Loading summary

Transcript69 lines

[00:01]
A
Welcome back to Limited Supply, the podcast where we get deep into the tactical and strategic side of e commerce, digital marketing and building consumer brands. I'm your host, Nick Sharma. I've spent the last nine years building, scaling and investing in brands. And through this show and my weekly newsletter at Nick Co Email, I'm here to share everything I've learned. The wins, the losses, the experiments, the tactics and the insights. All so you can unlock your next hundred thousand dollars in revenue. Today's episode is a good one, but before we dive in, let me tell you about our channel chosen sponsor for this week's episode. If you're a DTC operator and TV feels like a black box, that's because most platforms give you 20% of the actual available inventory and none of the measurement to see if it even works. On the other hand, with Tatari, you can buy linear and streaming directly, plus you can measure it like a digital channel. That's the reason over 400 brands like Manscaped Toqovas and Chime run performance TV using Tatari. See how they leverage the platform at Nik Co Tatari. That's Nik Co Tatri. Welcome back to another episode of Limited Supply. I'm your host, Nick Sharma and today I've got a very special guest with me and he himself is an AI expert and somebody that I've learned a bunch from when it comes to AI. And you know, one of my goals here is to bring constantly whatever I'm learning, bring it to you guys and in a way that's easier, easy to digest and easy to understand. So today I brought on a friend of mine, Jacob Pozol, who I met recently and have actually learned a lot from him and the company that he is building, which is really around creating kind of an operating system, an AI operating system for your brand or your company, your organization. And so we talk all about, you know, how to get on with AI, how to start using it as a company where there's onboarding failures or what succeeds when you're onboarding and just really like, how are different people within the org also using it to their advantage? So check out today's episode, let me know what you think. Shoot me a DM on Twitter. You can also DM Jacob if you got any questions for him. And I'll see you in just a moment. All right, Jacob, welcome to Limited Supply. I'm excited to have you here today. You've been all over the Twitter feeds. Honestly, we. We never really connected until recently either.
[02:24]
B
Yeah.
[02:25]
A
But we've been Twitter friends and Obviously, you know, we've seen all this, everything on the timeline, talking about hq. So I thought, you know, why not just bring you in and let's talk about hq is also somebody who's now been a user for a couple months and started to learn a little bit more about what you guys are doing. I thought it would be really cool to share what you're doing and how, you know, other brands and other operators can leverage it.
[02:47]
B
Yeah, I'm excited to be here. Thank you for having me.
[02:50]
A
Of course. So why don't we start with, you know, kind of like the one thing that I think will get people excited about Hearing more about HQ2 is what, what happens, you know, when a whole company starts using Claude or ChatGPT, and what doesn't happen, you know, like I was at, I was at a dinner recently and somebody was telling me about this new logistics tool that she built in the. In Claude, and unfortunately, her direct reports can't get access to it because she built it in Claude. So I want to start with that and then I want to go into more of HQ in your background and then let's talk about how different functions are using AI from what you're seeing as, as their clients of hq.
[03:35]
B
Yeah, yeah. I mean, I've seen. I've seen the insides of lots of businesses as they try roll out these tools, and it always results in this extremely weird fragmentation, which is obviously what we're trying to solve. But there is this extreme gap in certain individuals in their ability to pick up the tools. And so you can give 50 people Claude or 20 people Claude, and the distribution of how they would actually end up using that tool is completely different. And so you'll have the person like you just described there. They'll build something great, they'll build something cool, and they'll have no way to be able to share that with their team. That often results in a huge amount of duplication of workflows and of work, which is kind of the opposite of the efficiency that you hope to get with AI. You also have this, like, weird FOMO that happens where people start to get scared and they, like, see other people accelerating past them, they can't quite catch up. They get scared of the tool when it's not really necessary. And because it's so new and you kind of have to go to the edges to figure out how to use these things, you just. The differences are, like, are very wide. And so what I've seen happens is you try and you roll these out within a company, but you Never really get that nice compounding benefit that you expect. Like, you get people being more efficient and more effective, but you don't see it moving the needle for the business. But I think that's an infrastructure problem. It's not an AI problem. And so that's obviously what we have tried to solve with hq.
[05:14]
A
Yeah. And are you seeing companies becoming AI native? I actually, you know, like, obviously I feel like the answer is yes, like when. Whenever I ask that. But this. This past weekend, I met up with a friend of mine who's a YouTuber and I just randomly brought up AI to him and he's telling me that, like, from what he sees, you know, among. Call it normies, not us, you know. Yeah, podcast listeners or people, I guess, whatever. He said that more normies are like very anti AI, and so I wonder if that's trickling its way into the business world outside of our bubble. Like, I'm curious what you're seeing.
[05:53]
B
Yeah, I haven't seen businesses that are fully AI native and enabled unless they're very recently founder businesses or you have. You can have majority AI enablement, but that has to come from an extremely strong CEO founder figure. I'm always surprised by the lack of AI proficiency that I still see. The tools have gotten easier and gotten more powerful, so it's easier to be good with AI, but you don't see. I'm so surprised. Most of my circles, I try. When I try. When I socialize with friends, I try not to talk about AI. It's my only one time that I can, but it's obviously not possible. But yeah, I haven't seen that same penetration outside of the, like, Twitter sphere that I live within.
