Latest / The Chief Customer Officer Podcast / EP024: The Shared Brain, Forward Deployed Engineers & AI at Home

Transcript

EP024: The Shared Brain, Forward Deployed Engineers & AI at Home · The Chief Customer Officer Podcast

• Jeff Breunsbach: All right, welcome back to another episode of Chief Customer Officer dot IO. This will be coming out on Thursday, July ninth. Jay, what's going on? Jay Nathan: Not too much. It's hot and humid in Charleston, so just trying to survive inside the house. Yeah. Jeff Breunsbach: Never changes. I know. Every this is this is I always tell people this is like the six to eight week stretch that I just complain about Charleston. Like this is the only time that I will. Like I love it, you know, the other forty forty four weeks of the year, but this is like the eight week stretch that is just you're like it's too hot to go outside, you're sweating as soon as you walk outside, it's just it's brutal. Jay Nathan: Yeah, absolutely. Jeff Breunsbach: Especially with two well, you know, especially with two toddlers that like want to be outside doing stuff. I'm like, come on. Like I'm like you don't want to play with your toys inside, you know? I'm like trying to coax and like you know, I feel like a a bad parent, but Yeah, exactly. Well that's cool. anything you've been anything you've been working on the last week or so, with the AI that you've been building? Jay Nathan: Yeah. Trying to trick Yeah. Good luck with that. All right. so yeah, well, the actually I I thought about it this week. I haven't built too much over the past couple of weeks. There is something that I did start working on that I'll tell you about, which has been informative, but I'd been traveling a lot. So I was in Atlanta last week. I actually got the opportunity to sit around sit in a round table with a bunch of executives across different kinds of industries. like there was a You know, they're all technology firms to some degree, but one of them is a data company, one of them was a physical hardware type of company, hardware and software. one was a pure software company in the education space. and there were so there's about eight or ten of of these folks and done in Atlanta. It was r really, really interesting discussions. And I think, you know, one of the things that stuck out to me the most is that in big enterprise in terms of AI adoption, no matter whether you're in customer success or sales or product development, like the the big gap in terms of moving fast or the big barrier to moving fast is actually not certainly not the technology, right? The technology is way ahead of of where some of these companies are, but it's actually the it's the it's the legal questions, it's the cost questions now, like we talked about a million times on this podcast. Jeff Breunsbach: Yeah. Jay Nathan: and just sort of there there there's still a normalization happening to, you know, is it safe to put our data in these tools? And I think as you look at sort of how we got here, you know, these tools are pretty pretty safe. Like open I open AI back in the beginning, like when we started using it last year for my company, had a switch that said, don't train our models on your data, right? And so you could turn that off and on. You had to turn it off at the enterprise level, at the company level. We always did that. Claude has never even had that flag, right? Because it automatically does not train on your data. Now, does that mean they don't use your data, your prompts, your your files that you send in some way to make their service better? Absolutely not. I don't believe that for a second. They'd be crazy not to, right? You're feeding it all of this content. but I think, you know, part part of Jeff Breunsbach: Yeah. Jay Nathan: It it sort of goes back to the narrative of of AI that has been created by the big model guys and frankly, anthropic and and open AI and and the sort of the Doomer kind of story around AI. It's gonna take your job, it's gonna take your IP, like and that's so these are n these are normal reactions by very conservative people who are running in large enterprises who want to try to protect those enterprises. And so, but anyway, the the moral of the story is I don't think that most large organizations are moving very quickly on this stuff because because of those constraints. And that that came out in the discussion that I was part of last week. So it was pretty interesting. Jeff Breunsbach: Yeah. How do you think about I mean, I think this is largely what a lot of leaders are struggling with right now, is this whole idea of of moving I think we've used other terms, but like, you know, building the right infrastructure, kind of getting their teams to be multiplayer, like I think so it it almost seems there's there's almost like sh maybe two things that are coming out, which are is one, like you said, like these legal questions and like, okay, how do we keep our data safe and like are we using the right tools? And then the secondary is like, okay, now do we have the infrastructure to actually go you know, enable our teams on this stuff. So it feels like those are two probably substantial questions that like again just slow down the whole idea of adoption for for these teams. Jay Nathan: Hundred percent. Yeah. Yeah. You know, I think the do we have the right tools? I still see a lot of this. I was listening to a podcast this morning. It was very instructional, very informational. but it was very clearly single player AI, right? It's like, Okay, great, you built an app on your machine that does X, Y, and Z in your email inbox. My goodness. If I see another one of those, I might throw up. You know what mean? So, Jeff Breunsbach: Yeah. Yeah. Jay Nathan: But yeah, so it's like I I continue to be inspired by this idea of of having, you know, the right platform to do enterprise wide AI in a safe way. And I think there are many platforms out there now that that are off the shelf. You you buy them and you you deploy them internally to do it. I I don't necessarily think it's the AWS, Azure, GCP studios. Those are very much still developer platforms. Right or DevOps platforms. and so like I'm talking about commoditizing this stuff, like giving my head of delivery access to an agent where she can go build agents, you know, by chatting with it. Just like, you we've talked about nanoclaw, right? It's such an interesting and empowering tool to have running. but you know, imagine that on the enterprise scale in a in a team based environment as opposed to a single, you know, me, me getting value out of it personally, but it not. Jeff Breunsbach: Yeah. Jay Nathan: sort of going going into the team. E even us, right, Jeff? Like I'm using nanoclaw now to do help us do some of our podcast stuff, but we're not even using it well as a as a team. You and I, two people, right? So I was going to talk to you about like maybe we need to get Discord and or our own, you know, shared Slack environment, but maybe it's Discord, right? Because Discord's free and it actually works really well with NanoClaw. And we can both be in there. It can be multiple channels for the different things that we work on together. Jeff Breunsbach: Yeah. Yeah. Yeah, that's true. Jay Nathan: But like you have to have the collaboration layer with the agent involved. Jeff Breunsbach: Yeah. we we should do that. Actually it's a good it's a good call out of ourselves, that we're not using it well. 