Latest / The Chief Customer Officer Podcast / EP039 AI Is the Next Excel: Internal Builders, Token Economics & the CS Team of 2030
Transcript
- Jeff Breunsbach: All right. Welcome back to another episode of the Chief Customer Officer Podcast. This is coming to you on Thursday, October twenty ninth. Jay, how's it going? Jay Nathan: Nope. yeah. I'm sorry. It will be the twenty ninth. I was like, no, it'll be the thirtieth. Sorry. it's going good. How are you? Jeff Breunsbach: I'm doing good. Are you I I kind of have two questions. I have a real one and a fake one or softball. The softball question is Jay Nathan: Okay. Warm me up. Jeff Breunsbach: now that now that your kids are are older, do you are do you still dress up for Halloween? Are you are you are you out of the phase of dress up? Jay Nathan: I've been out of the phase for dress up for a while. I it's so I'm lame. I used to have this gorilla suit that I wore every year and then all the kids in Jeff Breunsbach: Just Jay Nathan: the neighborhood would be like, It's a when's a gorilla coming out? Like when when's he coming? And I used to chase the kids around. And of course we live in Charleston where it's nine hundred degrees on Halloween. And so I'd be sweating inside of this suit, you know, for most of the night, but it was a lot of fun. But I haven't done that in a while. Maybe I should go to the attic and see if we still have the the old gorilla suit. How about you? Jeff Breunsbach: I I'm in the other I'm in the opposite throws. I'm in the my my wife's very much in the while we can keep the kids in like family costumes, we're gonna do the family costume thing. So Jay Nathan: okay. Jeff Breunsbach: so we are gonna be Paw Patrol this year. My four-year-old's like favorite show right now is Paw Patrol. So Jay Nathan: Okay. Jeff Breunsbach: I'm Ryder. my four-year-old's gonna be Chase, who's the police officer. my two-year-old is gonna be the bulldozer rebel, which kinda is a fitting description for him. our Jay Nathan: Well, Elliot. Jeff Breunsbach: Our newborn is gonna be Chicoletta, which is like a little baby chick that the mayor carries around. And then my wife is gonna be Sky, which is a helicopter flying dog. So so yeah, we're we're you Jay Nathan: Impressive. Impressive. Jeff Breunsbach: know looking forward. And my favorite part is like we're all gonna we're gonna wear these costumes for all of 30 minutes, you know, an hour maybe, before like some of us have to go home to bed and others. But all right, that was my softball. Jay Nathan: All right. Jeff Breunsbach: a very quick I just want your take on this. I just read from a prominent VC kind of PE firm, he had a a take that I thought you'd find interesting. He said, over time, he believes that a software engineer who has AI is still going to be better than somebody who has domain experience with AI. So for instance, like he thinks over time that a software engineer who's got who's like prolific is going to be better. Let's just take accounting software, for instance. is going to be lightning speeds better at building accounting software versus an accountant who can pick up AI tools like Claude and build their own software. And I'm curious, Jay Nathan: Mm. Jeff Breunsbach: do you think the same thing, I guess, could happen? Jay Nathan: It depends on what your goal is. And it depends on the what sandbox you're playing in is. If you're building accounting software for thousands of people to use, then yeah, I would agree that the engineer is going to be better. If if if you're going to build it for one company to use, probably the engineer is better. I don't know why you would do that. But I think it depends on there there's some questions I have about that, right? About the context of that. I do think over time a lot of the engineering components get automated by AI or built in by AI. And what I mean by that is the security, the scalability. I mean, we even are doing this today, right? That you you have to know the right questions to ask today to make sure your system is really secure and it's scalable and it's extensible and not a mess behind the scenes. But if you know to ask those questions, you don't need to be a great engineer today. You just need to know how to and ensure that you need you need to ensure that your agents are accounting for those those non-functional concerns in the technology and probably using additional agents to cross check what your primary agent is building, if that makes sense. Testing, Jeff Breunsbach: Yeah. Jay Nathan: penetration testing, scalability testing. And it's all so I get I think all that gets built into platforms over time. And the need for the engineer to build the accounting software becomes less and less because you'll be able to build functional scalable systems based on the domain knowledge. So I guess I'm disagreeing mostly over time with with this perspective. Jeff Breunsbach: Do you and do you think over time do you think over time that businesses that can afford to are going to have internal software teams? Like am I going to am I going to see a Fortune 500 company that basically has a a pro a product manager who oversees all the internal tooling, who's got a a a development team alongside of them and basically, okay, now instead of going to buy all this stuff, I'm actually just going to to customize, configure it in-house with like this in-house development team. And like you said, like g give them I guess give it the the shine that it needs, in terms of, you know, security, scalability, all these things that you might, you know, want to pressure test. But I'm curious if you think like d do companies go as far as like having an internal team that handles this like twenty four seven, that's their that's their daily job. Jay Nathan: I we're already seeing it. And I and I think those teams and I so we w actually work in this area, right? A lot of non software companies working with their quote unquote product teams, right? Which are developing internal applications or solutions for their customers and their internal teams to work together and collaborate on however they go to market, right? Is really what these systems do. and they think of themselves as product people. They actually take inspiration from it. Now they're they're They're not like the kind of product person that you'd see at Instagram or Pinterest, right? But they're the kind of product person they they they're aspirational in their desire to become a product person, to find value in what they're building and to deliver value, not just like when I started, I was part of an internal IT shop. We were doing the same thing, but we were developing tools or deploying tools that we bought for the line of businesses that we supported, the lines of business we supported. And we didn't think of ourselves as product people. We thought of ourselves as IT project managers and program managers. And you know, we we engineered the solutions. But so I think long and the short