Build Mode: Becoming a Builder in the AI Era - Agents, GTM Systems, and When Not to Use AI

So thank you everybody for joining us today. I know that we are all inundated with information AI everywhere every day all the time and there's this new term with builder that has become more and more informal to the point that even at HubSpot our team which was the developer relations team is now the builder relations team. um other companies like Longo DB are following the same trend. So we know and believe that we've all are capable of ideulating solutions and now with AI taking the next step and that's what we're going to talk about today is what does that look like? So I am joined by the esteemed panel Julie Andreas sorry. Um okay and I I want all of you to do a brief intro but I want to start off with a very simple question over the last year and a half things have been moving very very quickly. What was the moment where you're like it clicked and you knew things are my business is going to be completely different? Eco, let's start with you. So, I'm Lea Landers. I'm the founder of Minometrics. Um, we build out a workforces for companies. Uh, the moment that I knew that this was going from different uh either end ecosystem, the emperor my story, my business was um, you know, rising and then I had uh, a downfall. Um I had plenty of clients that were on signature bank signature bank but wonder at soap did buy line retainers at the time and I decided to build that using only a high um and uh yeah that was the moment that it clicked um to be able to run a business using poly uh agents. We are currently at I say me directly and some of these AI agents and um we are uh building out a very specific kind of framework called brains. Uh I'm Andreas my company is double consult single person firstly operation a lot with it of AI and what I do is I code plan build and implement GTM data systems basically what does it mean big words that means that my clients have very bespoke data data that matters to them where they where their systems and their people need it and the most important part is where they actually use it. So it's very easy to get something that looks nice. It's very hard to get something that's actually running as a system or used by people, which is very close together, sometimes very far apart. That's what I do. And yeah, very happy to be here. Thank you for putting this together. Hi everyone, I'm Julie. Um, I'm part of the team at Geekly Media. My job spans client success, operations, sales, um, and partnerships with HubSpot. We're a HubSpot partner agency. Um, and from my world and my perspective, I have a few different points where I realized things for us were going to be different. Um, one of them is actually well before the last two years, I will say. Um, it's when someone put an AI notetaker in my hands and I came up in the agency services consulting world where um, my job was to take the notes and write the action items and do the things and if I didn't do my job well, there was no record of what had happened. And then all of a sudden there was a perpetual record and I still took my notes and did my things and like I'm really good at that. Um 12 words per minute. Okay, like I built the skill, I shall use it. But that was a really big moment of things are different and there's this huge utility and now we freed up all of this other way of working and using our brains. Um I also in a previous life worked and managed a product team and we had a really niche solution we were trying to solve where uh clients kept asking for a sentiment analysis in the forum software we were building. So it was community forum software and the engineer I worked with said, "Oh yeah, yeah, yeah. Here's how we do that. I'll just go do this. Bing bang boom. We need an LLM. This and that blah blah blah." And this was not in the last 24 months. And I said, "Cool, cool, cool. Um, that sounds great. I'm going have you go do that. I, however, do not know." And that was a big sign for me of like, "Hey, this is a research point. You got to figure out what's going on." Because if the first thing your engineer is doing is saying, "Yeah, we just need an LLM." Like, cool. What's an LLM? How does this GTM girly become whatever that is? Um, so I think those are probably my two biggest transformational moments. That's great. And I think Julie, you just got us to the next question, which is, you know, we're we're all here because we obviously are interested and believe in this AI thing, but that's not everybody. Like, what does it look like to become a builder? How do you get started? And in the area of going transitioning from, hey tech person, go do this for me. I'm going to do it myself. like what what does it take to get into it and then start to become proficient? Question. Okay, great. Um, sorry, I thought this question was maybe for my other friends. Um, and happy to have your input here. Um, for for me in particular, I've sort of built my career on figure it out, right? My entire exercise in GTM is problem solving. It's what motivates me the most. I'm doing my best work when I'm solving problems. And now with AI tools, I can start solving problems myself a little bit more. You've given the keys to the car to like a 15-year-old driver. That's I think who I am on this panel right now. Um so for me, becoming a builder is simply trying to build something and put it into the world and into