Time to Talk about AI and How you May not be Using it for...

Sa.

Good morning LinkedIn friends, Value First Nation welcome to another episode of Value First AI Daily Ear Collaborative Intelligence Report. It is Tuesday, February 24, 2026. Nico, George. How we doing fellas?

Doing good, doing good, doing very well. Cannot, cannot complain, sir. How about yourself?

I am excited. George and I had some fun yesterday talking about what we had built over the weekend. I was planning to talk about what you had started finalizing over the weekend, but we're here today to now you get, you get the whole stage to yourself, man. Which I, I'm excited for that because I feel like you're an OG when it comes to really building apps. Like with like as soon as custom GPTs came out, right, like, because we're going to talk about Hubbel perhaps today and I would love for you to start there about like, you know, one what that foundation looked like and how you've kind of thought about it as we get to where we are today.

Yeah, I mean it really, it was kind of a nice little jab and it was just a joke really. Kind of in the beginning of like, oh, hey, let's just see what we can do, right? Like when OpenAI was announcing that they were going to have custom GPTs, I was definitely geeked about that because it's like, oh man, it couldn't get any better than having your own like personalized resource in this way. And especially for HubSpot, it just made sense to have something that was just a very easy like hey, let me just chat to this resource to get the exact answer I need right away so I don't have to go like diving through pages and pages of stuff and try to sift through what the actual right answer is. And it started with just that really as soon as custom GPTs came out, just started very manually going and collecting data and going and doing a lot of manual scrape and figuring out how to write scripts to try to help with that. And it was a very, very massively messy process in the beginning of just trying to get it up and running and performing, you know, even, even to get it performing mildly better than chat spot at first was like pretty challenging. And that's where like a lot of the prompt learning and engineering and all that kind of stuff started coming into play was learning how to actually build those instructions and learn how to constrain a model that was already pre trained on something how to actually like drill down and focus on the data that I had to pull.

And this was like middle of 24, right?

Yeah, yeah. And yeah, it was, it Was great to be able to go head to head for a very, very long time chat spot finally upgraded to a Breeze and it's been an interesting challenge along the way. Not so much because of having to keep up with the information, but it's been like a forced learning process too because as the underlying models change I have to go back and change a little bit about the behavior. I have to change something about the instructions, maybe change something about the way it calls files or whatever it, it forces you to have to go learn about like the new, the latest techniques for the different models that are available whether you want to or not because you have to make sure that your chatbot is going to behave appropriately. So yeah it's, it's been a pretty big challenge over, over the, the year or two. Didn't want to say years but it's been a pretty interesting challenge. Just trying to keep up with HubSpot's data dumps is very challenging. But to also just keep up with the models themselves and the behavior and trying to make sure that more right answers are coming out than wrong answers when you don't really, you don't have control over temperature settings. You know, getting into, going from custom GPTs to then learning about like okay, here's how like actual assistants work under the hood. When we go into like the actual control panel in GPT and seeing that like oh there's you know, vector stores and all these other tools and fine tuning and all this other stuff that, that I could be using but I'm constrained to this, you know, platform and I'm constrained to this custom GPT thing and even though they actions and you know, all that kind of stuff, it's made it better but just not quite, you know, everything that I have wanted it to be, let's say. So yeah, it's, it's been an interesting journey. I'm really glad to say that Breeze and Hub Helper Harry are kind are pretty much neck and neck now. At least the GPT version and the in app version for Breeze.

Yeah, I think it's possible you might actually be running on a better, on a better model. So that makes sense. I remember like the first time I realized like this, watching somebody else solve their issue with Hub Helper Harry is when like I first realized how like where we're going in terms of just democratization of capability and not having all these barriers because it was in sprocketeer like 2024, I was in Sprocketeer like Slack like daily, right. And I got this question About, I think it was about reporting with Operations Hub or something like that. Like, somebody just direct messaged me and I said, you know, not really my. Not my expertise, not my strength. So go like, put it in the channel for, for Ops Hub. And like, if you put it there, then I'll tag some people and they can chime in and come and answer the question for you. And I think that might have happened. But I got a message 20 minutes later from the person saying, no, I just went over to have Helper Harry and problem solved.