of it is yes, absolutely. I think a lot of ops will will get that capability, ops teams, biz ops teams, because they're the ones that are actually building the tools to both run and measure and monitor the business in real time. And they're closest to those operations in the in the processes that are in place. So I think it ends up becoming an extension of operations in some way. I don't know that it'll all be centralized. The central teams will in enterprise, non-tech enterprise, they will become more and more like your DevOps team or your platform teams that put the right structures in place so that the lines of business can innovate. That's my perspective on it, and you already see it happening. Jeff Breunsbach: Yeah. I think back to a couple companies ago, a place I worked, we had a pretty substantial kind of chief of staff, a chief of staff model where like the chief of staff had a bunch of analysts that would actually be assigned to different teams. And then every month, every quarter, right, it was like a there was a backlog, a groomed list of products and that analyst would come in to help that team develop something. And so I kind of think about like, do do companies start coupling that with like like you said, like a there's like an internal IT or internal development team that is making it easier to build applications, making it making sure it's secure, safe, right? They're not maybe building the applications themselves, but they are building the infrastructure for the applications. And then do you have this kind of chief of staff team that goes out to these departments and basically is like, hey, I'm here as your resource to basically figure out how can I help you translate what you need into these products, into these tools for your team Jay Nathan: Yeah. Jeff Breunsbach: and act as that liaison and maybe that reports into a COO or something. But I I feel like that to me starts to become like a a real sense of reality of like, okay, great. We've we've secured how these things are going to be built. We've kind of we've kind of given people almost like a bowling lane with bumpers that's like, all right, we're all going in the same direction. We're all going to make sure they're safe. They're launched on Vercel. They're launched on whatever infrastructure we have. but then, you know, we're going to have People that go help to deploy these things versus just waiting for those teams to like, yeah, like I need this, let me go build it. Like kind of going to force the the hand, I would imagine. Jay Nathan: Well, th think about it, it's like internal for deployed engineers, right? Because they're actually helping deploy the solution that the line of business needs. And then Jeff Breunsbach: Yeah. Jay Nathan: they're then they're saying, Okay, well, what are the platform components that we need to bring back and make sure everybody has access to once we get in and learn what we have to do to deploy this, right? The security constraints, the permissioning, the what data sets need to be available. What do we need to put in our data warehouse versus just access directly from a system of record? Like that I think that Yes, we already see that model happening today. It you're you're Jeff Breunsbach: Yeah, makes sense. Jay Nathan: chief of staff. It's almost it's also a little bit like a PMO. you and I have had wor Jeff Breunsbach: Yeah. Jay Nathan: worked in companies where there's PMO and same thing, right? You don't have, rock star project managers in every part of the business. You might have a handful of those people and they're like a tiger team. They go around and you loan them out to the different lines of business for different projects to bring that expertise and that you know, skill set to whatever the initiative is, and then they move on to the next thing. So that I I think that's very realistic. We we and we see it happening today. I tell you what, I've got a couple of conversations coming up with CIOs because I'm really interested in the buyer side of all the AI stuff. we're we're gonna record those for Bellbella so we can we can share them here as well, I think. But but that's a that's a good question that I'll dig into with with some of those folks as well. Jeff Breunsbach: yeah, I I still think that there's largely even in even in, my company right now that's growing fast and rapidly, I I think we're realizing how maybe how much you need to put in place in terms of infrastructure, to make sure that people can move these things fast. Like it actually takes a bunch of time and energy to set those things up. but once they are set up, again, like we have people we have BDRs, we have BDRs standing up, you know, tooling. We have Jay Nathan: Yeah. Jeff Breunsbach: we have our our kind of an operations person standing up tooling myself. you've kind of got people across the business now that can easily stand up these applications. I built a whole, I forgot if I told you this, but I I just did my whole segmentation model for my customers inside of an application that I built in Vercell. And it's, you know, it's connected into our HubSpot and pulls in data every single day and refreshes and pushes and syncs back and forth. But it allows me to to look at the settings, change certain things and basically come back with a a way that I want to segment the customers and make sure that and actually I I think there's a lot more because I remember again back in the day we used to do this, you know, in Excel sheets by hand and it felt very kind of cumbersome. But now it's it's so fast, right? Like I actually can I can change the model and move it almost to the detriment, right? I can I can almost like move it too much where all of sudden now, you know, does C S1 mean anything or CS4 mean anything? So I'm not trying to do that, but Like the other day, somebody made a comment like, this type of customer that we serve actually, we think should automatically be a tier one because that is our main t our main go to market motion as we go in the future. And it's like, you're right, we should do that. And I went and made a change and boom, all of them are now tier one. And that was like deployed and very easy for us to go see like in a quick dashboard. So, anyways, I thought that was like another fun little application to build. Jay Nathan: Yeah, that's really cool, man. and but that's a great example too of what we're talking about. Like you your loaned out for deployed engineer would never like why should they have to go build that when you could do it in the flow of your work. We we you probably used to delegate, you know, processing big sets of data to people who had access to the tools to do that. And then Excel came along. Now everybody uses Excel. Right? AI is just sort of the next I'm I'm telling you, it's the next version of Excel. I don't quote me on that or don't clip it, but it's evolution wise, it's the same thing. We're getting access to a tool set in the business that has typically