the universe. Um, I work in a space of our agency works largely in the real estate and housing industry where software is antiquated at best. Um, no APIs, no connection points. We're doing some really old school things. And then as we talk about bringing in a tool like HubSpot, once we have the data and we can build on top of it, we suddenly have all of these opportunities. And now that myself and my team have the ability to take these wild ideas we've had in the past and bring them maybe into when we're doing client services, proof of concept before we bring in an engineer who can for real do things that we feel really good and stable about for our clients. Huge. We love it. Um I think maybe rest of panel I'll let you talk a little bit more about being builders too. I'm sure you have thoughts. Absolutely. So uh after not answering your actual question, what really changed which goes a little bit into I was focused on the microphone thing. Um what really and this goes exactly into that what there were two really three pivotal moments for me that was first when I discovered clay in 24 that was like oh this is how you can use AI to solve GTM problems and I didn't know how to do it but I knew you are able to get there and this is the other thing using AI is this yeah and this is the other thing using AI to like ask unafraid questions like it's such an amazing tool in the space we are building to learn. I mean this is different in other domains but everything we are doing is technical documentation. Everything is made for AI basically. Is that loud enough? I hope uh that's that's good. And that is like really what it means to me to be a builder these days is find problems that are worth solving and ask absolutely unafraid questions. The beautiful thing about AI in GTM or in Teasers, you can ask the most, let's not say stupid, but you know these questions that you really wanted to ask your coworker but you couldn't because you should have asked it a year ago and it was really odd to ask now. These questions you can absolutely ask Claude and when you have a nice custom instruction that doesn't even give you about it. Can I say that? Sorry. Um, and this is to me being a builder to to put it shortly is finding problems that are worth solving and then just have at it like this. I can get there is not the challenge anymore. Is it the right thing I'm building? That's the bigger challenge. But that's later in the conversation. That's what I would say what it takes to be a builder. Uh, I would say don't be afraid to have things break. uh it breaks all the time and that is part of the process. Um you want to iterate, you want to keep on working at it. I think if you start off uh by trying to build that most perfect thing, it's just not going to happen. So, uh don't be perfect. Just start um there will be a different version of your grand idea the next day. All right. So, we've heard prototype, we've heard fail fast, don't be afraid to break it. And that kind of gets to the day-to-day, but if we're always breaking things, how do we ever get to a point where we have like a full solution? So, like what what is the what is the reality? Like, what does your dayto-day now look like that you've transformed your business with agents and how has it changed? Yeah. So, uh I I started early on when I was breaking a lot of things all the time, uh building and then I realized that I needed some kind of structure, some kind of framework to build with purpose. I I really believe what Andreas is saying as well is like build with a purpose and that's when uh we developed brains which stands for build, run, audit, improve, normalize and scale brains. um because when you are deploying uh agents into your business, there can be a lot of negative consequences. However, uh things are never static. Uh the LLMs are improving. Uh you'll need to be able to switch up your LLMs. Uh the the APIs that your LLMs are using are being uh upgraded as well. So, I think there's a large part of what what you need to do is document. I spend a lot of time documenting how we're building. Um, and then breaking it down back to what is that purpose? What are we trying to do? Um, common thing that we say all the time is like what is the job to be done? So, anybody that's familiar with job to be done, that's most of my day is like figuring that out. Can you spray it? uh build uh run, audit, improve, normalize and scale. So normalize is is making sure that your uh other human team members are able to handle the AI uh employees as well. Um as much as I am very much for an agentic workforce, it's kind of like my thing, I don't believe that AI is going to replace human beings at all. And and I'm going to pause you there because I want to know what is that point? What is that point when it is too much or AI is not the best fit for? Andreas, I'm going to start with you. Like with your clients, when do you hit a point where that maybe AI isn't the answer? So I think there's two kinds of too much. The first thing is and I think that's the most solvable which is am I even so I think especially in GTM in that domain uh LLMs are so good that if you do the right thing with it it's rarely the thing that can't execute what you want to do. I would say the bigger thing is am I solving the right problem with it and I think you're very much drawn to have doing everything with AI and most of the time there is a solution that you can just do