'Cause we're actually we are using it individually, right? Like we're like, cool, I built this thing over here for like I built I built something for our newsletter and you built something for a podcast and it's on different things that we like. That's pretty funny. Yeah. well this this is this is interesting too, 'cause I don't know, did you did you see Jay Nathan: As much as we talk about like getting teammates, well, you and I are not on the same page. Yeah. Jeff Breunsbach: my gosh, I'm gonna blink on his name right now. The the CEO of Zapier or Zapier yesterday, he came out with a message. and he was talking all about basically private DMs in Slack and how they're trying to eliminate they're trying to eliminate private DMs in Slack because that's essentially lost context. And like if if they're using they're they're essentially like, okay, we're we're using Slack. Jay Nathan: Yes. What's his name? Yes. Yep. Yep. Yep. Jeff Breunsbach: as a I'll call it repository, but like they're essentially using that context for building agents and for work and for all these other things. And essentially they're, you know, he is alluding to the fact that like, okay, private DMs essentially makes us slower, it loses context for us. again and it's actually going to be a bigger part. So I think he wrote something about they have an internal leaderboard and they are trying to essentially show how many what percentage of your messages are are public channels versus private channels and he's and so that that leaderboard is is a way for them, almost like a leading indicator for them to try and get people to to share more publicly and to like have these discussions in public forums so that they can use that context in order to build better agents and and get better outcomes for what they're doing. Jay Nathan: Well, there's a really practical angle on that too. yeah, so his name is Wade Foster, right? and so yeah, here here's the here's the overview. They're they're they're banning DMs. And you you sort of made it sound like you know, we're gonna peer pressure you into it with the dashboard and that kind of thing. But this sort of reads like and this is the way I initially read it too. Like they're actually banning DMs, which I think you can do in Slack. You just turn off. Jeff Breunsbach: Way faster. Yep. Jay Nathan: Right. Nobody can DM. Yeah. So you have to have a channel for everything. And you know, by the way, there's some like really practic practical benefits to that too. And and again, the world's worst here. I I will spin up a DM message with three or four people. The three or four people, the exact group that I need to communicate with, I'll do that all day long. In fact, my chief of staff, Jonathan, had has given me hell about that before. He's like, Hey, can we just have a channel? Jeff Breunsbach: I didn't know that. Jay Nathan: For everything and I'm like, whatever. But he's right. He's absolutely right on this, right? Because you can't even find stuff if you do that all the time. So again, like I'm coming to you from a place of I am not the world's, I am no, no, nowhere near an example for any of this stuff, but maybe we can learn from it and go. But so preventing information loss, when things get in a DM, shared. Jeff Breunsbach: Yeah. Jay Nathan: Building a shared brain, which is actually the thing I've been working on that I wanted to talk to you about. And then the ninety percent rule. Wade believes that AI agents need deep context, do their jobs well, public conversations give AI the data it needs. Well, of course. and okay, yeah, so there is a transparency leaderboard. they track percentage of an executive's messages that happen in public channels versus private DMs, internal competition, and it's arguing. Increased public Slack messages from thirty three to forty six percent. So maybe they haven't turned it off. There's not a switch. That's sort of what you were describing there. It's cool. Jeff Breunsbach: Yeah. but yeah it's it and I think I think there's a multitude of other companies that have come out and said this too. You know, that they're trying to use this whole idea of a shared brain and in that like the context within Slack essentially is like gold. and I think about this a lot because junction at Junction, like we are a remote company, fully remote, and so we're all across different time zones of Eastern time zone into Europe. And so like I I I think about this like, I don't know, way too much probably about how we use our Slack. Like we just went through and like I don't know. We had like an executive discussion a couple like a month ago or so about like how do we name our Slack channels and like what's the nomenclature and all this stuff and you're like, wow this th that sounds silly, but like when you use it, literally it is like the number one tool that we all use in like so much. You're I mean you've got to think about these decisions and how it's gonna like, you know, organize work and impact like productivity and other things like that. So, but I think you're so t talk to me about the shared brain. What have you been building? Jay Nathan: Okay, yeah. So so I always talk about our call transcripts. I'm in love with our call transcripts. We have you know thousands of them at this point. And by the way, it's cool to do like infographics of of the growth of that because now, like every month, we're almost creating more every month than we had in the prior year combined. That's like sort of shows our growth. but so what I wanted to do is start start building, you know, you hear about these this idea of the shared brain. Jeff Breunsbach: yeah. Jay Nathan: Right, which is how do we start to take all those transcripts, sift through them, and make sense out of out of the information that they contain? Cause it's rich, right? We're communicating with customers, we're selling to prospects, we are discussing best practices of the tools that we support and the products that we work on. we are having partnership conversations, like so much richness is sitting in those calls. Transcripts. That's where the actual real work is happening, the real communication between us and our constituents. And that would be customers, prospects, employees, and partners. So so the idea of a shared brain is you start extracting that information and you create what's called an ontology. Everybody's talking about ontologies in in LinkedIn right now. And I hate it when we like latch onto these big fancy words because it's like I had to look it up. I was like, what the hell is ontology? Ontology. What is an ontology? Because I'm, you know, I'm a simple South Carolinian, Jeff. I just need simple words. But basically, it's like it's a definition of like the different things that your business interacts with, right? So we interact with people, we interact