speaker-0: Have Claude and OpenAI already won. speaker-1: OpenAI and Anthropic are trying to do to you right now with all these features. They're locking you in. speaker-0: I'm not gonna sign up for twelve months and lock myself in. speaker-1: It's never been a better time to be in consulting. speaker-0: Welcome back to another episode of the Chief Customer Officer Podcast. This is coming to you on Thursday, November 12th. Jay, how's it going? speaker-1: Hey man, I'm doing great. I'm in Atlanta, Georgia. Beautiful downtown Atlanta. Jeff, I I don't know if you knew this about Atlanta, but the nightlife, like Atlanta at night, downtown, is a really weird place to be. Why you say that? The very few restaurants, very few bars. The hotel bar, so I was with a group last night. We went to a a basketball game and we came back to the hotel that we're staying in. The hotel bar closed at nine thirty last night. They wouldn't even serve additional drinks. So in Yeah, I I actually don't drink, but it was we were gonna hang out at the bar. We ended up walking half a mile to find a bar open in Atlanta, downtown. So I I didn't realize that about downtown Atlanta, but now I do. speaker-0: So yeah. So I mean sprawling. I'm good. Yeah. We you know, just enjoyed a weekend of lovely weather here in Charleston and kids' birthday parties. I'm just in the I'm in the thralls of like you know, four year old birthday parties and I don't know, watch some football. That's pretty much it. Yeah. speaker-1: pro or pro or or college? Your Bills played last night, right? speaker-0: They did, yeah. it was a good game. I stayed stayed up way too late. Yeah. they ended up beating the Vikings and the Vikings have like the best defense right now. So it was a fun kind of matchup 'cause we have the best offense in some respects. So so yeah, it was a good game. Just you know, my gripe is that like they start these games at eight fifteen, eight thirty, and I was up until eleven fifteen, eleven thirty watching this thing, you know. And I'm in the I'm I'm in my in my pajamas, like waiting for the game to end just so I can press, you know, off on the T V and just like fall right to sleep. But yeah, it's good. one thing one thing I'm working on right now that would be like fun to chat through is we're doing a a bunch of account transitions. So I've hired hired new CSMs. and so we're just going through so I I guess a couple of things that I've been learning throughout this process. One is man, segmentation now having tools like Claude and hooking it up to ChatGPT or hooking it up to our HubSpot and even like Fathom Data and other stuff. I just remember how you know you used to struggle. To get data points that you could maybe use in your segmentation. and even things like the complexity of the customer, maybe, you know, or how difficult they might be. I can actually go look, okay, how were they in the sales process? You know, were they kind of fighting with our sales rep about a bunch stuff? Were they, you know, were they very agreeable? Were they trying to find So, anyways, I can find a bunch of stuff which has been interesting. And so, so that's been a fun exercise to go through now. Is and I've actually got that hooked up into customer OS, which is like this the customer success tool that I'm building for our team. And so and it actually runs our segmentation on deals that are in the sales pipeline. So I can start to see, okay, where am I going to segment these customers and you know do it in kind of the right way. So so that's been fun. I think and pretty interesting. and I would say the second thing that's been interesting come out of this is that you know we're shifting around a bunch of accounts and sometimes it is just like a great opportunity to drum up Some upsell opportunities and some some new business. and you know, I think our team is just kind of right now walking into a lot of conversations with the idea of just learning about this customer, right? Hey, this is a new customer for me. Let me ask questions, let me just go in and learn. And I think we're finding two things. One is I think something that I've always said maybe since I've been in software, which is that we don't tell our customers enough about like core features or core core uses that they should be like d you know. using of our platform and why it matters to them. Again, I think we kind of tend to tell customers one time in an email and then we expect that they're going to remember that. and so we're finding a lot of, hey, you know, you didn't know about X, Y, and Z, you can do that now. Or hey, you should be using this feature. We rolled it out a couple months ago. and so like going through and doing that type of inspection, I think is like one customer journey exercise that I'm like building my team around right now, which is like this idea of a feature matrix. actually I got this idea from ramp. So I saw Ramp did this with our finance leader. They basically showed a bunch of features on the slide. And it was very simple of red, yellow, green. Basically, are you using it? None, a little bit, or a lot. And it just had, you know, I think it maybe their boxes had 15 15 things up there. And it was basically, what is the feature? What does it help you do? Red, yellow, green. And that's it. And I just thought that was a simple. I mean, to me, that's a simple exercise. Like again, that I can have Claude or Chat GPT in the background go help us run this exercise. we have feature flags, I've got pr usage data. Can I basically go figure out which which features they're using or not using