in code where I'm using AI to produce a Python script or I produce something I always call it formulaic whatever but where it's like really just me using AI to produce code but it's very deterministic. Uh I think the problem is when people are trying to solve things with AI that should be deterministic and then you have these random outliers where it's like oh yeah it's not my fault it's AI yeah sure but you shouldn't have used it. So this is I think when you solve the right problems I think AI can get you pretty far and it's very capable. The other thing is too much you can ship so fast today that it's like impossible to catch on the other side things in my consultant world where I'm using AI for example I'm just making it very brief but it's super fascinating to me I worked with a client and I realized guys everybody you sell to everybody's on LinkedIn you do zero LinkedIn stuff that's blanket wrong you need somebody to help you with organic LinkedIn content and the co was like yeah sure like that makes sense And like 3 days later, I had like a request for for um proposals out based on my client's needs sanitized that I didn't give away who the client was. I had four conversations with people from my network and I used those conversations to package for my client what they do what how do they compare and I shipped it back to them and they were like dude it took like three weeks for them to have the bandwidth to do something with that. So I think there's two ways AI helps you to ship very fast and that can be sometimes too much because people need to catch it and when you solve the right problems it's very capable to actually do that. And Julie, so I know when I first started with AI, I I really wanted it to behave in a deterministic way and there were a few um tools that came out that sort of did that, but it was completely against what it was designed for and and so I know that feeling. um when you think about like uh an organization and um creating like best practices within the organization or like having a good sense of knowing when to bring in a developer like how do you build a strategy around that? So my my biggest takeaway is AI needs an editor. Um I have in the last few months received documents from clients and prospects some other folks in the world um 60 75 120 pages of specifications for a build produced with AI and you start reading and pulling back and I've been doing what I do for a long time and I have my own thoughts and opinions of best practices and things like that and AI doesn't have my thoughts and opinions thankfully for all of you. Um, but you start peeling back the layers and you realize this is overbuilt or overexplained or an actual human who needs to go in and use a system is not going to check 75 boxes to move a deal from one stage to another, especially at the beginning of the pipeline. Like it's not logical, right? Um, so that's not a builder type of answer, but it's this this example of AI needs an editor. And when I think about making sure my team and the folks I work with are embracing that mindset, it's just because you can build doesn't mean you should build. It's just because it looks good doesn't mean there's not something wrong with it. Right? I um have some experience outside of work in arts broadly. And if you've done anything in the arts, you've probably gone through some type of critique, right? Where you are critiqued, you critique others, you have to be receptive, you have to be thoughtful. And sometimes those conversations are not about what's right. They're about finding the thing that could be better, the thing that's going to break or that edge case. And those are difficult conversations to have. They're difficult conversations to build into your AI workflow if you're not mindful of them or comfortable with them. Um, but that's the type of work we need to do as a team, I think, to be successful. That's awesome. I I'm going to repeat one of the things that you said, which is just because it looks good and it's running, doesn't mean it's right. Is AI doesn't get rid of tech debt, it's still a thing. It still exists. And that's one of those nasty things that can surprise you a lot later. So, I I really love that. Now, you know, we're talking about go to market, we're talking about revops, we're talking about being builder in that role. When you started transforming your business, was there like one obvious kind of slam dunk process that you identified that you tackled first and how did you think about that or how did you know like, okay, we got this this absolutely nailed. Yeah. So, one of the first things that I built in it app that we're at a HubSpot event was a HubSpot agent that's able to do HubSpot admin. Uh, it's like one of those things who wants to sit all day and like fill out data and uh, you know, clean up the database. Um, so yeah, the first thing that I built was a HubSpot agent. Um, and uh, what's nice is that through Slack, it's conversational. So, I could just instruct it as a human being. Hey, we just have this import sheet. Uh, please clean it up and then import it and lay all the associations as is necessary. Um, it does a really good job also of uh understanding like where the import sheet might not be formatted correctly as well. Chairs, somebody had to say, uh, yeah, that was great. Oh. Oh, I totally lost track of my my thought that um that that time. So, um yeah. Now, on on the other side