with companies, we have projects and engagements that we do. we try to track results and outcomes for our customers. Jeff Breunsbach: I looked it up right now. I don't know what it is. Jay Nathan: We have partners, right? Very simple concepts, right? So it's just to basically think about creating a database structure. If anybody has ever created an Excel file or a database design, it's like, okay, this table is going or this worksheet is going to hold this information. It's going to relate to this other worksheet with this and with this key, right? So, you know, a company that we interact with. may be a customer, it may be a prospect. If it's a customer, then they probably have an engagement with us, right? And an engagement has this information associated with it. It's very simple, right? But what I found was so what I the first thing I did was I had Claude Cowork go in and just scan the last 90 days worth of calls, transcripts, to tell me what our what our structure is, what our ontology is, right? Jeff Breunsbach: Interesting. Yeah. Jay Nathan: And it came back, you know, it came back pretty, pretty accurate. it it did identify, of course, we interact with people and companies and we have engagements and all that kind of stuff. It also identifies some things that weren't quite right that I had to tweak and tune. So I guess you know, the thing I would tell you is that you know, people say, I just created a second brain, they make it sound so easy, but it does all this stuff takes work and and time, right? So I did some massaging with it. and Jeff Breunsbach: Yeah. Jay Nathan: So let's let's talk about what the outcome is of this. So what the output of it is is that we will have right now I've got a just a big folder full of of files, just little text files. They're called markdown files. Markdown is a a format that AI works really well with. And and we have like sort of a running history of what's going on in each one of those markdown files. So we might have a customer, right? Let's call it. Acme Corporation. So if I go into the Acme Corporation file, I'll see the summary. First, I'll see some fields at the top that are like the pieces of information that we're tracking on every customer. And then I'll have a summary of like what's the latest kind of interaction history? Like what are we doing with that company? Right. What are we working on together? What are we trying to sell? What are we, how are we engaged? And then beneath that, Jeff Breunsbach: Yeah. Jay Nathan: There's a history, like an append history. Like here's all the things that have happened over the past 30 days and then further out than that, calls, meetings, you know, all that kind of stuff. And so now that we've got that, that sounds a lot like a CRM, right? Just to be fair. but now that we've got that, the agent runs every night and it sort of just updates it with whatever's new. Jeff Breunsbach: Yeah. Jay Nathan: From the call intelligence system. So this is where you get into the concept of a loop. You want these things to be constantly improving. Now, where this becomes powerful is when that file structure is now available to all of my agents. Okay, whether that's Claude, Code, Cowork, or you know, GPT, Codex, whatever it is, right? Even if you're using Glean, if you're using some other agent, you want to have that context available because then Jeff Breunsbach: Yeah. Jay Nathan: When you go ask it questions, it's going to be very efficient in terms of going to say, okay, like we're talking about Acme Corporation. I'm gonna go pull the latest of what's going on there, the interaction history, the engagements we've had with them. You could do that with your CRM too. Like, don't get me wrong, right? But when you start doing that, that's a it it gets a it gets more expensive to token wise to to do that. So we're we're gonna build this out and then we're gonna figure out how it benefits us. Jeff Breunsbach: Yeah. Jay Nathan: Right, in the future. You can certainly ask it questions, you get answers back a lot faster than you than you could if you were, you know, sifting through your CRM, even if it's with MCP. But but that's what we're doing. So we're creating the the Balboa brain, if you will. Jeff Breunsbach: And then and then I guess like the I guess in the future as you start building this out, like is it that you're running essentially a nanoclaw or something that like your team can access in Slack that's you know essentially, hey, go, you know, go at nanoclaw or whatever and like you you can essentially then start communicating in the channels that you're already in and it just like you said, it has full context. So now everybody that's you you know, everybody at work essentially now can Almost like answer their own questions without having to go bother somebody else, about something. Jay Nathan: Yeah. Yeah, absolutely. Like get the context. You know, a big a big piece of it that maybe is a little less obvious because what I just described sounds like a CRM, but it's also pulling out best practices. So we're we're sort of advising and consulting with our customers, doing hands-on work for them. And as we, you know, do that work, we're leaving an exhaust of conversations about, hey, this is the way to do this thing, or this is the way to think about this thing. So it's actually Pulling out best practices into a knowledge base every every night as well and updating them where it where it finds them. So, so those are now available to the entire team, right? It's like, okay, if if I'm running up against this problem I haven't seen before, let me first just go ask Claude via Slack to your point. Like there's Claude tag now that can be connected directly into Slack. And let me see if there's an answer there before I go. Jeff Breunsbach: Yeah. Jay Nathan: 'Cause it's a lot faster to do that than to have to go meet with somebody and describe the situation and or to send a private DM in Slack right to go figure to go figure it out. Jeff Breunsbach: and how did you I think like what's I think what people are finding interesting is also just like the the how behind these things. So like when you say when you are like, Hey, I've got you know, I'm building this second brain, like is that in your kind of AWS like you you've essentially built tables or databases that essentially are housing this information? Like how how are you structuring this, I guess, and where is it living? Jay Nathan: Yeah, okay. This is a really good question. So here's here's the problem. So today this is mostly I did it in cowork, but I did it that way to to get started quickly, to build the right data structure for our brain, the ontology, and to spend time refining it before we deployed it to our cloud infrastructure. So it is file based today. It will be database. It will all this will live in a database eventually. Jeff Breunsbach: Yep. Got it. Jay Nathan: And that database will be, you know, to use a technical term, it'll be vectorized so that it's really easy to do, you know, search across those entities, semantic search, which means it finds things that are similar to what you're asking