l been limited to the IT organization or the product development organization and it's not not you know, it's not a constraint anymore. We just said to your point, you have to put the guardrails around it and then let people loose. In fact, you know, one of the conversations we've been having this week is how do we just get everybody on my team clawed code and get them using clawed code and not just Claude and Claude Cowork. or, whatever tools we're using, everybody needs to be writing software now. In especially in our world, but, in general I think why not? Jeff Breunsbach: And what's the w I guess what was the determination for you guys of getting everyone cloud code? Is it again, mainly just trying to almost like what's the what are the stepping stones in order to get people a kind of over the hump of like, hey, this isn't as scary as you think it is once you start talking about code? Jay Nathan: Yeah. Well, we haven't actually done it yet. We've we've given it out to a few people and the ones we've given it to have done amazing things with it. And so now I'm just I'm wondering like what's wrapped up in the minds of my team that's not coming out because they have no real outlet for it. We haven't r unleashed that creativity yet. And that creativity is really important in our business, in any business really, right? So, Jeff Breunsbach: This is this is something that I I've been going down the wormhole on too. And so again, I I think like the maybe the approach that I've had is consistently asking like what's something that you do every single day, or like what's something that comes up multiple times a week, what's something that is just what's something that you like are are thinking about that is just kind of like always there. And if like we document those things, I almost guarantee you that there's a way to for us to be to think about this process or procedure or whatever, with just like AI in mind. and and another example is a push from our CEO. He kind of has we have a a weekly meeting with our executive team and he has kind of founder notes that he puts in there and it's kind of I'll call them like proverbs, right? They're kind of ephemeral where it's like, hey, we should, you know, he's trying to to get us thinking about okay, what's next and what's on the horizon and how are we really pushing the business. And I thought one of the things that he put in there yesterday, which I walked away with and was like just it it resonated very much with me was basically are we thinking about this as the next iteration of something or are we reimagining how this thing works? And the problem is that we often think about, okay, what's the next iteration of this? hey, this process looks like this today. Let me just go insert something else. Let me go insert a different button that I press and then it just automates it, right? But what if I didn't actually have the process to begin with? Do we need the process? Right. do we are we actually reimagining this stuff with AI at to like to a degree that I don't know that that really warrants us actually pushing the business to where it needs to go? And so it's another, I don't know, reminder to me of like thinking a bit bigger, thinking a bit more also of like, especially now, I think more than ever, you can actually reimagine things versus just the next iteration. I think it was harder before, like to your point, For s for an individual contributor to really remove a process or or dramatically change a process, you had to go figure out so many different things. And now I literally think it is like, you know, you can go to Claude and basically say, Hey, here's the process I do today. If we were to rewrite this entire process, or if this process was to eliminate as many human steps as possible, what does this look like? and like to me, that is like the place you want to be living in. Jay Nathan: It it's first principles thinking, right? all right, what are we trying to do here again? It's like starting from scratch. But I think the caution there is you can't start from scratch with your teams every day, right? Jeff Breunsbach: Yeah. Jay Nathan: 'Cause that gets exhausting. And some people like you and I, I think are wired in a way that I could start from scratch every day. I do start from scratch every day, right? I'm like, my gosh, I don't like how that you've seen me iterate on presentations and segmentation models and everything. It's like, Jeff, we put all this time into this thing and I hate it. Right. And I want to start from scratch and I'm going to do it again. Right. I I've done that, I can't tell you how many times in my career. I'm sure you've lost me out. But for for people like us, I think that works. But for I I actually think a majority of the population of the working world that doesn't work. They want some semblance of consistency. And so I think it's being We have to be thoughtful about how we roll that stuff out too. Now, and it sort of gets into the adoption. Now we we were talking before the the call about sort of how do you get people to adopt the stuff that you're building, which is a near and dear conversation to my heart, but it's it's all it's all it's all part of it, right? It's all part of y you've got you've gotta we still have to handle the human side of all of this. Jeff Breunsbach: Yeah. Like the this isn't fully I think almost in some cases people think like we're jumping to like agent to agent communication like tomorrow. And again, I I still think back to the chart of like the hundred dots and like AI is really like two dots on the hundred or so, you know, so far. but yeah, and I I think that is like one thing, and I'm I'm curious how maybe like how you how do you think about this with your customers? Because I think this is I think this might be a challenge you have with your customers that I'm I'm having a challenge with some of our internal teams, which is again, the possibilities now are almost endless. And so I find myself walking in and saying things like, we can configure this however we want to, we can customize this however we want to. And again, I think to you and I, right? Like I'm wired, like, that is freeing. That's cool. We can we I can think of so many Jay Nathan: Yeah. Yeah. Jeff Breunsbach: things, right? But to your point, I think a lot of times people look at that and say, Hey, I'm just trying to get my job done. And I you know, I I I don't have the mental capacity or I don't want to, or maybe it's not intriguing enough to like. help you think about all these things that need to fit into this software or platform. And so I'm curious because I I think this is a customer constraint too, right? Like oftentimes as CSMs, we bring software to customers and we say, hey, you can configure this. And then it's like, well, what are your what's your best practices is the what the customer says back to you. Or like they say, hey, what do yeah, what what emotion Jay Nathan: Right. They don't