of that, you know, sometimes it feels like flavor of the week AI. I know that feels that way at work a lot. I mean, the number of times I've like transitioned skills and automation from one tool to another tool and um actually creating a lot of work. not just for myself, for for other folks because there's no delay in the feedback you get. So immediately I'm saying, "Hey, can you review this slide deck that obviously I created it because there's no typos." Um, so how do we think um Julie, how do we think about tool adoption specifically like what do you pay attention to, what don't you pay attention to? Um, and how do we handle this just inundation of information that we're getting? We try to approach everything not from the top down. So I would venture guess that some folks here have been part of some AI initiative at their company where it was like we're going to use this specific tool. Use it as much as you can. This is your job now. You use AI. And I try really hard not to be that person, right? Um, but at the same time, our team has real pain points and real client work and things we need to accomplish. And in doing that, it doesn't always have to be this old way, this brute force of, well, let's get on Ella's calendar. Ella, by the way, is our engineer. She might be a witch. I'm not sure. She's incredible. We all want to be on Ella's calendar. Her time is limited. Um, but how can we enable ourselves? And when it comes to tools, where are we comfortable and confident working? And that may be different for different people on the team and different ways they work. So right now as a company, we don't have I don't know if I'm supposed to say this out loud, we don't have one dedicated tool. We say to the team, we'd like you to be using a tool. We're a Google Workspace or you have access to Gemini. Gemini works incredibly well for team members that are doing work with data and spreadsheets and all of those things where maybe it's not just cleaning up or doing this, but it's like inventing some data or mapping in certain ways or I have a client where on a monthly basis we've been manually mapping six different reports together while we build a bunch of integrations for them. Um, and that type of work Gemini is great for. My content team not so hot on Gemini. They're Claude Claudette fans. I like to think Claude is Clawudette. Um it just makes me feel really good about her. I say, "Hey girl." Um so they live in that world because it works so well for them. Um and then we say, "Hey, let's have this conversation. What's working well for you? What parts of your day-to-day are still problematic? they feel timeconuming and how do we do that better right the next thing on our list on that road map is we do some um reporting and analysis for our clients still somewhat like manually old school review what's going on why is this happening why is this happening the the five wise of GTM analysis right um and then we turn that into a month of deliverables for our client of here's what we're going to do next and then that has to get built out into the project management system and you apply the templates and assign the tasks and do the things and we have the structure and the framework and who wants to press all of those buttons because it's not me. Um, so the next thing we'll solve is how do we improve that workflow? Because once we do that, that's 15% of like two people's days or weeks. Um, and then they can do work that's more interesting, more meaningful, and more impactful for our clients as well. By the way, it means our prices get cheaper. So it's more about team how can we use this in your life than saying team you have to put this into your world and it's been working pretty well lead them to AI and they will use it right hopefully so I think probably not a surprise to anybody in this room but one of the common things that all of you have said is documentation being deliberate there's strategies there's frameworks there's like all of this, you know, um, adultting stuff that you also have to do with AI and it's not just all fun and games. I want to hear from each of you, starting with Andreas. What is one of the things that you see folks doing wrong most often? So I would bet that maybe not in this room but when you Google is around for two decades and I bet most people still don't use Google as powerful as it can be. It was just like smarter searches. So I start with this is or like googling stuff that's ungooable and then wondering that the answer is weird. But like the what I think is like very often done wrong is asking AI questions that are not the right questions to ask AI. And I'm just obviously simplifying the the example, but like hey like what's good for my company is just going to give you a very bad answer that reads very well. and like giving AI the guidance and the giving it the constraints that it doesn't go off the rails and giving it the guidance of what it needs to know to give you a quality answer like context is a very overused word sadly but I think it's really really important to use AI not just in chat GPT and having some memories in there knows you live in New York and like you're whatever years old and a man um but like what is what is your thing what do you want to what do you want to achieve? What do you want to make money with? What do you know? What don't you know? Like, um, these things are so often not happening. It's used as this oneoff