and then be able to build that slide for every single customer that we have and then figure out, okay, should we be going to talk to them about these things? so that's like one, I think, interesting thing that's come out of all these account transitions is just this kind of feature usage. How do we get adoption on certain core products? and how do we get better at convincing people to use those things, right? In the future. and then the second thing is just around upsell opportunities. You know, I think it's just a Not not that they're, rampant on every single call, but definitely as accounts transition, as somebody gets in there and asks questions and is curious. again, I think there's just a a natural, hey, you know, I'm interested in this, and maybe now's a good time for us to talk about it. And so it just reminds me, I guess, of of how sometimes we get caught in the mundane, like, okay, this account's been with the CSM, we've been having regular calls, or maybe we've been pushing updates to them and like, they've just been doing the same, same, same. And I don't know, a little bit of variation is is helpful to I guess get them out of the the normal thinking. speaker-1: Yeah, it's like it's like, you know, fresh blood, new life into the account when somebody new is introduced. And there's a lot to unpack and everything you just shared. But a cup I wanna I wanna hear more about like some of the segmentation that you're doing. Like I would call that psychographics. Like what is it about? Like Yeah. Nothing nothing demographic, but like how do these people behave? And you're absolutely right. We we we we could not really see that before. You might have you might have a couple of options and somebody would have to choose from a drop down list, like these people are pain in the ass or they're really easy to work with. interesting side note on that, what one of the characteristics of customers that I've picked up on over over time that makes them really good customers, almost in any industry, no matter what software you sell or what domain they're in, is what's the tenure of the team that you're working with. So if you're working with people that have been working together for five, seven, ten, twelve years, it's almost guaranteed to be a better customer outcome. Interesting. Somebody who has a lot of turnover because There's a maybe a cultural issue in the organization that doesn't, have a long term focus on on strong outcomes with their technology partners like you are if you're a technology firm. So anyway, some of some of the best customers that I've ever worked with have had teams that have been together for a long time. So it you can't control that as a vendor, but you can be aware of it and you can respond accordingly, right? And and manage risk accordingly. So that's one of the things I thought of when you were talking about the psychographic segmentation piece. speaker-0: Yeah. I I think of this too a lot, I guess w we used to talk about like building these micro segments, you know, and and having some flexibility of like, hey, maybe for in my mind at least, right? Like maybe there's a period of time when a month or two that I just want this customer segmented for this reason or like to be able to sub segment into like these these smaller segments for this reason. and again, I feel like that was very hard to try and do, right? Okay, I have to great, I have to go ask my Salesforce admin to create a secondary segmentation field, then I have to figure out what the options are and it feels very limited, right? And now Literally I could I just told Claude, hey, I want to go find customers that are doing X, Y, and Z with the platform. Go build that as a field inside of HubSpot. let's segment that out. And now I have a list of customers that are doing that and you know, can use that going forward. But I I don't know. So I like these micro segments are are also a thing to do. And then to your point, like the just some of the psychographic and other information that you can pick up on is super helpful. Another thing that we just used, I don't know, Claude and some other stuff for, which just popped in my head is We're going to a conference. we're going to a conference. And so we got a a list of attendees and company domains at this conference. And we went and, you know, again, rather than I think in the past I would have thrown that sheet over my team and said, Hey, can you find your customer and like which one is here? You know. And now literally it's a five minute exercise. Hey, tag the CSMs here. Hey, these customers are going. we spun up a we spun up a calendar invite for this like one of our CSMs is going to be a customer representative for us. So we spun her up a calendar invite. She sent that out and just said, Hey, we'd love to book fifteen, thirty minutes with you while you're at the conference, talk to you about this. And now, you know, she's got tons of meetings on her calendar for customers that we're gonna see while we're there, which is awesome. And then I just told her, behind the scenes, I just went and scraped our customer OS and basically are I'm giving her account account plans for all those customers so she can just have a hey a flip book of like, this customer's coming up and basically attaching it to all of our calendar invites. Hey, this customer's coming Again, one liners, what are they using it for? What products do they have? how much ARR is it worth? But just simple stuff that they could be using. and now, she has that and then I'm I'm sending that to the rest of the the team that's gonna