know. Yeah. Jeff Breunsbach: what do most of your customers do? And we'll we'll probably just follow that. Right. And so I'm curious if there's a way, you know, like you said, you guys are building some AI solutions like How are you bringing some of this I guess to bear for customers where like I I guess like you don't have to get them thinking, okay, I have I have too many ideas that I can't even start. I can't even write anything down. what's the approach been for Jay Nathan: That's paralyzing. Jeff Breunsbach: you guys? Jay Nathan: So the approach for us is real-time prototyping, which you can totally do now, right? I mean, Jeff Breunsbach: Okay. Yeah. Jay Nathan: so think about you know going into a conversation to to with with a here's how I would structure it, right? Hey, we're we're building an agent for let's call it customer success. That's an easy one. We're building a customer success agent. We've got a starting point, here's what it looks like. Let's talk about. What we're gonna do, what we need to do differently, different than what this thing is showing us, right? And you know, how do we manage customers? How do we segment them? What are the kind of communication cadences? How do we think about health? You know, hone in on a couple of ideas that are unique and bespoke to like real differentiators for how we do things in our company or in this company, and then literally come back with a prototype an hour later that embodies it in something technical because. Telling somebody that they can configure anything something any way they want is great, but they don't that does not resonate until they see it happen. Jeff Breunsbach: Yeah. Jay Nathan: Right. And what we're finding is that if you're gonna sell agents, if you're gonna buy agents, if you're going to implement agents, they have to have something tangible you can see associated with them so that you know what it is you're going to get and you can react to something. We had a a demo or a I should say a a readout with one of our largest clients. last Friday and we proposed building out several agents with them for their line of business in addition to doing some of the pendo work that we're doing. And we built a prototype it and it did two th just a wireframe prototype. We put them in slides. We didn't even click through them, although you could, but it did two things. For us going through the exercise of building a prototype for this particular part of the business made it real for us. more so than it had been. And Jeff Breunsbach: Yeah. Jay Nathan: they looked at it and she was like, wow. Yeah, we absolutely need to do this. And no, it's not the area that you prototyped over here. It's this other area that I want to do. So putting that picture in our stakeholder's head allowed her to iterate with us faster. Now she's sort of like us. Like she's first principles thinker. Like she's really moving quickly. But that that's what I would say is like we have the ability to rapidly prototype now. We should actually do it. Jeff Breunsbach: Yeah. yeah. Jay Nathan: In very robust ways. Jeff Breunsbach: I I think that's actually been my approach as well. you either using claw design or just clawed itself. and then again, kind of taking the platform, giving it the concept or the idea, talking through some of the workflows, like and basically just, hey, I need a mock up that I can go show people. and it's become very good. And to your point, like having that, being able to like walk through that has been pretty profound to like walk into meetings with where people say, Okay, like now I have an understanding of like what at least Maybe more of an understanding of what's possible, than they did before they walked in the meeting. Jay Nathan: Let me show you something real quick while we're talking about this. And people who are listening will not get the benefit of seeing what I'm showing you. But we have a agent design sprint. And the way we talk about this, because yeah, you might want to have an agent that's got its own custom interface. So you're gonna build a complete new app, net new, that teams are gonna use, they're gonna it's deployed on Purcell or deployed on Azure or deployed on AWS, whatever, right? But it's A new application that you're building from scratch. Although it might not be, right? It might be you're interacting with this agent in Slack or in your email. Okay. And so we can mock that up. This is the same agent, by the way, I'm showing you. Or maybe you have Service Now. Here's a mock up of the same agent in the ServiceNow interface. Maybe you're running Salesforce. Here's Salesforce. you want you want HubSpot? Great. Here's a mock-up of in HubSpot. So like it did. You can flip between these things effortlessly, right? And you can even mock them up in just a pure HTML prototype to put the idea in people's head and then mold that. and you know, can do it quickly, you can do it inexpensively. It's basically free to do. That's the way we should be designing workflows for our team, not just software, right? But like the whole workflow. Like that's where you start and then you build the interface is just a manifestation of it. So you you got me going on a on a topic that's near and dear. So Jeff Breunsbach: there's a co there's something in there that I just I'm gonna take away and go do myself, which I liked, which is Jay Nathan: good. All right, cool. Jeff Breunsbach: one aspect of that is just like what's happening almost like the notification system, right? what's happening that you don't know or what's happening behind the scenes. because like I still find that the CSM wants to know those things, right? Okay. Even if it if it changed X, Y, and Z feature or like setting or kind of field. they still want to know it, even if they don't have to approve it or whatnot, right? there's still there Jay Nathan: Yeah. Jeff Breunsbach: still is power in them having that context of why did it happen? Is it right? Is it accurate? So anyways. Jay Nathan: That's the big learning. That's the big learning for us, Jeff, is that we all talk a big game around agents, but agents are not the user interface cannot be clawed for an agent. Cause Jeff Breunsbach: Yeah. Jay Nathan: the and here's here's my thesis on that, and it gets back to your question. The mental leap you have to make between a chat interface and a structured, durable process that needs to be followed fifteen times a week, that's a big leap. Right. And Jeff Breunsbach: Yeah. Jay Nathan: the reason that structured interfaces exist is to basically guide the user or guide the the business person down the path of the thing that they that their judgment needs to be interacted upon. Right. Not the right words, but you know what I'm saying. So Jeff Breunsbach: Yeah. Yep. Jay Nathan: th then, like we were talking about before we started recording, if you want your team to do things a certain way, if you want to implement change in your business, then You could talk about it. You can