chatbot experience, which is just like such a far cry from what it can do. But it's also really, no, I wanted to say really hard. That's actually not really hard to do, but you have to be very deliberate of knowing hell, what do I do? Like, what do I want? What do I know? What don't I know? That is actually not that trivial. But when you start there, it's like like and I'm not talking about your AI chatbot, but I'm just saying your AI chatbot is not my AI chatbot. When I ask cloth, I get like answers pushed back that look at second order questions that I didn't ask and it surfaces these to me because it knows where my weak spots. Um, and that is like incredibly powerful when you do it and I see it most of the time not being used like that with like casual AI users. I would say yeah I think for for in my situation a little bit different but also very similar is that I'm looking at it from is it a replacing an employee. I think one of the biggest uh fallacies and and and and uh myths that is out there is that AI agents that work 24/7 can run autonomously. You never have to check in on them. You can go sit by the beach and you know sip sip a cocktail. Uh I don't think that exists at all. Uh you need to make sure that your uh agent stays smart. Um unlike hiring an an actual person, AI agents don't come with any kind of expertise built in. You need to train that. I think it's very similar to what Andreas is uh saying in that you have to make sure that there are guard rails in place that they uh that they're tapping into the right information. I that's why I love HubSpot because that is the business context for our clients. So, uh I call it a dynamic rag for anybody that is into rag architecture. I don't know if that term exists, but for me, uh HubSpot is a dynamic rag. holds all the contacts o of your business, which clients are active, which clients have churned, um what deals are in flight. Um and um and yeah, we still need that human in the loop, but I think uh a lot of people expect it to solve everything and it doesn't. You need to stay in the mix. I once had an art teacher who every time we were working on a final composition said, "You cannot get the good paper." You've taken an art class, you know there's good paper. You cannot get your hands on the good paper until you sketch out that composition in at least five different ways. and he said the first one is not going to be it. You are not going to put that into the final 90% of the time. The second one, the third one, the fourth one are probably really interesting and the fifth one probably takes it a little bit too far. Right? So the point of that was there are so many ways you can solve a problem. The first way you think of may not always be the best, the most interesting, the most elegant. exactly what you're going for. But somewhere in iteration 2, three, four, five, you're striking gold and really interesting things are happening. Um, when we start talking about using AI, we don't always have to flex that muscle. And it is a muscle. We have to build it. We have to practice it. We have to be aware of it. Um, and the mistake I see people make most often is we go with the first thing. This looks good enough. I could use this. I could send this. I could ship this and shipping is a feature. A client said that to me once and I said, "Absolutely, shipping is a feature, but if we can do it a little bit better and it doesn't take a lot more time, shouldn't we try it one other way? What about one more way? We've done it three ways. Which feels best? Do we need to go for four? And what a better product we have now?" So, shipping's a future, but make some iterations. All right. So, we're gonna turn to questions from the audience here. um right now very shortly um they're going to ask the really hard questions. I was just, you know, teeing you all up with simple questions there. And um yeah, I mean, we've heard best practices, strategies. We've heard all of these great things about, you know, how to make sure the adage garbage in, garbage out, garbage context in, garbage AI response out. Um, and I'm I mean, I'm babysitting agents here. No, just kidding. It's Instagram, don't worry. Um, so I want to hear just rapid fire starting at at the North Leica. What are you most excited about in the near future? And and maybe you can take an opportunity to briefly t tell us about OpenCloud, but what are you most excited about? Rapid fire back to me and then we're going to open it up for three to four questions from the group here. Um I'm really so uh HubSpot just uh launched custom tasks or custom properties on tasks um which I think is absolutely incredible and uh it allows me to build human in the loop uh points within task structures. That's amazing. Uh and and I'm going to look into that. What I'm really excited about, I live very much in cloud code and like all these it's a lot of my work is about the organization of knowledge and and the processes that change knowledge that update it that substract that that add knowledge and a lot of that infrastructure is being built into cloud code and it makes working just on what I want to do and not so much on the infrastructure of that that's just I think it allows me to take on one more client than without that. That's gets me really excited. So, I love a HubSpot report and someone on my team just figured out how to take data that lives in HubSpot that's not accessible and