be site. Hey, if you see these customers, here's like little one pagers for you. speaker-1: You know, we that's very natural to do that kind of thing for an event because events are very expensive to put on, right? And it's expensive to fly people around the country and you want to get the most out of an event that you possibly can. as you were talking though, it dawned on me that it it i let me go even further to say even before Claude and AI, it was worth the time spent doing what you just said because you did spend so much money on those events, you want to get the maximum you know result from them. But the take what you just said and now you can extrapolate it almost to every marketing campaign that you're doing. If you're doing a new product release and it's just, you know, you have a webinar, or if you're, you know, releasing a new capability or feature, or you're you're bringing on some thought leader onto, you know, an office hours call, whatever it is, right, you can actually now go spend more time on those activities. You could have well the customer engagement agent, if you will, that goes and and drives, you know, the follow-up from those Those pieces, making sure your clients that that need that saw it, making sure that you had a conversation with them after they got the notification. The other thing I was thinking, as you were talking about the scorecard, the capabilities of the platform that you're using, not using the red, yellow, green, two things. One, I like the simplicity of that. You ca we have to even though we can produce a lot of content now, we still gotta really work hard to make it simple, right? Yeah. It's the whole Abraham Lincoln, I didn't have time to write you a shorter letter kind of thing. You gotta really still do that so that People understand what you're sending them. But the other thing that that struck me when you were talking was communicating on multiple channels regularly is the way to get the word across. Get the word out about whatever you need to get the word out about. If it's like, hey, you're not using this feature, tell them in the application itself, send them emails, you know, have webinars that you invite them to specifically about that capability that they need. And then Another channel is human, which is what you described, but that can't be the only channel, especially now, because we just don't have that much human capacity relative to the number of customers that have to be served. Right. The other thing about doing a simple view of this is you could actually send it up and down and across the chain too. hey, here's the stuff that you're doing and not doing with the platform. It's a way to maybe get around your if your champion is a blocker, it could be a way to get around, right? Make it just look like a marketing email that came from the product as opposed to an email that came from my CSM Jeff. Right. So it doesn't look like you're hopping the the organizational hierarchy, but that you're you're actually, the product is telling the organization, hey, you're not using me well enough. Let's talk. speaker-0: Yeah. Right. Well, it reminds me too, if you keep it simple, then you could also push it into other teams very easily, right? Can support understand what this customer is using or not using? And even can they make again, maybe they help s fix a support issue, but can they also, hey, you know, fix this issue, but I noticed you weren't using this feature and this is actually something that could help resolve this in the future, you know, that could help whatever. So like can you make it simple to do that? And then one other one other tie in, I guess, for the segmentation piece that goes without saying too is just starting we're starting to push the segmentation piece across other teams here. so everybody's like speaking the same language. Okay, we know what it means if we're talking about a tier one customer. and, you know, we we understand like what what level of support we're giving them, kind of an idea. So speaker-1: Yeah. Yeah, totally. And that's another reason you gotta keep the just because we can segment by a bunch of different things doesn't mean we should. Yeah, maybe you have like your master segmentation, like with market segments, you know, maybe how much they they pay, maybe you know, the use cases that they're using. But then ultimately, you know, some of those more micro segmentations are campaign level type stuff. speaker-0: Yep. so yeah, anyways, just playing around a bunch and I guess like some of the the real stuff that my team needs to to execute on daily. So it's been helpful. speaker-1: That's cool. yeah, l lot of good stuff in there. all right, so maybe a little bit of a separate topic. I'm curious where you have been building your agents. I I was I actually had the opportunity to go to a sales kickoff of an agent company last week and it was really so I I I got to sit in the room with their sellers for a full day and sort of hear some of the challenges that they have getting their product into market, they sell an agent platform, right? So think of it as something that would be maybe competitive with Claude or with OpenAI's platf platform capabilities. Maybe Databricks Genie, like some of the other, you know, subs Palantir potentially could be a competitor. Glean might be a competitor here. and there's a lot of these, right? So they're they're not they're not isolated. But I it got me thinking about and and we had a little bit of a discussion in in the pre-call about this too, but where do you actually build agents and what are