Right.

That was back in 24. Right. And I think it's like, if you're in the ecosystem, especially if you're in our echo chamber of, you know, consultants and agencies, like, you don't have to look very far for people that have, you know, gotten help from Hub Helper Harry. So. But you mentioned some things weren't the way that like, you wanted them to be. Can you speak to that a little bit?

Yeah, it was really the case where like we, well, I say we, my team at the time, we were looking for like, again, like an easy way for us to be able to access information that we needed to when we're with clients or like double check ourselves or whatever the case was. And using Chat Spot was where like I thought everything was going to change and it's like, oh, cool, now I have this thing in app that I can use in order to just reference whatever I need to. And it was just, it just felt very unfortunate because it's like, huh, okay. That doesn't seem to be the case. Actually. Every time I ask Chat Spot something, it's like 8 out of 10 times it doesn't know. And I'm asking it very basic questions that are like default HubSpot things that, you know, should be able to answer that, no problem. So it's a little weird that like it's getting tripped up on it. And after that happened enough times, it was mainly like, that was probably the drive. The, the biggest driving motivator was just like, hey, the, the default tool should be better and it's not. So I'm just going to go make a better tool since that one isn't. And yeah, just. It started there, kind of got out of, kind of got out of hand real fast because I started just grabbing every and every, like any and every resource related to HubSpot as well and Salesforce information integrations information, you know, just. It started with, hey, let's try to make a great HubSpot resource to help people with HubSpot. Issues and it turned into like, hey, let's just make like a really great HubSpot centric martech resource that people can use.

So as you've, you've been talking about transitioning from a custom GPT for a long time while also getting fairly good at just building whatever app comes to you in the moment all outside of, you know, custom GPTs. What does what through that process of learning how to like build apps and just all of this development activity that's been happening, things we talk about almost every day at this point, like where does that take Hub helper Harry?

Sorry. So like where does the modification or like the ability and the, the future modifications or whatever, where is it going to take them?

Like how are you thinking about what like standalone, you know, not attached to specific GPT or you know, anything like that version?

Yeah, there's a. So like so there is a standalone paid version that's just, it's on its, it's on the doorstep that people are going to be able to use very, very soon that is anthropic based. So it's not sort of kind of moving away from the GPT platform for a multitude of reasons, but primarily for data security and safety. And outside of that, yeah, it kind of turns. You know, it's one of those things where you, you thought that you already had a diamond when you were looking at it, but you didn't quite realize that there was something even more special underneath. So in creating this, this newer version of Harry and migrating him over to, to Anthropic, there's, there was quite a bit at the core that I kind of had like glossed over in terms of like, oh, it would be awesome or it could be awesome to use him for like these different outputs or use him for these different applications. I would get into it, but it's going to be some like front screen stuff that when you guys do see it, and I didn't realize that outside of like not just HubSpot but like outside of marketing what the potential was going to be leading up to this year. So there is definitely something independently coming that has Harriet the core that is definitely going to be a major, I hope, I hope it will be a very major disruptor for the plugin ecosystem. So it won't necessarily be for like SaaS. So like don't, don't get too scared but it'll definitely be like a plugin ecosystem type thing where like if people were going to be developing that, it will likely be Harry doing that for you. And yeah, there's what I look forward to is the fact that like, this is only the beginning. So like I was, I was reviewing it last night. It was just one of those things where it's like, ah, you know what, maybe we take like another couple extra days here because there's like this, this aspect of agent, of like agentic functionality that's going on now where we were talking about ourselves like putting together like our own little internal malt bot and that my brain kept like jumping back to that in terms of like, oh, wait, but we're not going to like pigeonhole ourselves into this too quickly and then we're not going to be able to like, pig of it later. And so I was like, all right, well there's definitely the space for the chatbot for like people to use for reference and stuff like that. There's obviously a space to look at in terms of coding and CMS and all that kind of stuff. But then there, you know, there's also another aspect of a problem that I've been hearing people talk, talk about a lot lately, right, which is like your own CRM management and like how you would CRM customization. So there's like some other larger stuff that's coming down the road too, but I see it as being like a multi. Being able to facilitate multiple aspects