combinable with other parts of HubSpot data um using data hub to turn it into a data set and activate it for reporting across the CRM. So, we can do that all inside of HubSpot without any other tools and I'm so excited to do it more and it's very very cool. Awesome. Isn't this panel great? And like we and and there was no please talk about HubSpot sort of thing. That was all from their own experience. But thank you very much you three. Let's take three to four a couple questions and then we'll let you all get back to the food and beverages and you can come up and talk to our panelists. Um but anybody have a question? Make it really hard. Andre, go ahead. Yeah. Oh, yeah. For whoever wants to take it. Um, how much are you using HubSpot's native AI tools versus like flaw and then Squad connector called Quaders? And I don't even know if there's another part AI that connects to HubSpot other than quad, but there you want to just explain about that. So, uh, this is I love HubSpot. I try to consolidate all my tech to put it into HubSpot. We use the projects uh object. Um and I have been a part of some of the beta features for the breeze agents. Um I decided to move stuff over to OpenClaw just because I was able to iterate faster. So one of the things that I trade for that is security. Um, HubSpot has built the pre breeze agent to be very secure. It's limited for a reason. Uh, for me personally, that wasn't enough. That's why I ended up in in OpenClaw. I wanted to be able to just push it a little bit more, push a little bit faster, have more access to the API. I know that HubSpot has recently announced that they're going to open things up a little bit more as well, but they have uh they have a you know this HubSpot they can't just roll out something that's not secure. I don't think there's a lot that we can talk about. You know, AI security is a whole another section. What is insecure about? Uh so, so for example, in OpenClaw, if you scrape the web, currently uh there is malicious code on websites that are being deployed. If your agent ingests that, it can mess up your system. So, HubSpot is coming from a position where they can't have that happen at all. So they have to lock that down. I appreciate that. It just means that they iterate a little bit slower. They have to lock it down. Same with emails. Ingesting emails, you can have malicious code can mess up your your AI agents. Thank you for the question. Let's Oh, nice. This is great. I we're not going to unfortunately be able to answer everybody's question. Um I I'm one two more. Um sir, you had a question. Um, what are some of the top use cases that you found to school uh for your favorite use cases you found for RevOps and um I guess just small businesses as well? Yeah. So, this is top RevOps and small business use cases. Who wants to go first? Oh, I'm putting you on the spot. I I fear this is boring, but still kind of cool. Um, one of the things we do with our clients in the property management space is we bring in their listing data to HubSpot. So, their current available properties for lease for rent. Um, and we're able to use chat, AI chat in some different ways because of it, right? Because we can tap into availability, price points, different markets, etc. Um, which is not the coolest thing you can do with AI, but it's very cool for them because property management companies try to run very lean. It's not a high margin business most of the time. So the more they can have like the easy stuff answered by an agent and then hand it off to someone to who can actually coach you through finishing your application and the information you need, the more we can get documents collected in an efficient way. And to the extent that AI can help us with those things, the more their team can do kind of the the harder stuff and focus on being closers instead of engagers. Um, and that's a, you know, I think you can apply that across the board, but it's I don't think very cool. I'm sure Leica, you have something cooler. Uh, it's, it's one of the reasons why I don't have claude co-work agents. Um, that's why I also went on open claw. I believe that a lot of building that's happening with uh, agents being built on individual desktops, that means that data is not openly sharing to other parts of the business. Uh if you have OpenClaw, you can interface with that uh through Slack. Um so your team can interface with agents through Slack. Uh and then on top of that layer, you have HubSpot um as as that dynamic context as well. uh a very concrete uh I believe pure revops use case that was just impossible to do at that quality and scale before is like all these uh messy situations of like a lead coming in a contact coming in like which entity does it actually belong to this is like not a solved problem um and the the extent to which you can dial that in on the unstructured information about the world what is this person's thing what is even their actual job we care about when they have like three opportunity or three uh things on their LinkedIn whatever deciding that contextually and using that and then decide where in which of the entities that all have the same domain but they are different business entities the company sells into. You can solve that and you can't only solve it with like okay let's put it into the property with the most contacts or the most recently contacted or whatever stuff like that that you could do in code but you