what are they? So i i'll I'll start with maybe a definition, my definition of what an agent is. And I I like to revisit this every once in a while because I it helps me clarify my own thinking on it too, because we're getting into jargon land again with all the AI stuff. We I mean jargon is just part of the industry, but we gotta be clear about what we're talking about. So to me, an agent is actually just the new new terminology for a almost like a SASP. product or a a piece of software, right? Yeah. and it's got UI that helps me understand what's what the intelligence is what what the system is doing for me automatically in the background. It's got different points of of for me to me and my teams to interact with the intelligence that's being provided through that software. and then it might actually have things running autonomously in the background, doing work. that either needs to be checked by a human, informs what a human needs to do next, that facilitates some process for me, right? So agent to me is just new software. And I think a lot of this software is being customized per company, right? So everybody's building their a lot more of their own software these days. There's gonna be more software in the world than there has ever been. We know that. But then the question becomes where do you build this software? Where do you where do you run it? Because you can do it in claude. Right. You can go build this stuff in Cloud. I think Claude to me still has a single player mode problem for teams. Yeah. Which is, there are a lot of capabilities in Claude now that are that are making things shareable, artifacts shareable. I don't know if you've seen their they have like the equivalent of Google Docs and slides and everything now that can be shared across teams. Like how many hundreds of millions do you think Google spent building that stuff and Claude replicated it in like months? Yeah. which is interesting, meaning All software can be replicated, it's cheap, it's value less, you know, because it can be generated. But anyway, point is like we still have a multiplayer agent problem. And I think, you know, building software the new way with an agentic mindset is the way to get around that problem. But I'm curious, where are you building and employing your agents right now? So speaker-0: So I think the I guess this is the qu this is the question I'm asking myself right now, which is, how do you I think this is larger than just me. This to me seems like the company wide decision that matters of like, okay, are we all picking a platform that we're all building agents on? Is each department picking their own? Is there benefits to that or pros and cons? And and I think there's a real sense of like making the wrong decision here because you feel like stuff's just moving fast and you could get kind of lost, you know, left behind. so I guess my short answer right now is I'm building them in Claude and it is a single player instance. and I guess like for me that's fine right now, basically because I'm building these agents and then I'm just pushing this information into like customer OS and a platform that my team can can operate from. And so, you know, to me, some some simple things that I'm looking at right now are like some agents that are running are an opportunity agent. So something that's running in the background for Upsell, cross sales, renewal pipelines, and it's listening, it's looking for new calls from Fathom for those customers. It's looking for Slack messages, emails, any sort of context. And then it's trying to help the CSM update key deal notes or update next steps, make sure it's staged appropriately. But basically, how do we make sure the opportunities are all kind of housed or categorized in the right way? and again, to me, kind of a very easy use case in the sense that like it's not customer-facing information, it's really just for our internal purposes. There's there I don't know. There's very little that we risk right now by running something like that. because it really is I guess like the forecast is really the the thing that would change. And again, we look at the I look at that every single day. So if something is awry, I can I can quick kind of quickly solve it. But that's like one example of something that I'm running right now. and I guess like the reason I chose Claude is just that our engineering team has done the hard work. Like we have an enterprise license, we've done the hard work of connecting kind of all the tools that we have available to us. Like it it honestly is just like everything's there. Like why I don't know. Why make it harder on myself, when everything's already connected and and hooked up? It's like, wow, it's pretty simple for us to just go in there and start using it. speaker-1: Absolutely. and I think I mentioned last week my my entire team is in the process of getting certified on Claude right now. Because honestly, if you can work with Claude, you can work with any agent platform. Yeah. Right. I mean there there's just not that much difference between what these things do. Every pitch is the same. but it to me, like so then I I look at something like like what you just described and say what are the gaps? I mean you you've got multiplayer, right? Because you're pushing the outcome of what the agent is doing, the re the results of that, you know. Agentic process, you're pushing that into the CRM or you're pushing into a list that that the customer success managers or that team can manage through customer OS, your your customer product, right? Or your customer success