do presentations. You can, you can try to train people and enable them on what the quote unquote new process is. But the way we've always implemented processes that stick is how we deploy them into the systems that those people use every day. So the Jeff Breunsbach: Yeah. Jay Nathan: workflow sits in Salesforce now and it's got the right choices and the right list. Like same thing still applies. The technology is just a little bit more pliable now and we can deploy those workflows a little bit faster and in whatever different user interface is where the where the user actually lives day to day. Jeff Breunsbach: Yeah. Jay Nathan: So I think AI built applications like you're building for your team are the way to change execution and culture and process and everything. So Jeff Breunsbach: Yeah. well I know we've got about five, six minutes. What's what's the what's the last topic you want to hit on? I know there's been Anthropics S one, which is like, you know, old news at this point and hasn't too many hot takes. Jay Nathan: Yeah, I think everybody everybody's hot take that one out. So it's interesting. I I think it's interesting that they they don't they didn't have gross margin in their in in their S one. So what does that Jeff Breunsbach: Interesting. Jay Nathan: mean? but yeah, I don't I don't have any hot takes on it. I'm just I'm sort of tired of hearing about it, to be honest with you. We'll let everybody else hot take that one. Jeff Breunsbach: Yeah, I'm with you. what else? Anything else that you guys are building or seeing, conversations you're having that you've like found interesting as of late? Jay Nathan: I t two funny things. I I am with a company this week who I'll I'll just I'll leave them nameless, but we were talking to one of their leaders and he was joking last night because he got an invoice from finance and one of the people on his team had racked up a thirty thousand dollar bill on Claude in a month. So, last week you may have seen Accenture is starting to clamp down on their hundred and ten thousand employees using Claude. They're just going down to like the base Claude subscription for twenty Jeff Breunsbach: Gosh. Jay Nathan: nine dollars a month and not letting them spend, you know, thousands and thousands of extra dollars every month. I think more and more we're gonna see teams as we go into, you know, twenty twenty fiscal year twenty twenty eight or, you know, twenty twenty seven, I think you're gonna see teams that are starting to rein this in a little bit. I don't know what it'll mean for anthropic and open AI. I'm sure they're still gonna do fine, right? They're still gonna grow because there's so much demand for intelligence. Jeff Breunsbach: Yeah. Jay Nathan: But I just thought it's funny, right? And this wasn't an engineer. It wasn't a ops person. It was just, you know, some mid-level manager just cranking it out. Jeff Breunsbach: well I guess in in that case too, do you think it starts to become more likely that you it's almost like you need to I guess like the old school term is like come with a business case. Like do you think that there's now they're they're gonna start to like lock these things down more and more to a point where all of a sudden it's like, okay, I'm gonna have to come present a business case about, you know, how big is this project? How how do I what you know, what are the the guardrail? What are the guardrails I'm gonna put in place? What models am I gonna try and use? How am I gonna try and execute token usage? there's probably gotta be some version of that that's coming soon. Jay Nathan: Yeah, probably. And the cost of this stuff will continue to go down too. But one of the things I was thinking about over the weekend was you you know, we have sixteen people on our anthropic subscription. And I can guarantee you, like even right now, like my usage for this I use it all morning, but I guarantee you my usage is at like two or three percent of my premium subscription for this five hour session or whatever it is, right? And if you look across my team, I'm sure it's the same. They're using it all the time, but we're not using it very much. So my my my I started thinking about how could we leverage all of the all of the tokens that are available to our entire account, you know, that that are just Jeff Breunsbach: Yeah. Jay Nathan: going away every five hours because our that session expires. Those the you know, the opportunity to use those tokens does not come back around. So this really speaks to the idea that, you know, as you're thinking about what you're building in in Claude or OpenAI or whatever you're building in, you need it running all the time, right? And it's maxing out what you have access to because, you know, there are there are times when you're gonna want to do more and you're not gonna be able to do more. So I don't know. I there's a deep thought. Jeff Breunsbach: It's no, I I think about this a lot. I actually before I st before I start we started recording this podcast, I opened up my my work computer for clo with Claude because I had a couple things I wanted it to do. So I just asked it to go start doing research on three things that I I know I need to go do and build in customer OS. It's actually mocking up a few things for me. And I'm not like actively checking it right now, but I just told it to get to a a stopping point, said, Hey, go build these mock ups the best way you see fit, get to the stopping point and then like wait for me. But Like to your point, I I oftentimes I I go to bed sometimes thinking about man, think there's eight hours that I'm not using this that I'm are just gone. Unless I Jay Nathan: Yeah. Yeah. Yep. Jeff Breunsbach: were to come up with a like you said, come up with a system that is like, okay, how do I how am I architecting and working on this thing overnight? So yeah, I think it's an interesting problem. The the other thing like to your point that I often think about is like is yeah, it like you know, Jay knows he's not gonna use his tokens for the next hour. Can Jay put his tokens up for like to the rest of the team, right? Like how do you start getting like shit? Like to me, it seems very much like they've basically built individual licenses into Jay Nathan: Yes. Jeff Breunsbach: a a game that is multiplayer. And you're just like, okay, at some point there's gotta be this this flip, right? like there's just gotta Jay Nathan: Yeah, yeah. Jeff Breunsbach: be so much demand or or piping out that it happens. Jay Nathan: So I've got this idea for something to build here and I just it's not high enough on my priority list to do it. but my idea is Basically build some kind of centralized MCP server that stores jobs in anybody's clo and then there's a cowork job running on everybody's computer and it's constantly running and picking up jobs off the list Jeff Breunsbach: Got it. Jay Nathan: and just running them. Now you have to have the right access and to all the data sets and all