for, just the way AI, you know, does normally. and so it this is it's it's sort of step one right now. Let's get the files locked down and get them distributed in a way that everybody can connect them to their cloud. But step two is no, it's its own little system sitting out to the side and your cloud instance connects to that system and then eventually our own agent platform will go connect to that system to to query in real time. Jeff Breunsbach: So even this file structure that you're building today, is that in Google Drive or is that just on your personal machine? okay. Cool. Got it. And so then people c people can technically I guess like so so other people could technically point to it right now within the business, but but like you said, it's not I don't want to say productized, but it's not, I guess, to a a scale or to to a point where you want to essentially like roll it out into its own infrastructure. Jay Nathan: It's on GitHub. Yeah. So it is in the cloud, so to speak. Correct. That's right. My team, like not everybody on my team is an engineer or a developer by default, right? Now everybody's sort of learning GitHub and and we're getting them familiar with some of these more developer oriented tools, but they, you know, though that's not the first place that they're gonna spend time in their day, right? They're gonna be in Slack, they're gonna be in Pendo itself doing work for clients. They're so you know, part of this is is enablement in tools that may not be native to the workflow yet. Jeff Breunsbach: Yeah. and then when you think about I guess like the 'cause I I think you've referenced second brain, which like I think is a older it's a it's a idea that's been out there before. like I think people have talked about building a second brain. man, I'm forgetting his name too. There's a guy that I know became famous for like how to do this. But the I I would say like the concept of building a second brain was essentially Jay Nathan: yeah. Jeff Breunsbach: For years I think people have s have said, Okay, I've read all these books, I read all these newsletters, I watch these movies, I you know, have a personal life, I've whatever. A anything that you all the things that you've accumulated in your life, essentially they've tried to figure out, okay, how can I How can I turn this into databases that I can essentially access and and kind of you know all the things that I've I've experienced in my life now I can kind of go back and access as I want to and and have notes and all this other stuff. So it's like that that's been something that I think people have been talking about and trying to figure out how to do. Like I know a lot of people tried to use Notion for that in the early days of Notion, was like, this is a a new way to do it. and so now it's you know essentially the same concept just at a business level. Like how can I how can I bring this to a Jay Nathan: Yeah. Yeah. Jeff Breunsbach: level that again multiplayer that people can all access that agents can now access to which I think is the other interesting part too is you think about I think we've talked a lot I I guess about multiplayer and and kind of getting into like multiple team members using like more AI but I think also like you you mentioned you've got this agent running at night and you've I think we're starting to talk more and more about agents. So how do you I guess like right now when you are running agents Is that running just on cowork or clawed code or like how are you running agents right now? Jay Nathan: Yeah, great, great question. I we're doing a lot in in cowork, but again, I think most of those agents are single player agents. When you're doing stuff in cowork and claude, it's pretty much a lot of a lot of times single age single player. so that's why we built the agent command center that we did, which is so we could run we could run agents across the team, run the same processes for everyone. So when you're done with a when you're done with a call. Jeff Breunsbach: Yeah. Jay Nathan: It's gonna get, you know, classified and treated as a, you know, whatever kind of call that is across the board consistently. The output's gonna be the same for everybody. the artifacts that are created based on that call are gonna be the same for everybody. That that that's the difference. So, but yeah, that that's that's why we're we're creating the the agent command center effectively. Jeff Breunsbach: And I I mean I think too, like a as I've thought more about agents as well, or I don't know, I I guess I I'm starting to break down agents into like agents to me also isn't that new of a concept, I guess. Like because we've had processes that run, right? Like that are like structured that run every night that have a time schedule or time based or something, right? Like you've kind of had these things that exist. But now I think when we are thinking about agents, like the idea is that we can introduce our AI into those agents that essentially is enhancing the ability to do that repeated task or changing the ability or, you know, doing something. So for instance, like I think like what you're referencing, like you have this call agent, right, that like is going through and looking at these transcripts and classifying it in certain ways. It's sending updates to certain systems and tools. And it's making determinations in order to do that. this is a customer. Let me go update that customer's file. this is not a customer. I don't need to go update that, right? Like it's making these I guess it's making these determinations. Yeah, based on like certain certain aspects. So I think also like, you know, we've kind of had repeated we've had ways to, I guess, do repeated processes and have these things run on certain schedules and whatnot. But now, again, you you can take that up a notch as you introduce AI in order to Jay Nathan: Yeah. Judgment calls. Jeff Breunsbach: make determinations about certain things or prob you know, there's deterministic and probabilistic, but as you've taught me, I still don't I still don't know the like how those come into play or where those are. But but I think like you've essentially now, you know, as we got agents, you can start having these things update multiple systems and tools, and being able to do that I think in structured ways across the whole whole business is like what you're trying to go for. Jay Nathan: Yep, that's exactly right. I mean, here here's this really simple example. Like, so we've got this feed of of calls happening, right? Again, like all types of calls, right? We have partnership calls, we have prospect calls, we're interviewing people, we're having working sessions and discovery sessions with clients. Like the the the call recording system doesn't know the difference necessarily, right? But when you sprinkle in AI on it, it's like okay, I can If you give me a sort of a definition of what a call looks like, then a certain type of call, then I can I can sort of identify that on a on an ongoing basis. Think about what we would have done before we had AI, right? You're you're right, you could run that process before, but