platform. It's connected to other data sets already, which is a shared resource. That to me is the harder part, right? Yeah. Over time, like you know, do we have everything connected in to allow the agent to To go discover all those resources and pull in those resources and make sense of all those resources and do you need to do any fine-tuning there? The only gap that I can see, which is not a big one, is that when your Claude account gets shut down, when you leave the company, does that agent stop working? So it's a question of operationalization of these agents, right? And putting them into a production environment that we can rely on. But, you know, that's where a platform like the one I was talking to last week comes in, or Glean. But the The question I have is that worth replacing your platform for or is there a way to do it? Because Anthropic has platform capabilities that I can go create a shared agent, right? It's a little bit more of a developer task to do that as opposed to setting it up in cowork and putting it on a schedule and just going. but it's a question of permissions and and how you and and sort of the how how how consistent that thing needs to be from an operationalized standpoint. I would say the bigger the team. that you're doing work for, the more that needs to probably out of clawed and into in the more in the more s mission critical that process is, the more you need to be out of clawed and into a proper agent platform to run it. Just so that it is monitored and if somebody's keeping track of it just like you would any other systematic process that's that's going on across the organization. Somebody needs to make sure it's up and running every day. speaker-0: Yeah. Well it's interesting too. I I I think I was just listening to podcast with Palo Alto Networks CEO. in speaker-1: Yeah, he's been all over the place. speaker-0: Yeah, he's interesting. And basically I guess like one of the things I took away from that is that like security threats and attacks are basically just gonna go up, right? With the more advent of AI, the again kind of the the misuse of it by us business folks versus like true developers. And to your point too, I think that's another reason. It's like I I've like stayed in the cloud environment 'cause it's just like, cool, my my engineering team has done the work to set this stuff up and, you know, is essentially monitoring this rigor. I feel like if I was to go buy a tool and and do this, right, then then all of a sudden like that becomes another operational burden on me to go run and figure out and keep up to date. And are we using the right permissions? Is it connected to the right tools? Like like that to me, you know, is a kind of another example of like, okay, that's Again, it it's almost gonna add more to my plate if I have to go buy a separate platform versus can I just can I make Cloud work for me for right now? And then eventually does it get to a point where you know, does it get to a point where then it's like, cool, this is a this is becoming a com a company wide, you know, effort and we actually should have a centralized platform where everybody can see these see these agents running, can be able to manipulate them or change them if we need to, like, versus it being on one person's machine. speaker-1: I I think it also I completely agree with that. And it also depends on how seriously you're taking the this whole thing. What I see r a lot of right now is task replacement versus role replacement with with agents. And I mean look, I'm not trying to to work anybody out of a job, but the reality is the the companies that are really doing this well and what we're doing with our clients who want to implement agents is or who we're convincing to implement agents in a lot of cases. Is you're we're building you a role specific agent. That role specific agent is going to do several things for you. It's basically going to take work off of your plate and tell you when you need to go engage. So let's take, for example, our CSM coaching agent. Like that thing is just working all the time in the background. It's evaluating calls, it's evaluating communication and Slack and email. It's actually assessing the team in real time every day against their communication and a rubric that we set up for how we want the that that team to behave, right? And then it's providing that data to the manager. So it's almost like a it's a CSM coaching agent, but that that coaching agent is an extension of the CS leader. Right. So it's a CS leadership agent. and I think when you get into that it and by the way, so then it's it's Putting that information in a place where it can be accessible on demand if the CS leader needs to actually go look into somebody's performance, or it's saying, okay, it's time for your quarterly review for this person. Here's the information to review. It's basically teeing it all up for you. So it's just like boom, boom, boom. Okay, we're going to have conversation about this with the CSM now. I'm not going to find all the data at this point. I'm not starting from zero from scratch. I'm actually getting ready to do what only I could do, which is have a conversation with this person about their performance and how to improve. That is a much different mindset, I think, than hey, I need a call follow up task doer agent. Right. And so like how do we how do we well i in that environment, I think you do need a little bit more structured and durable and operationalized of a system because those are gonna become mission critical systems at some point. They're speaker-0: Yeah. speaker-1: gonna become part of the fabric of how you run the business. speaker-0: Yeah. And I think the I guess the other thing in in my mind about these agents or, you know, how you're thinking about some of these some of these aspects is that I guess like in my mind I'm trying to optimize for two things. One is more time with customers or more time with people internally. And like if I can optimize for those two, like yes, I want I actually would rather you, I don't know, be on a call with our product team talking about certain things versus just like automating a message to them, right? Does that actually get get across better. Do we actually have, more productive meetings and other things? and like, do I get more time with customers? Like, you know, can we be sitting in front of customers much more frequently if if we're able to take some of these things off your plate. So so I think that makes a ton of sense. And it's funny you mentioned the the coaching agent. I am in the midst this week of setting something up like that. Like I have a rubric for my team that I'm like rolling out on like the skills of a CSM here at Junction and like then I'm going to assess them across those and it's basically pulling from Fathom, Gmail, Slack, all the tools that we're connected to. Nice. And and then trying to provide some like clear, you know, concrete examples. and again, I guess the idea in my mind is that like it could do a much better job of going to surface all these things. And then I can, like you said, I can be the one to basically basically at the end of the day confirm and synthesize this down and then have the conversation. But like no longer do I have to go spend time gathering all of these examples, pulling all this stuff together, which is again, it was useful like we needed to to develop the team, but I have have a much better way to do that now and a much more consistent way, right, that I'm doing that across the team. Where it's not like sometimes I would probably get subjective and pulling certain examples here and there. And now it's literally like this is just done for me. And I think like that consistency to me is a big thing too that you get. speaker-1: Yep. One hundred percent. And it it's if you train it to be objective as much as possible, it's always gonna be subjective to a to a point, but you you're gonna get objectivity, right? that's to your to your exact point, y it's gonna be done the same way over and over, which is why you also have to continue to train these things, make sure they're still doing the things you want. So you know, I think the you know, as I think about I think about every day, like our business, we're we've been a you know A solution provider on one specific software platform, Pendo, for two years. And we will continue to do that. There's no question about it, right? There's a very healthy business there. Pendo's a healthy company. But what what I'm starting to lean on more and more in we we are helping our customers use that data from the from from that from Pendo. and we're working in those roles, product management, product marketing, customer success. So The way I'm thinking about it, if we're gonna succeed as a company, we have to be top-notch domain experts more so than we have to be AI experts. Now we have to have AI expertise, we have to have AI engineers that can go do the data work, figure out how to deploy this stuff, you figure out which tools to use, which connectors to use, all all the things that you've described, right? That your engineering team has already figured out. Like we do that as well. But to me, the winning play here is it's just It's just domain expertise. It's never been a better time to be in consulting. And I think we talked about this last time too. But I think now every saw every traditional SaaS company needs to really think about how to turn their product into an agent, go do forward deployed engineering to do the work that their product was, you know, supposed to support in the old SaaS world. Now just do the work for the customer if you can. You gotta become a BPO for that role and for that function and for that process. Business process outsource. You to own the outcomes. speaker-0: It's interesting. The the other thing I'm curious about as you've gone it down this route more and more too is again, I'm kind of like early on this. I would I would say I probably don't have a lot of the controls around some of these agents, but I guess are you really finding that like evaluations and feedback loops and like training data, like are those things becoming I guess more standardized for these agents? Like are you seeing that like those things need to be in place and that like you should really be like I I'm assuming the answer is yes and asking a leading question here 'cause I know I need to get there. I'm just not I'm not doing it yet. but I guess I'm I'm curious like if you guys have explored those things right now and maybe like you know if there's a a learning or two that you're coming out with. speaker-1: Well it's it's the only way to make it's the only way to make the autonomy truly autonomous is to learn and iterate on itself, which you have to do through the feedback loops. Evals to me, you know, we've we've been doing evals for a long time in software. They're called unit tests, right? It an eval is just a way of unit testing the output of a of an L L to make sure that it still gives you, you know, a directional an answer that's within a a range of what you would expect because it's not deterministic. It's not like you write code