that kind of stuff, right? there's there's some permissioning and qu questions and authorization questions that that you have to answer to do that. But I don't know, chip chip. Told me it wasn't possible. I think it's totally possible. So Jeff Breunsbach: I don't know, Chip, I think Chip Chip sometimes is the realist, He's the one who's like, yeah, yeah, man, you're like, but Jay Nathan: Yeah. Jeff Breunsbach: but no, I like I think about that like often. especially 'cause like there's been times when I've run out of tokens and I'm like, like it, you know, a five hour session and I'm out like and it's like a Friday, Jay Nathan: I know. Jeff Breunsbach: it's like Friday at like three PM. It's like, cool, I'm not gonna log on at eight PM to do more stuff, you know? Like I'm not gonna wait for Jay Nathan: No, no, no, no. Jeff Breunsbach: so so yeah, I d I do think it starts to become a problem though of like This, yeah, I'm spending something and it's just going away, right? or I I don't know, can't you? It's almost like, is there a resale market, right? Can I can I take my tokens and resell them because they're they're mine, right? isn't that in theory? I guess it's probably part of the it's against the terms of use, but or terms of service, but like there should I dunno, you I can start to see a world like where I'm actually just using or I'm actually using a harness that is agnostic to. any of these models, but I have access to all of them. And then there's a marketplace where I can trade and share. okay, hey, Jay Nathan: Yeah. Jeff Breunsbach: I'm willing I'm willing to give away anthropic token tokens for Codex tokens. Like, I don't know. That Jay Nathan: Yeah. Jeff Breunsbach: that to me seems like there's like there is a world that that's going to happen, I think. Jay Nathan: That's so interesting. Yeah, yeah, yeah. I I'm with you, man. That's very interesting. But you know, the model providers have a veste vested interest in not letting that happen because, like, let's face it, they probably don't want us to maximize all of our user accounts. And I bet you if they looked across all the user accounts in their in their platform, I bet you utilization is quite low relative to Jeff Breunsbach: I was gonna say, do you think that's a risk that they think about is if if do you think there is a level that they're willing to to take a bet on that basically says because I like you said, I imagine if every account was to use a hundred percent of their usage in a five hour window, that would have to melt the world, right? there's that that would have to be a doomsday scenario. And but to the Jay Nathan: Totally. Totally. Yeah. Jeff Breunsbach: I maybe maybe to the point, like they've got enough. I mean, they have millions of subscribers now that like they've seen they have enough historical data that basically says, like you said, may maybe the top one percent of users. use their five hour window and that's it. And that allows them confidence to do it. Jay Nathan: Yeah. I think it is a risk, although nobody's gonna not have the security policy of not having an anthropic account or a an open AI account for their team, right? But I think the bigger risk in the short term is what I said earlier, where okay, anthropics, you know, they're driving tons of growth, but a lot of it is on these usage spikes that are gonna be Jeff Breunsbach: Mm. Jay Nathan: those those usage spikes are gonna get quashed, right? Like the CFO is not gonna stand for a thirty dollar $30,000 random bill coming in down Jeff Breunsbach: Yeah. Jay Nathan: from Anthropic from one user, right? So who's not even an engineer? Yeah, that's the other thing. It's like, so what I'll do is I'll Jeff Breunsbach: And you don't even and you and you don't even really know the efficacy or the output, right? you're just kinda like, Okay, we spent thirty thousand dollars for what? And you're like, I could have, done so many different things with that. Jay Nathan: add that, I'll add that $30,000 to this person's salary and and you tell me, is this person worth that to you? And are they worth that much more than they were last month when we did not have to pay? That's $360,000 a year. So now that person, let's say they make, I don't know what I have no clue, right? I'm just making it up. Let's say they made $150,000 a year. Now that person all of a sudden, if they're pulling that stunt every month, they cost you half a million dollars. Are we getting that value? So yeah, yeah. It's a very good, very good question. Jeff Breunsbach: I'm curious, like one of the other things that one of the other things that I have been, I guess, thinking about quite often too is just like I guess interfaces to begin with. Like, do I think you've you've kind of come out and said basically that you think interfaces still like need to exist. Like that you don't think everything's just gonna go headless or like everything's just gonna go to a chat interface. like I guess what get what gives you confidence for that is is it gonna go back to the point earlier of like, hey, we still need these interfaces where humans can get work done, but also you can kind of direct these workflows and like actually have kind of these things baked in. Jay Nathan: It yeah. at some point the constraint isn't is the human. It's not what the agent's doing. I could tell you right now, I have more files in my downloads folder on my computer than I've ever had. It's like my main working folder inside of my computer now, right? Jeff Breunsbach: Yeah. Jay Nathan: do you have that too? I Jeff Breunsbach: Yeah. I was actually just joking that all of a sudden I have random folders that I create, like local like random you know, like I have now I have Jay Nathan: Yeah. Jeff Breunsbach: local folders where like I'm I'm doing stuff and hosting work and and other things and like to your point I've I've I've I've kind of not really there you know, there's no strategy to it. I'm like, okay, sure, I need a new folder for this. Let me just create it. Or like, yeah, let me just download all these artifacts to this download folder. So yes, I'm chuckling 'cause I have the same problem. Jay Nathan: Yeah. I just I just counted. I have almost fourteen hundred items in my downloads folder. But think about it, these agents are doing so much work on our behalf, but but to the extent that tasks are not fully autonomous, there still has to be a way to organize the work. And not all tasks are fully autonomous. In fact, very few are fully autonomous, relatively speaking. Right? And even if they are autonomous, the other thing that, Auditability, observability of what the agents are doing is is critical. And again, that could be through an agentic interface itself, right? Not to be too inception-like about it. But yeah, I people are still accountable. The the the the thing that's emerging is the people are still accountable for the business outcome that that the agents are helping you achieve and they're responsible