then you'd have to have everybody who had a call mark the call as whatever type it was, and then you process it appropriately. Jeff Breunsbach: Yeah. Jay Nathan: In reality, that would never happen, right? Because getting people to do something every time they have a call, like to update data, there are a few people on this planet that are really good at that and who will do it, right? But the reality is that the majority of people aren't great at that. And I'm I'm one of them. I'm I'm the world's worst at that. Unless there's a scoreboard, then I'm good, right? But Jeff Breunsbach: Yeah. Jay Nathan: Just as a matter of process, like I don't want to follow process for process sake. so so yeah, so it in in by the way, like one of the things we did in our as as I tried to define the structure and the ontology, like if we're gonna log all the calls that we've had with a company over however long years, then I wanna know what type of calls those were. Right. And so I actually had. Jeff Breunsbach: Yeah. Jay Nathan: Claude, go in and just analyze all the call records, literally thousands of them, and just tell me like what are the prevailing types of interactions that we have? It came up with 14 different types of calls across delivery, sales, partnership management, employment and recruiting. And it did a really good job, right? Of course. So now every time another call comes in, it's like, okay, well, which one of these does this sound like? It's going to take a quick look at the At the transcript and it's gonna scan it for you know the closest match and it's gonna tell us which kind of call it was. And it's of course, you would not be surprised at this point, it's incredibly accurate. Right. So and then you could do downstream stuff with it, right? You could take the unorganized data of a conversation and turn it into a draft deliverable, for example. Right? Those are the things we couldn't do before all these Jeff Breunsbach: Yeah. Right. So and then three stuff with it, right? Yeah. this yeah, the th I I think the thing too that's like sticking out to me about this as well, i and again calling my own self out, is like I think the the thing that becomes important for this too is like the actual mapping of processes. Like the idea of like, okay, do I know the steps? Do I understand what these steps are? Like, because if you start to do that more and more and more, right? If I start to visualize, okay, here's a little mini Figma board or whatever, and here's my process, right? Then you can start to imagine, okay, like, like you said, like, okay, are we following this every single time? Probably not. Okay. If we're not, then where can I introduce AI? What can I do? Like where can I basically, you know, not circumvent but enhance like this this process or this step in order for us to you know have a hundred percent outcome on this, you know, on this thing. And so I think that's a like a good call out, but but that's cool. The second brain. I like it the or the shared brain. but it I I the you know we we were talking a little bit earlier and I'm curious we were talking a little bit earlier about the enterprise and how, you know, it's hard to adopt AI and I feel like all these companies now are coming out with kind of their own services businesses to try and you know, their partnership organizations and services businesses to try and, hey, you know, open AI and cloud and everyone's trying to go to, you know, the Fortune five hundred companies and say, okay, hey, let us bring our services company to you to help you enhance and figure out how to use AI and whatever. And I think another version of that is this idea of the for forward deployed engineer. Like it feels like they're they're making a big bet on that too. So I think there's some some articles about that. But what what what's happening, I guess, in that that realm right now around the forward deployed engineer? Jay Nathan: Yeah, so I guess our our key story here is like seven point five billion dollars invested in four deployed engineering companies. so Amazon just announced a one billion dollar investment in an organization like this. Open AI, four billion dollars, anthropic a billion and a half, seven and a half billion. That's just on the those three organizations, right? Not to mention every other, you know, AI software company on the planet. Like Sierra, like, you know, whoever. and so I think, you know, again, like the the issue here is that the capabilities of these tools are so far beyond what the organizations are ready to adopt that you need an engineer to help do the deployment. We're actually doing this now. We have four deployed engineers on our on our team. And we are having conversations and working with clients to help. Jeff Breunsbach: Yeah. Jay Nathan: Do some of the things that we're learning how to do for our organization because it's all moving so quickly, right? A year ago, we weren't talking about agent harnesses. We were hardly talking about agents. Some people were because they saw it coming. now it's like, no, these things are going to run autonomously in your organization. They're going to be running inside of a sandbox that we call a harness, which knows about your data, it has access to your brain, it hacks has access to all your MCP servers for all the tools you use. But how do you make all that work, you know, consistently? How do you make it work in a team-based environment? How do you keep costs under control? Like we keep seeing more stories of teams utilizing their entire token budget within the first, you know, four months of the year kind of thing. Meta was the late latest one to they said, hey, we're gonna cut back on our on our anthropic spend. Next year, like they're significantly cutting their budget because they were liter I the number was crazy, Jeff. It was like I want to say tens of billions of dollars, I think, that they were spending on tokens, right? So I could be a little bit off on that number. You'll you'll find it, correct me. But but that's what so the part of what the idea is like it is, you know, these organizations and in, you know. Jeff Breunsbach: Okay. Yeah. Jay Nathan: No no pitch here, but like Balboa included, we are we are deploying four deployed engineers into our clients to help them figure these things out. Jeff Breunsbach: they said meta used seventy three trillion tokens. So I don't know if that equates to dollar wise, but that sounds like a lot. and then my my other Jay Nathan: yeah, I don't even think. Yeah, anything with a T Jeff Breunsbach: My other comment about our our lead story too is I I wonder who's doing the accounting when when companies come out and say, Hey, we're gonna make it a billion dollar investment in like hiring four deployed engineers or like, you know, it's like I wanna go back and validate like did Amazon actually spend a billion dollars like a year from now? You know, like how I'm just joking. But like, but yeah, it it continues to I guess like the the the question or the thing that comes to mind for me is it I I guess or like the the prevailing thought is that essentially you need Jay Nathan: Yeah. Jeff Breunsbach: you need the business's context and you need the people in the business, like because they've got all this history and knowledge and context and everything about the business, right? And then like essentially if I can bring that engineer into that environment, then can the engineer bring the the processes, the systems, the reliability, all these things that like are necessary, right? And if you combine those things, that's where the the power is. And so I think like that that continues to be the I guess like the recipe of of what needs to happen. But but I think this is similar. I mean, I guess this again like this this idea or this concept to me hasn't it is isn't new, right? Like I think many businesses have been using services from companies, you know Jay Nathan: Hundred percent. Jeff Breunsbach: There's agencies that have existed for things like, let me help you with your Salesforce instance, let me help you with your your HubSpot instance, right? Like these things have existed. It's just I think like you said, like this is just taking it to another level where, hey, you know, we need you to essentially like this technology is moving so fast, you know, the guardrails need to be put up and therefore like we need to rely we essentially need to hire somebody to come do that for us because like the effort is worth it and the risk is high and so we need to be able to do this reliably. Jay Nathan: Effort is worth it. That's the most interesting point, right? Because it's very easy, I think, to get in and we where we've been over the past six months in particular is in a state of euphoria around this stuff, which is like, hey, AI is gonna s solve everything. We're gonna invest, doesn't at all costs, like we're gonna go invest in it. But I think that, you know, it's just like everything else in AI, the evolution has been very fast and go going back to the token budget piece, like Jeff Breunsbach: Yeah. Jay Nathan: You all of a sudden have boards and leadership teams saying, okay, like, yeah, we're all for you leveraging AI and doing a lot of spending on it. But those are board level conversations now, right? It's like, hey, we're gonna increase our our AI budget to be able to do more. And that's there these these companies are coming back and saying, or these boards are coming back and saying, that's fine, but to what end? Like, when are we gonna see the returns on that token spend? Like, what are you building? Is it gonna allow us to have more products to to sell? One of the you know, one of the interesting things that's happening with our primary partner, it's a software company. You probably know who it is if you listen to this podcast, but they are releasing so much software that their internal teams are having to run to keep up, right? Much less their clients, right? So again, it's the enablement and the capability. well, it's the enablement is is the hard part of all this. Jeff Breunsbach: Yeah. Yeah. Jay Nathan: 'Cause now the software itself, which was always the constraint, is not the constraint anymore. It's you know, the people who need to get it out to other people. Jeff Breunsbach: Yeah. That Yeah, that's th I I think that like speaks to the chart that I think we talked about this maybe last episode or episode before, but I forget the chart that's out there, but effectively like we've developed more apps than ever before, but the usage of those have stayed basically steady. Like it's stayed which me you know, essentially saying that like, you've developed all these more things, but you know, nobody's using them more, and so like essentially, you know, we've we've kinda missed on the enablement piece. what's interesting to me too is is like I and I I'm curious, you know, as you think about I guess like for deployed engineers and I don't know, like I I guess like we can we can release more than ever. But like the question that comes to my mind is like shouldn't we be like this idea of of customer effort score, the CES back in the you know, we've we've talked about this before too, but like shouldn't we be I guess like reducing I guess the redu almost like reducing the software. Like shouldn't we be like figuring out how to use AI to basically say how can I get this job done as effectively as possible with like the least amount of software? Like it feels like we can release as much as ever, but like the maybe the question really isn't like should we should we be releasing more, but more so like what is required in order to deliver this outcome and how can we basically hit that mark. And that that's where we should stay. We shouldn't, I don't know, release more features. Jay Nathan: yeah. Yeah. I mean it comes back to a classic like product management type of well, I say classic product management, but like the best product managers in the world don't look at shipping features as the outcome. They look at was I able to drive some metric for my customer in the right direction, right? And and yes, that has to be at this point there would have to be some level of efficiency. Jeff Breunsbach: Yeah. Jay Nathan: As part of that that metric, right? Like, can I just do more of the thing that I'm doing? Let's call it claims processing, right? Can I just handle more of the claims processing on behalf of my customer? That that's what people are going to pay for in the future. We've talked about that too, right? And that's not that's not even novel to say at this point, but people are gonna pay for the outcomes that we drive. And I think we have to look at outcomes as little O, not big O, right? Little outcomes, like yes, I am going to successfully. Jeff Breunsbach: Yeah. Jay Nathan: Complete the processing of a claim, right? Of course that feeds into a bigger business outcome, right? Which is happier customers, you know, you know, claims process correctly, you know, drive savings for an insurance business, whatever it is, right? But the you know, the product itself is going to have a discrete small outcome that it that customers are buying. That feeds into the bigger business. Jeff Breunsbach: Yeah. Jay Nathan: Sort of like in line with what you're saying, part of the Amazon story here is this idea, their methodology forty five, forty five, forty five. Did you see that? Jeff Breunsbach: you did, yeah. Yep. Jay Nathan: I like I like this because it so 45 45 45 is 45 minutes to define the problem, 45 hours to validate it, and then 45 days to productionalize it. And I like that methodology. It matches with sort of the way that I've seen this play out in in our own work, which is like you can get on a call and define the problem very, very quickly. Then you've got a transcript, and then you spend the next 45 hours building it. Right, building and validating like how we're gonna go solve that problem. And then the 45 days to production seems long relative to all that, because you go from like one week to essentially six weeks. But there's a there's a building and tuning effort associated with all of this stuff, and it does not come for free. There's nothing for free, even in AI. Right. It it it will give you very impressive results