and the same thing happens every single time. Like a unit tested. So but to me a an eval, and I'm sure there are people who would, you know, shred my definition of an eval, 'cause I'm not this not my personal area of expertise, but that's what it is to me. So I think yes, you have to have those components. Guardrails would be another thing, right? Like when when is when when is it not okay for the agent to take action? in what circumstances do I have to shell out to a a human or some kind of review process or maybe another agent, right? Because that's a good way to use agents too, is like have them check each other's work. so so yeah, I think all those components are becoming important, but they're not as difficult to implement, especially if you come back to the platform concept that you talked about. Like what is the platform for for deploying this stuff? I mean, you can do all this with skills today and with you know, how you how you basically write the prompts for the agents themselves. But I think having some structure to it and having a way to manage the exceptions, having having a way to continue to train and optimize. Well, one of the simple things that I do whenever I build an agent in Claude is say, Hey, look, I want you to to remember, you know, all the decisions we've made along the way. And as I correct you, I want you to update your your memory on that. And you know, sometimes I'll even have it store that in a markdown file. That's like lessons learned and it'll then applies the lessons learned every time it runs the agent the next time. That's a simple way of doing it. But but it still, has the same effect. speaker-0: Yeah. all right, last last thing I was just curious about. Have you seen ramp and their new concept of glass? Have you seen this speaker-1: No. No no no. speaker-0: ramp.com slash glass, I think you is what you can go to. Okay. and it it's an I guess like it's another interesting thing to me. Like it basically is like a they're trying to go after I think the opportunity you're talking about, which is they're trying to be a multiplayer AI sw they basically built a harness, I think internally is what I'm picking up on, and then have basically figured out like, hey, should we actually release this as a product? Because, you know, again, their whole shtick is like how can we help you save money? And so like glass one of the ways that glass can do that is routing tasks to the right area and basically building in controls for you to do that and kind of having a multiplayer AI. I don't know, I just I feel like ramp is always just popping up at h every day for me. And I just saw it on LinkedIn and have seen a bunch of ads for it. So I was just curious if you had seen it. But it seems I guess it's another interesting I guess it's another interesting thing to me because again, like almost every company is converging on the same thing, which is like, okay, you we basically need to get you to use our platform, connect all your data sources to it, and become the router of like where all this stuff is going. and it I don't know, it's just interesting because it's like there's can you feel like there's only handful that are gonna win and and I guess has have Claude and OpenAI already won because they, you know, s effectively got us to use their harnesses and connect all of our tools to it and it's hard to break away from that. But it still feels like the market's so big, right? Like there's really not that much AI usage as we think out there right now. And anyways, I thought it was an interesting play just because also, you know, think of Ramp as a finance company and I don't know, here's an interesting slant on their product where they're like, Hey, yeah, this is still kind of built around finance 'cause we're trying to save you token usage and dollar usage and trying to make this as transparent as possible and help you route things let's route the tasks to the right models that your team isn't overspending tokens where it doesn't need kind of a thing. So anyways, I thought it was interesting. speaker-1: Well, d do you know how few companies in the world are actually concerned with that problem right now? Back to your point about AI usage overall. w we are, we think about that every day, but the vast majority of companies are in the late majority on all this AI stuff. We're still in early adopter territory, in my opinion. Everybody's using it as an individual. Not everybody's using it as a company yet. Not in a structured way. So I did see glass, but I didn't know it was called glass. I saw it when they're when they first announced it and their CFO basically came out and did video and talked about how their token spin got out of control. I also think token cost is going down like really, really fast at this point, although we're using more tokens, but it will continue to go down, down, down. Now we'll see what affect the US elections and the data center moratoriums. Like we'll see what effect that has. But to me it's a foregone conclusion. We're gonna build more data centers. It's just gonna take a long time. Yeah, we're it's gonna it's gonna be a painful slog. All that to say, back to your point and your question, when you think about an agent harness, why why can Ramp do this? They already have the most confidential data that you have as a company, which is your financial data. So basically I can trust them with anything else. It's about trust. Yeah. Sort of sort of like building agents is about domain knowledge, not to technology. I think where you're gonna put your Data is more about trust than