can now do it with semantic understanding of where does it actually belong and that also the whole elaboration to get there the process to get this little thing in this huge data pipeline right is like I'm hyperventilating it's like super amazing um and it's an actual problem right it's Yeah, and I'm by no means a RevOps expert myself. That's why we bring them in. But one of the use cases we see a lot is prioritization for reps and also helping reps um open the door. It's like it's it's really hard to start a conversation that's lukewarm, but AI using AI to find um compelling content or uh compelling talking points to start a conversation or restart a conversation that's um started to go stale. Okay, another question. Did you No. Okay. Gentlemen, you um yeah, two questions. One, especially for you because I used the open claw. How do you manage costs and how do you scale that across different clients? Then the second one is for everybody. Um over the last two months like I think everybody's seen u flawed like pulling back saying like hey the you know monthly subscription is not going to count for Oprah so that gets cut back. Then the outages now space X invested there's a ton of money into it again. I've been trying to like diversify, right? So like using Codeex where I can um using Claude, I've got like Quinn. But I'm just kind of curious like one the cost perspective and two like as we go forward how reliant are do you think we all should be in terms of these frontier models as they start to get like more walled gardens and like how are you experimenting with these other models to try to have use a tool like open router to say like hey I know you do this at clay I'm trying to conserve cost I don't need 4.8 right I'm going to use clay. Yeah, it's a great question just repeating um cost management with openclaw and linka if you wouldn't mind just giving like a quick definition of what openclaw is and then the next one is kind of removing yourself from the proprietary nature of getting kind of tethered to an LLM like that that's something I think about a lot. Yeah. So open claw is a harness uh as as is code co-work as well. Basically that means it's a very fast way to create AI agents. My open claw is hooked up into HubSpot. Uh cost is definitely a thing. Uh I learned the hard way when uh you know Opus was uh just released. Um it's also one of the reasons why I don't use or am not a big fan of MCPs. As much as I love that everybody is creating MCPs, I the way that I structure the agents is to break it down into SOPs. So each agent has a job to be done. What is the SOP? And then we do API calls and that allows us to be more efficient in the context uh breaking up that that context window because open claw can run on the longer a session runs the more expensive it becomes. So that's where the architecture comes in uh to make sure that it's it's doesn't go crazy. Um I've I've gone I've done the locals that are free, but I've moved back into the to the APIs for the LLMs. Uh and I am paying for it. Um the cost that I incur right now is much much much less than an employee, but it's worth it more stable. I get better benefits out of it. So and the next part is the dependence on a specific LLM. So uh the dependence is a very multifaceted thing. So the question is how do you make sure you don't have vendor lock in with your with your LLM with your AI and I think it might be very specific to the work that I do that is not like I don't produce like per se software applications right where that might be a little bit more ingrained but like especially for the GTM stuff everything is quote unquote just markdown it's so transferable it's kind of the opposite of vendor lock in when you build first the system of what is the purpose why are we doing this what is the thing we are doing here what are the steps what are the constraints what's the guidance and that's a markdown file and you drop that into another LLM and honestly they are converging so much in capabilities I just use claude if they like implode tomorrow I can switch over to codeex um because the gold is not the actual execution of the LLM but what do I tell it to do in like a for a prompt or an agent or whatever so I think that's actually a very very good development that when you document doument well and you document AI ready and consistently which is what I talked about that that is being built into cloud code that you have to build less of that architecture yourself that's very powerful because basically drop a bunch of markdown files into another LLM say figure it out I want to pick it up where we left off and it's going to work reasonably well of course it's going to be a headache but like it's a possible solvable headache that's what I would say I agree Hey Andreas. Oh go wow. Jeez we are done. No more questions. So we we will do one more question but I did want to comment on that even more because um you see third party tools like versel that has the AI gateway and the AISDK to solve this specific problem and they end up becoming like kind of like a broker for LLMs. And my first question is are they marking it up? and they have a very specific article um about how they're not, but it is a really important consideration um to avoid that that vendor lock in. Um so, let's make sure we get one for Julie because she's just trying to get out of it and then so one more question and then we're going to um move on