quickly, but then you start digging in and you're like, okay, there's a Jeff Breunsbach: Yeah. Jay Nathan: bunch of stuff that's got to be tuned here. And that the fine tuning of these systems is what takes the most effort, which is why that 45 days to production I think is really interesting. So anyway, I I thought that model was pretty cool too in in conjunction with their FDE. Maybe we'll adopt that directly. But we we already have a version of that which is, you know, it's so funny is we're trying to we're we're we're taking these services to market, these FDE services, and you can't sell like It's not there's no such thing as a three-month engagement anymore or a six-month engagement. Like, no, no, no. We're like, what are you talking about? Six months? These are 60-day engagements. That's what literally what we're pitching right now. and it sort of matches up with this 45, 45, 45 thing. So I think it's interesting to have that kind of methodology wrapped around it. Jeff Breunsbach: Yeah. Yeah. there's Yeah. Th there's like I guess like two two thoughts that come to mind for me. One is have you heard of I'll have to go find his name too. my gosh. It was a well-known Twitter guy who did marketing services and he was a designer and his NMJ, something like that. But anyways, he came out this is now maybe a year or two old and I'm you know paraphrasing and tr probably getting s parts of this thing wrong. But he basically came out and said, I'm gonna charge you five thousand dollars a month and you basically I'm your creative I'm your creative person for your business and Like you can use me however you wish. And people are like, well, wait a minute. Like my agency tells me I only have like forty hours. So how many hours do I have with you? He's like, unlimited. And they're like, wait, what? And Lee's like, yeah. it and all these things. He's like, I have in all I have is 20 customers and I do this. And people are like, That's crazy. You know, you're never gonna be able to it's a one person business. And it kind of reminds me of this idea. Like I I think you can almost create a forward deployed engineer like if you were yourself were like a forward deployed engineer, you could probably go create your own little one person business where you're like, Hey, I'll charge you, like you said, whatever, sixty-day engagement. Blah blah blah, whatever, just five grand a month, and I'm just your forward deployed engineer for however long you want, how many hours you need, whatever. I actually think you could probably go do that. Like and I I I think like that could actually be a a pretty viable business because like you said, I think there's plenty of businesses out there that need your services. Design Joy, that's who it that's what his name is. Jay Nathan: Okay. I love it. Jeff Breunsbach: I'll go f I'll go find this, but and I'll f I forget the number, but roughly that's like this the the idea is basically he went and basically said, I'm a one person agency and everyone's This k this can't ever work. Like what are you talking about? Like, you know, my agency quotes me hours or they have a project that's time based or whatever, and he's like, Nope, this is it. so I don't know. That just reminded me of of like this idea that you could go do with a four deployed engineer. if you're if you yourself are interested in doing that, I think you could go like build a one person business. and then the other thing that that just came to mind for me too is this that like like you said, this idea of of forty five, forty five, forty five. I I don't know, I I continue to be obsessed with this idea of the inputs and like the inputs I I think like people have under appreciated, I guess, getting good at prompting in the inputs and they've, you know, basically just, hey, I can just go chat with it. Cool. I'm just gonna go throw in like a a one-sentence prompt or whatever. And I think like you said, all the stuff that you're building, the second brain, all this context, right? Like this thing actually is like all of these things enhance the outcome. And so this idea of like Like you said, defining the problem and I think getting down to this, I'm I'm kind of obsessed with the this PRD and like this idea of like, okay, how do I define something so well and so crisp that the AI can go go build it? and so I think that's like a big part of of how this stuff actually gets gets done. Jay Nathan: Well, I I think the people who think in terms of systems are going to thrive. Like even individuals right now who think in terms of of systems and structure are really gonna thrive with this stuff. more so than those who who don't. And and it brings to mind a and I know we're running out of time here, but a quote from James Clear that I'll leave you with, which is this it's you don't Jeff Breunsbach: Yeah. Jay Nathan: Now watch me butcher it. we don't we don't I'm gonna Google it. Jeff Breunsbach: I know what you're I know the quote that you're looking for. you do not rise to the level of your goals, you fall to the level of your systems. Yeah. We're gonna leave this in the podcast too. We're gonna leave that in. but yeah. Jay Nathan: There you go. I was like so eloquent up until the point of actually I knew what I wanted to say, it just was not coming out. Jeff Breunsbach: This and it but I think that i I like it it rings true. I remember reading that in Atomic Habits. and I've been thinking about that a lot lately now that I've got three kids and trying to figure out like what are our systems of of our household, you know, how do we make this stuff run efficiently? Jay Nathan: You know, you know what my big challenge to myself is? I I actually I thought about it this morning and as as I was at the gym and I was like, this is what I'm gonna do. My goal, and don't tell her this, but my goal is to teach my wife how to use Claude and Claude co-work to do stuff because number one, because I think when she gets her hands on it and figures it out, she is going to be unstoppable with it. Because she's a system thinker, like So she just doesn't she she gets the power of it. She uses Claude like in Chat GPT, like a consumer today, right? She queries it. and chats with it. But I think if I could teach her how to, you know, in non technical terms, build an agent, like use Claude Cowork, I think she's gonna crush it. So that I'll I'll keep I'll keep I'll report back on that initiative. Jeff Breunsbach: I just sent you I just sent you a video from Clairvaux's podcast, How I AI, and it was a woman who basically runs the household now with Claude in a Mac Mini. And it's got like ki and she's basic like you said, she's like developed agents that are like mining their emails for like updates on their kids' school or dates. It's adding automatically adding dates to the calendar of like, you're you know, the the whatever the the play is on Saturday at ten A. Like it's adding that stuff to the calendar. Like she so she created this whole I think she's homeschooling I I believe in that video too, if I if I remember that one correctly, she's like homeschooling her kids with AI, like in in this whole