for what the agent produces. Jeff Breunsbach: That's yeah, I think that's a good reminder that like what it's basically you have an extra employee, right? these agents are extra employees. Jay Nathan: Yeah. Jeff Breunsbach: So if it goes like you said, if it goes off the rails and spends thirty thousand dollars, they're not going to the Jay Nathan: To you, bud. Jeff Breunsbach: they're not going to the agent's desk saying, Hey, why'd you do this? Right. They're gonna go to you and say, you didn't put the guardrails in place, you didn't put a cap on spend, why didn't you do this? Right? Like, and that's your responsibility now. Jay Nathan: Yeah. And and so then extrapolate that to, okay, we've got an agent approving insurance claims. You know, low dollar insurance claims. If you approve, you know, five hundred low dollar insurance claims that you shouldn't have, like who who takes the accountability for that? The person who built the agent to do that will have to that line of business is ultimately re accountable accountable for it. And the people in it that made the decision to build it and, you know, test it. And so, you know, these things are not It's not a panacea. And so yes, I do think you will need user interfaces. Like I said last time though, your user interface will be your user interface, I think, right? Jeff Breunsbach: Yeah. Jay Nathan: Whether whether it's in Salesforce or ServiceNow or or custom bespoke, web platform, it will it will be your interface so that you can have some, you can tailor the way that you work your your processes and your secret sauce and your business. Right. And and they'll and even even when we deploy these agents for our customers, we're bringing a starter kit, right? So you don't ever really have to start from scratch. I think there'll be more open source. Like you and I have talked about some interesting open source ideas, right? 'Cause nobody wants to really start from scratch either. But everybody wants to be able to take it where they want to go with it, ultimately. Jeff Breunsbach: one other thing I'm curious about too, just in your travels of the world right now. I was I was talking to two different friends. One works in Fortune 500, or they both work in Fortune 500 companies. One is in manufacturing, one is in banking. And actually you can you can throw a third in there. you can throw a third who is in I would call it like like manufacturing services and like building services Jason. But they're Their usage or the way that that I talk about Claude and Chat GPT and Codex and is so wild for them because they're just like, we don't even have some of the stuff. We we basically have very locked down environments where they can use a Gemini maybe for a couple of things. And I'm just curious if like again, it's still kind of is one of those moments for me where I was talking to them over the weekend at like a party we had, and I was just like, man, like a reminder of like, wow, like the adoption on these things still is like so myopic to where we are and the the people that we interact with all the time. And I'm curious if you still see some of that still happening where you're just like, you know, man, like these companies are so behind the times. And it just I don't know. I'm curious if you like still see that. I I guess I walked away from that conversation thinking, again, maybe people now that are exploring with AI tools are using them, are becoming extremely proficient, almost have like a leg up in the next 10 years of hiring and job cycles. And if you if I think if you've got this skill set of hey, I've built applications before, I've used, any of these tools to codify processes and automate things, like it kind of feels like you're gonna have a leg up in the job market of the next 10 years because again, I I still think there's a large faction out there that are like at these companies that are legacy and are just kind of like, again, I'm just gonna keep doing my job. Right. Like that's the the general sentiment that I get is like, hey, it's Yeah, sure, there's endless possibilities, but I'm just doing my job because my my company hasn't mandated has not mandated it to me and they've also not opened up access to me. And so therefore, I'm not gonna go personally explore it myself. what's the advantage if I just it's almost like adding more work on my plate. So I'm just gonna keep doing my job. Jay Nathan: I think there maybe there's some of that, but even for the people who are interested in it. So I in really interesting timing. I talked to somebody this morning who works in a manufacturing business and works actually in the plant. You probably know who I'm talking about. And this person told me that they're getting ready to deploy this AI solution on on a on a series on one of their lines, basically that's gonna be able to give them tons of information. Diagnostically about what's going on with the system. They're gonna train it over time to get better and better at at repairing and keeping Jeff Breunsbach: That's cool. Jay Nathan: things up and running. But then you know what he said to me? He's like, the the thing we're trying to figure out before we deploy all this though is how are we gonna keep all of the knowledge about our process out of the hands of OpenAI? So it's a it's an intelligence sovereignty question, more so than it is a my job question. They're trying to figure out how to keep their trade secrets, trade secrets without handing them over to the model providers because that is the other thing that's happening. They don't want their prompts going over to them. They don't want to give away the secret Jeff Breunsbach: Yeah. Jay Nathan: sauce on the data that's coming out of their machines, none of it, because they know they can interpret it and figure it out and maybe turn around and sell that information to other people. That is a very real thing. And so I think you'll can anyway, the downstream impact of that is they will continue to. Press into open weights, open source models, Jeff Breunsbach: Yeah. Jay Nathan: and running models in their own environments, which is why, and you see, like Nvidia is really leaning into this. They bought Hugging Face, they are pushing, they're they're actively pushing the open weights models and supporting the open weights community. They're complimenting or they're they're they're commoditizing their complement, which is the actual model, right? Jeff Breunsbach: Interesting. Yeah. But and to your I guess to your point, right? NVIDIA NVIDIA almost like needs to serve both sides of the market, right? they need to help they need Jay Nathan: yeah. Jeff Breunsbach: to serve the, the the models themselves, but then also the pushing the open weights probably helps some of their top customers who want to deploy certain things themselves, right? Deploy their own servers and systems and training. Like that's interesting. Jay Nathan: Well, their top customers, no doubt, are there's three or four of