that ambition. That's you know, that's completely fair. So, but vendor lock in is one thing. ⁓ I think ag agnostic to any of the models. So actually I wrote another post the other day talking about how Claude and and Chat GPT are just not good enough. So what we all need, so the model is one thing, right? Models are one thing. And and I think the fable five thing, we won't go into detail on that 'cause we don't have time, but the whole debacle with the models this this past week, ⁓ Jeff Breunsbach: Yeah. Jay Nathan: Is is really telling about where the value is actually going to end up residing. I mean, these models are gonna be commodities, no matter how much money we're spending on them today, from a CapEx standpoint to get them stood up and get the data centers ready. They're gonna become commodities, compute commodities. So what's really gonna be valuable is the middle tier, the harness layer, the layer that says, okay, here's where all the business logic lives, and here's Jeff Breunsbach: Yes. Jay Nathan: How the decisions are going to be made as to which model I'm going to use. Here's how the the data sets of the business get connected to the model for, you know, intelligent decision making, all that kind of stuff. And that's, you know, that is the core value of my business, right? The way the the processes, the the way we interact with our team and coach them, the way we ⁓ hire, the way we go to market. That's all the secret sauce. And, you know, do I really want to? like stick that in some some tool that only the engineer has access to. The answer is no. So there's a usability piece that that was my other reason for doing this, both from a, you know, Claude doesn't really do it the way I want to do it kind of standpoint. And the hyperscaler platforms are very technical and have a lot of crap to wade through to actually do it well. Jeff Breunsbach: Yeah. Yeah. Going back to your point too about the the models, I think what's also as you kind of look back, right? Like what's interesting is you can see how OpenAI pretty early chose the consumer. Like they they went almost like D to C, right? And that's why like in anthropic, all of a sudden I feel like to people kind of came out of nowhere, but they chose a different strategy to your point, which was like they went to secure business contracts. They went to how can I go basically lock these people into Jay Nathan: Yep. Yep. Jeff Breunsbach: ⁓ instead of, you know, going to each individual user, let me go to a Fortune five hundred company, a Fortune one hundred, Fortune one thousand, whatever, and like let me go that route. And so you can start to see how those are like you know, now they're they're I think they're trying to compete at both you know, they're all trying to compete at I guess multiple lenses, but ⁓ but ⁓ you can see how that was like as part of their early plans. Jay Nathan: Yeah, absolutely. and well, we we don't really need to get into the details of it right now, but they've sort of shot themselves in the foot a little bit with the way they've handled this whole Fable Five release because I mean, good grief. Like, talk about an own goal in terms of yeah in terms of like shooting yourself in the foot. It's it's crazy. ⁓ they had such a commanding Jeff Breunsbach: Yeah. Jay Nathan: lead in the business side of things. But even us now, I'm just like, good grief. Like we can't place our trust in any of these model companies. They're they're too erratic. and they actually don't understand the business workflows that sit on top of their tools. And that's that's what I need to solve for. Jeff Breunsbach: Well, I just I was reading an article about how that's become like I I think like what you're alluding to is becoming ⁓ I guess more and more known. 'Cause like you know, the the KPMGs of the world, I think that's you know, that they're trying to go for that space too. Hey, how can I go own this customer? And basically, like you said, let me help set up the systems when Cloud's trying to do the same thing and open AI is trying to do the same. Like they're actually all competing now for the same business, ⁓ at the end of the day, which is interesting. ⁓ All right, last question on the the ⁓ agent that you built. Jay Nathan: Yep. Yep. Jeff Breunsbach: ⁓ or the agent command center is ⁓ I'm curious like or you you mentioned like the usability piece ⁓ as well and it sounds like I guess now you guys have meetings on Friday. So how are you trying to perpetuate I guess like building and using these tools? Is it through really like a lot of show and tells like hey almost like weekly you guys are coming to the table saying, okay, I built this who built you know who built something else? Like just trying to be like as open and ⁓ because I feel like that's probably gotta be something that you feel like is driving momentum internally is like, ⁓ people are just seeing possibilities every single week because it's like, ⁓ Jay went and built this and Chip went and built that. And like, you know, Keith's doing X, Y, and Z. Like, and so like that almost is just like the I don't know, the weight 'cause I feel like people are