thing. so yeah I'm I'm on the same page though. I'm I'm Jay Nathan: Dude. Yes. my gosh. Jeff Breunsbach: You know. I've got like four days three days left of paternity leave and I'm like I wonder if I can figure out how to, you know, run my household on AI. Like that's Jay Nathan: Hey, I'll tell you this, man. I've been working with my nanoclub. My son sent a, he's like, you need to create this agent. He sent me a markdown file. He's like, go create this. And it's a basically a it's a health and fitness coach, essentially. And I did it. It I called it Jack because that's his name. So now Jack coaches me every day. And literally after before every workout, after every workout, like during the day, I'm like, hey, what do I eat? What do I eat next? Jeff Breunsbach: Okay. Jay Nathan: They're like, okay, you're here's where you are in terms of your protein for the day, your calories for the day. Here's what you need to go do. Did you just work out? Okay, you need to do this within 30 minutes. And like, I'm telling you, man, I can there's a difference already in just in the past week of doing this. I'm like, wow, I had some things completely wrong about my nutrition and fitness that an individual trainer, you can't interact with them enough to teach you all these things, right? So, like personal agent. Jeff Breunsbach: That's cool. Jay Nathan: Such a powerful concept, right? If you're just completely constantly chatting. Okay. All right. Jeff Breunsbach: send me that send me that markdown file because I want to do that too. 'cause I was just gonna say I so I bought this Google Fitbit. They just came out with this Fitbit Air, and it's like basically the equivalent of Whoop, but they have undercut Woo Whoops pricing. Basically Whoop like three hundred bucks a year, I think, four hundred bucks a year, and you never own the device. You you know you essentially are renting it every single year with the subscription. So they came out, this is ninety-nine bucks to buy the device, and then you have a then I can Jay Nathan: Okay. Okay. Jeff Breunsbach: I can either subscribe or not. So I can if I don't I can subscribe to their AI services for ninety-nine bucks a year, or I can just basically have the device and get all the data. And so that's the reason I did it is I was like, cool. If I want to and so now I'm testing it for one year for like the AI. And so far I thought I actually sent a screenshot to my buddy because I thought the AI's been it's been pretty funny inside of the the the tool. So here I'll I'll just read this to you and like I just think it's it's pretty funny. But it's like that early bedtime gave you a solid 81 sleep score. It's a great recovery win, but yesterday three cardio sessions and twelve thousand six hundred fifty five steps push your workload high. Like it just it's got these prompts and stuff in there. but I wanna I I w I've been trying to think about how do I use this better. Like you said, how do I like incorporate food and all this stuff and it has the ability to do it, but I I'm I don't know. I still find the Jay Nathan: Wow. Jeff Breunsbach: the friction with these types of things high and like using an app and have to log my food and all stuff. If I could just go like you said, if I could just be texting a nanoclaw that's like, Hey, I just ate this, go log it. Hey, I just did this. Like Yeah, okay. Send me that markdown file. I wanna do this. This is cool. Yeah. Jay Nathan: That's what I'm doing now. Yes. It's so much better. It's so much better. And by the way, you know how much like if I could so if you could connect your there is a way to connect your Apple Health data into this agent. It it involves a little bit of a manual step, but like with a tool with a open data source like that, then this this thing can get even more powerful. So yes, I will send you this. I'm excited to see what you do with it. And I'll tell Jack, you'll be excited that you're using it. Jeff Breunsbach: Okay. Yeah. Jay Nathan: So he actually also created a fitness app and he deployed an instance of it for me. And there's an MCP server that the that the agent knows how to talk to and it just goes and logs all this stuff, all my workouts, everything. I'm like it's it's incredible. So like yeah, there there's never gonna be a day again where I'm entering in the food that I just ate into a stupid, you know, my I'm not gonna name anybody, but like it's not gonna happen, right? Jeff Breunsbach: Yeah, that's cool. That's cool. Jay Nathan: It it's just gonna be I'm chatting with it and it's gonna pick it up on its own. Jeff Breunsbach: I know we're already about time, but I I'm gonna keep this going for like two more minutes. But have you heard of Cal AI? Did you ever hear about this school? So it's like the 18-year-old kid built built this and like they're and they had this little team and he like going to class and all this stuff. I think they sold it last year. Maybe like end of last year. And and in my mind right now, I'm like, what what timing like Jay Nathan: yes, yeah, yes. Yep. Yes. Jeff Breunsbach: What glorious timing for them because they sold it it basically what you do is you take a picture and then it would analyze the picture and basically say, I ate, you know, chicken. Here's seven it looks like seven ounces of chicken, looks like this, whatever. And then it would just like basically come up and say, okay, here's like the here's what you ate. J you know, do you want to change anything? No, okay, log it. But now in my mind I'm like, cool that's cool. But like now I'm like, they couldn't I don't think they'd be able to sell that business for that much. I think they sold it for fifty or a hundred million dollars. Thirty million Jay Nathan: Thirty. I think it was thirty to my fitness pal. Yeah, I think. Jeff Breunsbach: Okay. Yes. And so now my mind kinda like I don't think they'd be able to sell it now because like I feel like you said, I feel like all these tools are available. and so like you know, what a great what what great timing for those kids to be able to sell that. Jay Nathan: Okay, do you know the rest of the story about that? They sold it for thirty million dollars and then subsequently it got hacked. Go go read about that. And I think they stole all the data that was in there. I th I it was it was pretty catastrophic actually. And it was based on it was based on some like vibe coded like security piece around it. But I think it got I think it got t like hacked pretty bad. Jeff Breunsbach: No. No. no. Man, three point two million users, yes it did. Jay Nathan: Yeah. Yeah, how crazy is that? But yeah, I'm talking about a moment in time. 18-year-old, set for life, and now he can go do whatever he wants, sort of wrap it up on that 30 million. I think he owned most of it too. Pretty impressive. Well, hopefully there wasn't a routine tied to all that because he might have lost it with a hack. All right. Go go get it done. Talk soon. Bye. Jeff Breunsbach: Yeah. Sold it. Yeah. Man. man. all right. Yeah. Yeah, exactly. all right, my wife just texted me. I have to go help her or something. All right. Alright, we'll see ya. Thanks.