for

like all levels of users in a lot of different directions. It's a really cool thing where my partner and I were just talking about the fact that a lot of companies seem to be staying too narrow in a time when you have the ability to go wide. Right. And still have control over it. So I'm looking forward to it. I know it seems, I know this stuff seems very generic. It's tough to talk about specifics before you actually release anything. But essentially what I'm trying to get at is that like chatbots are definitely going to turn in and you know, they're already in most cases, but if they haven't, will be very soon, just assistance. And then assistants, which were already like kind of a big thing or kind of, you know, you could think of them as offices or groups of agents at this point. And then of course you got your, your software build outs. But I think, okay, I can't, I can drop a little something.

Do we get a.

Just think of. Yeah, well, yeah, I can definitely show you the, the, the chatbot, but I would say think of it like, think of it like maltbot, but it's iPass.

Yeah, that's all right.

Like, because like, why not like when I thought about it. I was like, you know, it kind of seems like if, if you just stripped it all away, it's really just, you know, dude, from, from mail room, taking stuff over to the other place. Right. And then bringing back the response or what have you. So it's like, well, I'm not really sure why you need IPASS altogether. I think you could just, I think you just create a resource that can actually just go talk to the other resource directly as opposed to. Yeah, so it's like a nice little door knock feature too.

Yeah. I mean, it's kind of like what I'm hearing and especially when Dharmesh posted about Hub code, the, the cloud code IDE experience that we've been enjoying, like, I think is on the table for any. Anybody building stuff like this. I think that's what you mean as far as going wide. Like if you have, you've established the vertical expertise, right. Like, why stop at just like helping answer the question.

Right, Right.

Like, it's like one thing to go get knowledge help from Harry. But that sounds like it's, it's possible maybe in the future to say, okay, Harry, like, sounds good, go do it kind of thing, right?

Yeah, it's, it's definitely going to be the case where it goes beyond the chatbot. Right. And I think for, for a lot of Hub Helper Harry users, this kind of helps to bridge that educational gap too. Like, takes people from the seat of like, oh, I was just using it for chat to. Oh, wow. I can use it for this too. Because the, I think one of the things that was like, really most important to me about building this out was helping people with that transition. And so a lot of the like, suggestion chips that you get afterwards instead of like, some of them are follow up, but a lot of them are like, you know, if you're actually doing something with code, it's like, oh, build this out for me. Or like, hey, can you, you know, can you ship this? Or like asking additional direction. I'm like, okay, well, how do we, how do we actually go about implementation? There's aspects of attempting to try to keep it more realistic. A lot of people's thoughts and comments came to mind when working on this. The whole AI native aspect of things. I think one of the bigger challenges still, because it's not, it's not quite what I wanted. It's not quite talking the way I want it to just yet. The biggest challenge has been getting it to something we were talking about earlier last week has been getting it to Shake off trying to use like really old school academic, like marketing stuff, right? And trying to like, trying to keep it from leveraging too much. Like corporate really, like the old school stuff that, you know, either doesn't work or isn't relevant anymore or is just tried played out. I think that's been the biggest challenge is getting it to understand, like trying to get it to fight against that norm and to inject the reality that like, hey, timetables have changed. The way that we work has changed. Thinking a little bit after that.

The training, AI collaboration, right. And like the way it's built, like the math behind it, like, makes sense why this is such a challenge to overcome the training data. But I'm curious George, when you hear that as far as like the HubSpot, specifically hub helper Harry, that not only can, can help answer questions but can actually execute on those answers,

like, where does my brain go?