getting stuck in this idea of like, ⁓ I just give my team the tool and then like they're not using it. But like this feels like a little bit of the weight weight behind it is like you're you're kind of forcing it through like, hey, Jay Nathan: Yeah. Yeah. Jeff Breunsbach: Every week I'm gonna show you something and like I'm gonna expect that you guys all come to the table with something else about how you're doing your job better. Jay Nathan: Yeah, and I think the the other part that's becoming more and more important about this, and I feel like this is we've talked about this a lot, but is it's really important now to start prioritizing the problems you're trying to solve with these tools. Like let's solve something valuable now, right? It's not this isn't playtime anymore because this stuff is just too expensive to to be doing playtime. Yeah, yeah. Or generating some HTML pretty HTML file for the sake of, you know, like Jeff Breunsbach: It's not just writing an email. It's not just like hey hey hey, write me a response for a customer. Jay Nathan: No, like let's go automate something that we have to do every day. Let's go build, you know, something that like let's actually try to take work off of our plates. ⁓ I I did a ⁓ inter we just closed in a deal yesterday and I did an internal knowledge transfer kind of ⁓ doc because I've had does you know probably a dozen calls with this client, ⁓ have countless emails, all the calls are recorded, of course. And so I went and pulled together a full you know, internal knowledge transfer. And I was like, well, this is the last time I'm gonna do this this way, right? But it but the way I pulled it together was I just wrote a prompt in Claude, because everything's connected that we need, told it what I wanted, had it bring back the results in a basic, you know, text file, markdown file to begin with. And then I'm coaching it through how I want it to format that. And then that's going to turn into a skill, right? So but anyway it So yes, I think that that's we are we are trying to keep it top of mind enough to where, you know, people are inspired by what they're seeing and they're like, ⁓ my God, well if you can do that, then I can do this, right? But we we are trying to keep it grounded in things that are actually going to save us time and improve the quality of our work or improve the quality of our work. That's the other aspect of it, right? Building a coaching agent for our teams will allow Amy to basically see where we have opportunities to coach our teams and what's the most urgent coaching that's needed across our team. So it will help her, it will help her prioritize and, you know, spend time where it really needs to be spent. Jeff Breunsbach: Yeah, that's cool. Yeah, the last thing I'll mention real quick, ⁓ so I know we've got to wrap up is ⁓ I I heard from somebody last week who started to think about this concept of like planning as a business. Like we should actually be having a a backlog of AI projects, like you said, across departments. And then like the department leads should actually be looking at that pipeline, starting to say, Okay, let's groom this, let's start to like flesh this out. What does this really look like? Which ones take priority? But like effectively now, like almost like treating the business as a little bit of a a product manager where you're like, Okay. Like we need to start thinking about like said, like who's gonna get the most value and what order of operations are we going in? Or like, you know, what has dependencies on each other, or like how do you start to ⁓ 'cause like you said, like I think there's just too much to be too much at stake now. ⁓ cool. Jay Nathan: Well, and the one other one other piece of that which I'll tie back to the command center is having one place where all the agents live is is an important thing now because we I guarantee you all of us are duplicating our spend all over the place right now. Every tool has got AI in it, doing the same stuff over and over. You know, you've got your agent platform, I've got my agent platform. ⁓ you know, you've got your departmental tools which have agents built into them. Like it's a it's a mess. Jeff Breunsbach: Mm. Jay Nathan: Right. And it's a duplicative spend mess. And I think we have an opportunity to, you know, streamline some of that with how we're how we're thinking about consolidating all all of our agents in one in one platform. So we can leave it at that. Jeff Breunsbach: Yeah. ⁓ I'm I'm actually living I'm living in that ⁓ mess right now too. We have multiple tools, multiple agents running across things, and I'm literally now thinking like, okay, we just go rip that out and start. Like we might as well just start do it now, like while we're still early and young. Like we might as well go like start fresh with something like this, command center. ⁓ cool. ⁓ all right, well it's good to see you. ⁓ this is fun to talk through. I think like these live projects people enjoy. So ⁓ I think these are to me like the coolest parts. ⁓ all right. Talk to you next time. Jay Nathan: Yeah. Yeah. Yeah. Awesome. All right. Talk soon. See ya bye.