right? They're Jeff Breunsbach: Yeah. Jay Nathan: spending billions. But but even still, I mean, look at I mean NVIDIA is so much more than than just a chip maker now. They they actually host services and provide all kinds of cloud services for for enterprises now as well. So yeah. Anyway, but to your question, I think I I think it's less about the I just don't wanna do it and more about okay, is this what's the security of this for our IP and therefore I don't have the ability to just tinker like we do in some other industries. Jeff Breunsbach: Interesting. Well, that's cool. all right. Well, maybe a good stopping point for now. Jay Nathan: Yeah, this has been good. Jeff Breunsbach: Yeah, I like to you know, coup a little bit of a ramble, but I think we we made it to topics that we we cared about and that were interesting to us, so Jay Nathan: Well, hey, can I can I throw one more topic in real quick? Jeff Breunsbach: Yeah. Jay Nathan: We'll do it like a lightning round because we only got a few minutes here. I've got a facilitate a session today with some Jeff Breunsbach: Okay. Jay Nathan: customer, customer executives. And the session is: what is the customer success team of 2030 look like? So fast forward three years. What what is the customer success team of 2030? Here are the questions that I'm gonna ask. I want you to tell me. I wanted you to tell me what what you like, what you hate, and one thing you would change. Okay. So question number one: what part of your CS org won't exist in 2030? What what part of today's customer success org that you run will not exist in 2030, two and a half years from now? Number two, do you have forward and employed forward-deployed engineers today and how do they fit into customer success? Is it a permanent role or a bridge to adoption for AI? Number three. Where do you have where in your business do you have agents doing real work today and where have they flopped or missed expectations? Number four, how should what the customer success team is hearing on the front lines shape what's actually getting built in the built in the product? So customer led versus sales led concept. And then what's your big bet for twenty twenty seven and your team? What do you love? What do you hate and what's one? One question you'd change. Jeff Breunsbach: I love the number three. Like what are they where are the agents doing real work today? because I think it'll be interesting to see. I think in the room you're probably gonna get like a a dispersion curve of like leaders, laggards, and then probably like a bunch in between. And I think the leaders like, you know, autumn I've got agents doing X, Y, and Z. You probably got the middle that are saying like we've we've got an agent doing one thing, maybe, and then like the laggards, they're like, We you know, we haven't even explored it. so I think it'll be interesting to see that that question play out. I think the I'll I'll just say the one I hate and I hate the words hate is strong, but I think forward deployed engineers just to me seems like so misused as a word. it just feels like it's just been it's just been thrown around because everyone just assumes this is what a forward deployed engineer does. it's kind of like saying here's what a customer success manager does, right? It's just like every company has a different different definition. Jay Nathan: Yeah. It's lost its meaning. Jeff Breunsbach: yes. And then I think one thing that I would I I don't know if I'd change it, but one thing I think is pretty interesting is on the how should what CS here shape what gets built question. our head of engineering just posted in our executive channel yesterday that he he or two days ago that one of the things he's working on right now is to try and allow any non-engineer to change code. And so he's like that that's like something he wants to do is enable that before the end of the year. And again, I think he's not just saying, let me go open up access to everyone and give them a GitHub account and blah blah blah. Right. But he's I think he's what he's thinking about and like a very real example is like, our website is hosted by our website is hosted in GitHub and it's like built by ourselves. We've just custom built it. And like a real example was like, our our some of our people team just went and changed something on there yesterday. And he was like, Look, real example of like somebody merging code, like they they put up PRs, they know what to Jay Nathan: Wow. Jeff Breunsbach: do. Jay Nathan: Yeah. Jeff Breunsbach: But and so now I think he's trying to think about how do you take that how do you take that example further and further and further, right? To where I I gave him him an example of I feel like I could go change this one thing Jay Nathan: Yeah. Jeff Breunsbach: that customers have wanted in our platform. I could go change that in the code. and so like I think we're gonna try and figure out how to, you know, how do how do we safely do that? The guardrails, PRs, all the things. But I think that's just an interesting question to me about it almost is, you know, how should what CS here is shape what gets built and then How are you gonna go enable your engineers to or your CSM to or your CS team to go build it like themselves? Jay Nathan: That's awesome. That's so cool. Just quick blurb and then we'll we'll cut it off. But I listened to a podcast yesterday on my trip where at Reddit they've actually built a they've built the tool to make prototyping for their designers very simple. So they're actually able to deploy new versions of Reddit in a sandbox on production data, like with this harness that there's no setup. They don't have to figure out how to pull this repo down or do this or that. They don't have to hold their tongue a certain way. They just log into this thing and start going. And then they can iterate on the actual Reddit site to prototype. Jeff Breunsbach: Let's go. Jay Nathan: So I think that's coming. I think it's for number four, how how should what CS here shape what gets built? It'd be really, really interesting. And it's not just CS, right? I heard the same thing. Ramp lets their sales team prototype features. I don't know how, I don't know how much that actually happens, but I mean, it think about it. you've got customers on the phone, you could actually do prototyping in real time based on the transcript for that call if you really wanted to. You know what I mean? So I think that's Jeff Breunsbach: That's cool. Yeah. yeah. Jay Nathan: coming. I think you're right. That that's that's very cool that your your CTO's already doing that. All right. Cool. Thank you for the Jeff Breunsbach: Yeah. All right. Jay Nathan: feedback. Jeff Breunsbach: You got it. I'm excited to hear what the session what comes of the session. So make sure and come back around. All right, cool. All right. Enjoy the time. We'll see ya. Jay Nathan: Yeah, I'll let you know. I'll let you know. All right, man. Talk soon. Bye.