to social time. um feel free to come up and conversate with our amazing panelists but sir what is your question also a two question first thank you for being here so first part of the question is so first there should definitely a human rule because you've got like first is very important since there should be some people so I'm curious like um AI is really tolerating it's able to give us a lot of information I'm curious just how to read like judgment so we could make the most out of the information so that we could have a good motion market plan that is going to overdo the burst. So that's the first part of the question. The second part of the question is that how do we like um how do we kind of like improve it so that we don't have a skill that AI can use in the future. Ah. Oh wow. The second one is so um if I if I tell me if I get this wrong. So the first part was really the human skills so that they make smart decisions about what they're receiving from the AI and then acting on that to make effective gotom market decisions. Um and then the the second part is really also kind of around skill advancement. Um so that we we are not consumed by the AI that we build. We are more working with them and like Leas you built a business with the you have a workforce around agents and humans. So yeah those are both really powerful questions. Who wants to go first? Uh I think good judgment is just generally tough to teach. Uh anybody who has had kids knows this. Um, so how do you how do you build good judgment? I I I I think uh we're going to head into uh an era where the human aspect of life is going to be really important. Uh EQ is going to become important. I think these are the things that are going to differentiate companies from one to another. So uh you know we uh yeah good judgment. uh we need to make sure that uh we have it and um yeah, I'll leave it at that. I I would say for me it's being very aware of my blind spots where it's where I wouldn't have this this intuition. Oh, wait, wait, wait, wait. This sounds really nice, but it's actually pretty wrong. Uh so that I'm aware this is not my area of expertise and I kind of don't trust it on a default. um or where I don't trust I'm building in the right direction. Let me think about like very adversarially is this even the right thing we are solving. Um so taking that step back and the the second question is how to not get afraid of becoming redundant because basically I'm teaching what I'm doing today to a large degree will be in the world knowledge of LLMs and it will be able to do much more complex GTM stuff that I still have to manually wire together now and it will be pretty good at that in a year or two. But what I'm seeing so first um I have no right for the word to stay just as it is because I'm very happy it went to this point and I have no right to ask oh but now don't develop further like now you're far enough like let the rest be to the humans right so and honestly you have to be very flexible and on your toes on how to develop yourself what I thought and the last thing I'm going to say to this two three two to three years ago I thought I just had to be super amazing at clay and all these tools and I would be like really sad. That's not at all the case. But now the work that I do levels up into higher order things. Now CEOs I work with asked me, do I have the right team? What who do you how do you suggest to structure my team? So I got to like grow into that. So the the line of what value you bring to the world and to the market is is always going to shift and it's going to shift upwards. And I think when you're just aware of this, you're less afraid of it. That's what I would say. Um, yeah. So, once I had an art teacher, No, I'm kidding. Um, um, I think that the use good judgment message for me is really that same message of you need to have a critical eye. And that's so hard to teach. Um, and it's going to take us back to humans, right? It's back to those conversations of I'm solving a problem. I've got my buddy Clawdette with me and it's us against the world, but we also have colleagues. We have a network. We have clients we can talk to. We can talk through these things. And just because it feels like me and Claudette against the world doesn't mean it is and doesn't mean we can't talk pros and cons. And I think as a oh I hate this phrase I'm saying as a business leader part of my job is to make sure teams have the the psychological safety to have those conversations about like I did this thing I'm working on this I need to talk through it because I have uncertainty or I think it could be better or is this overtaking where I think it should be. Are we doing too much right or is it just enough? Um, and I think to Leica, to your point, like bringing humans into it and thinking about that element is also critical. What I love about closing out with that is because I'm hearing a lot about just being really healthy at having good communication, being open, being curious, which I think leads to companies are hiring different and they're going to value emotional intelligence. They're going to value these things a lot more than they ever did. So, thank you everybody for joining us. Thank you panel. SJ wants to You want a drink or you want to share? I just want to take a moment to thank all of you for being here and to a huge round of applause for our awesome panelists. They were so great. Thank you so much.