it is about than it is about feature functionality per se. Because the feature functionality, to your point, is all going to be the same. So I actually think and a lot of people, the one the really forward-leaning companies actually don't trust Claude and they don't trust OpenAI. Yeah. Because they're not sure that these guys aren't going to use their data to go train their models in the future. And if I have some proprietary, you know, formula for how to make chicken feed that creates superbionic chickens that are really healthy and big and sell in the market and are good to eat, then I don't want to give that trade secret away by sticking it in clod somewhere. speaker-0: Yeah. speaker-1: So, and make no mistake, that's what most businesses is like that kind of stuff, right? It's not all digital like we think it is sometimes. Yeah, yeah. Yeah. But it's about trust. And I think every industry will have every industry, multiple domains, you're gonna have agent harnesses that that come up to be domain specific, industry specific in in all sorts of places. So I think it's gonna be very, very diverse in terms the functionality will all look the same, but who you trust as a company with your data is all going to look very different in terms of the logos involved. speaker-0: It's like to your point, y earlier too, right? Anthropic, right, has already created Google Docs and Google Slides and all this stuff. You it's like you know, so clearly they're gonna try and go, you know, use their data use the data that they have to their advantage, right? And so it is interesting that the trust word to your point becomes so so much more than I like first anticipated when you start thinking about this. We're like, shit, yeah, where am I push where am I gonna put this stuff and can I actually trust it in there? and it's interesting about ramp, right? Like they do have Yeah, we can trust Ramp because we've we've got all this financial data there that has that they've been and then also I think what works in Ramp's advantage too is their whole shtick since they've come out as a company, right, has been how do I reduce your spending? Right. We want to we want to be building tools like again, most credit card companies that have come out have said, let me give you rewards, let me figure out how to make you more spend more with us, right? And their whole time they've been saying, let's let's do the opposite. So again, like I actually think like them picking that at the beginning now even has even further advantages because like, I I actually trust them so implicitly because They have been built their whole company around this idea of trying to reduce my cost and trying to make it make it, easier for me to get transparent spending. So of course I would trust them with this glass product that they're like trying to do the same thing. speaker-1: Yeah. It's so funny that I I forgot to mention this earlier. You were talking about you're afraid of making the wrong decision and and you know, choosing a platform that you get locked into. Well, guess what? Guess what open AI and Anthropic are trying to do to you right now with all these features? They're locking you in. They're locking me in. I realized it this week. I was like, man, like these new like docs functionality, this is really cool. We just do everything here. And then I was like, whoa, whoa, wait a minute. The minute I start doing that, like what happens when I do want to shut this down? Do I lose all those artifacts? Yeah. Microsoft 365. You're just not gonna do it. So and that also goes to prove that it's not the model that's the the valuable component. It's still the software on top of the models that that's where the money is because the model companies are building software on top, codecs, clawed, right? They're building these harnesses and these tools that sit on top. So speaker-0: My last my last little anecdote for you and then I've got a drop. The the last point that I think is gonna be interesting is I think with again kind of making the wrong decision what we were talking about earlier, I I was actually wondering how annual contracts do those actually do do customers now want so much flexibility and do customers have the advantage, right? Do do like as a consumer, as a as a buyer of of products and services, do I actually have such an advantage? that I can actually force cu force these companies now to say, Hey, I'm not gonna sign a yearly contract, like I'm gonna go to monthly because I want the option the flexibility to be able to move quickly and and change. Again, I don't know if that will will happen, but I I think that's coming sooner rather than later. It's like you're gonna get a lot of people in the market that just say, you know what, I'm not gonna sign up for twelve months and lock myself in. you know, just can't afford the risk of doing that. Like, it's almost like the opposite problem that we've had. But speaker-1: But I think I think these companies are gonna know that it's painful for you to switch and they're gonna give you just enough pricing incentive, just like we would, right? we run these kind businesses too. Just just to give you enough financial incentive to to not do that to make it not worth the headache. Sure. So but it's a very interesting conversation. All right, let you go. I got one I want to talk to you about next week. a friend of mine building something that I think is really interesting. So I'll I'll put it on the docket for next week. And talk to you then. See you soon. speaker-0: All right, good to see ya. speaker-1: You too, bro. Bye.