Yeah, I always want to know where your brain goes.

Well, there's a couple things that I'm sitting here thinking about. One was like, and, and I went kind of just as far as timetables. I was like, man, I would just set a hardcore rule of whatever time you think it's gonna be. Divide that by three and then say that instead, please, because you're, you're way off. Here's. So I'm actually, I'm battling with. I haven't seen anything and I'm a visual guy, dang on it. And like, I want to see it, right. So I'm battling with that, but I'm listening to like everything that's been said. And, and the part that I'll say that I've really been enjoying is the talking of the journey to get where we're about to be. Even though Nico's doing a really great job of creating this like, canyon of. I don't know what the f he's talking about, but it sounds cool. Like, you know what I mean? Because like most people, by the way, you said, imagine this, but I pass. And you guys are like. And I'm like, I don't know what the that means. Like, what, what? Like what what? And so like I'm, I'm hanging on for this ride and I'm enjoying the ride and I'm like, well, what can it. And then also the other part of me is like, well, doing like, I've, I've, I've. I shouldn't even be thinking about that. But like, I've got doing down. Like I built two things in the last three days in HubSpot. That your boy George B. Thomas could have never built in HubSpot historically. But it was like, I'll use the word easy. But then I temper myself and I go, well, maybe not easy. It was fun and challenging at the same time, therefore it felt easy. But I, but I sit there and I go, now hang on a second. Take me out of the equation. Listen to the story that's been told. Think about the canyon that's been created. And for most mere mortal human HubSpot users, this might be like the unlock that makes their day not suck. So that's kind of where like I'm, I'm, I'm wondering if something is about to be released that is so helpful that like you could enjoy your job again as a HubSpot admin versus feeling stressed to flip out and not, I know not all admins are administrative stressed out. I'm just saying some things that I've seen in the ecosystem. But yeah,

I think, whether I think whether I want to or not, I'm actually going to launch tomorrow and just deal with it and cobble as I go along. I think I'll give everybody seven day free trial so they can try not to make me go bankrupt while having a lot of fun. And I will use that time to do some refinements and whatnot. But yes, so far, here's what I can say and I'm actually, while I'm a little scared to say it, I'm actually also very excited to say it. There's been a really, really, really good amount of beta testing going on in the background up to now and Harry hasn't gotten a single thing wrong yet. So really, really looking forward to it. Definitely now that I even say it, want to drop it tomorrow and super, super excited for everybody to get their hands on it and try to break it as much as possible.

So does that mean we get to unbox it tomorrow?

Oh, yes. Yeah.

Hell yeah.

Yeah.

Wow.

All right. I will say, Nico, I'm glad to hear it has nothing to do with what you're doing or what you're gonna drop, but I'm glad to hear you say what you just said because I basically the entire time the thing that I showed yesterday was this like massive page 2 downloadable campaign, like 10 purchasable like PDFs like, and everything that I've been building on the spiritual side of leadership.com has been built in the open. Like there have been spaghetti stains on the T shirt and there's been a rip in the jeans and but like it was just like, no, I'm gonna get it out there, and it's. And you're gonna see ugly, and you're gonna see good, and you're gonna see it mold, and you might hate it, and you might come back. You might. But, like, it was literally just working in the open, and it's. It's very interesting when you allow yourself to have the freedom and flexibility to understand that people can come along the messy middle journey with you is. And you don't have to control perfection.

Yeah.

Like, it's. It's a. It's a nice place to live, especially with all of that we're doing and talking about and being able to be flexible and nimble and, like, ideate and creativity and listen and launch and, like, it's just really important to live in that space.

I do.

Yeah. Well, I'm excited to unbox it with you tomorrow. So if you're excited to see Hub Helper Harry, is it like, what, 2.0, 3.0? Does it have a point zero?

I guess he would be 4.0 at a version. Yeah.

All right.

If he had a version.

Let's go, everybody. Have a great day. We'll see you tomorrow. Thanks.