[06:40]
A
Yeah.
[06:40]
B
But again, like, the. That sphere is mostly founder, CEO types, you know, like, even within their companies. It's not the same. It's not. It's not. It's just the people at the top that are really trying to push it because they're the ones who experience the ROI from it.
[06:56]
A
Yeah, I feel like too, like, that is. Yeah, that's actually probably what I've seen too. There's usually a few people in the org who are super AI pilled and they're trying to bring everybody else with them and kind of get the org set up. Okay, I want to go into your backstory a little bit. So you were in engineering and then you sort of made your way into the marketing sphere or the kind of like the tech sphere then, you know, and I feel like your journey with AI probably evolved. I'm curious like what, what did you think you were getting into at the beginning versus, like, where did it evolve to? And like, how did you guys get to hq?
[07:39]
B
I mean, my AI journey started very early. I was. I was doing a. Basically an AI machine learning master's degree when, well, I graduated right before ChatGPT was launched. And so I was already in the machine learning space. I got a job as a software engineer and data scientist at an E comm logistics company. And I was helping them build basic forecasting models, like sales forecasting models to help forecast demand within the warehouses. And then ChatGPT was launched and I basically just quit my job instantly. I was like. Because I think I saw the writing on the wall a bit sooner because I saw how powerful this was. And so I literally instantly quit my job and try to start an AI company. And I was trying to build an AI logistics company for E commerce, which is very different to where I am now. But it just wasn't that powerful back then. And I think you could always see where it was going, but it wasn't quite there. And then in a roundabout way, I got connected to Common Thread Collective with Taylor Holiday. He kind of took me and took me under his wing and I worked with them for a while. And again, like, we always had these super ambitious visions of what we could do and what we were trying to do. And it felt like we were always waiting for the next model to release. Like, okay, now we're going to be able to do it. Now we're going to be able to do it. And then that actually happened, like around December last year, that actually happened with like the Opus 4. 7, Opus 4. 8 launch. And then we had this amazing moment. We were able to just do everything we wanted to do, build what we wanted to build. And I kind of just went crazy just doing as much as I possibly could. The way that HQ came about was Corey Epstein, who, who I was connected with again also through the same, like, Twitter world. Him and I had been in touch for a while. He had actually created this personal tool called hq. At the time, it wasn't a team tool, it wasn't a company tool. It was his way of being able to manage AI personally more efficiently and more effectively. And he had started converting all of his friends onto it. So that's how he got me onto hq. And I was like, corey, this is sick, but this needs to be a different tool. This needs to be for companies. This would be so much more powerful if all those context could be shared across companies. And that's how we teamed up, co founded this company and now that's how we have hq. And now it just feels like we're able to do whatever we want. Like it's just, it's this unbelievable feeling. And now we're building infrastructure because we've experienced so much across so many different businesses. I've seen so much across so many businesses through that journey. So now we're just trying to build the best infrastructure for companies to run on.
[10:19]
A
Yeah, that's pretty amazing. And can you explain like to a fifth grader level person how the product works?
[10:27]
B
Yeah. So when you have a whole company using Claw, using chatgpt or using AI, what happens on my chatgpt is just on my computer, it's in my personal account. What happens in yours is just on yours. And we have no clean and easy way to share that with one another. And HQ basically connects all of our different AI tools together so that we're operating on top of shared context. So if I do something, I can share it with you. If I build a workflow, I can share it with you. If I connect to a new system, I can share it with you. And so that allows us to work together, that allows us to compound what we're doing across an organization and it allows us to get those people who are lagging behind to be on the same level as us. Because if they just use ChatGPT connected to HQ, they're automatically inheriting everything else that we've already put in there from the rest of the company.
[11:20]
A
Yeah. I also describe it as like Dropbox sync across the company for all things AI related.
[11:27]
B
Yeah, yeah, that's a good one. If you. There's a good way to compare it to other companies. It's like Dropbox for agents, 1Password for agents and then like agents for agents kind of thing as well. But yeah, those are like a good, good ways to think about it.
[11:44]
A
Yeah, I use it to basically just stay up to date across all devices and all team members across all projects. It's pretty amazing. And the other thing I think is cool and not talked about enough is your memory system is like the way it's built. Is it called vector memory? Is that right? Yeah, yeah, vector memory. So it's not just reading markdown files, it's like a whole different way to recall memory, which just makes all the context and makes the output so much better.
[12:13]
B
Yeah, like it's really important that. Let's just say you built that logistics tool and I started building a logistics tool. It's really important that Claude is able to recognize that you built it first and to find it and, and to do that we had to give this, create this company wide memory which uses some really cool searching and retrieval techniques that we've built.
[12:33]
A
Yeah, very cool. Now I feel like a lot of companies right now are popping up with their or you know, like this. Everything started in the chat window and then, and then a bunch of other softwares have integrated models within what they already do. So like you know, influencer searching software now has a model to help find better creators maybe. And then it's going toward more like open source software where you plug your models in and you, you know, have your thing work however, which is probably where it sits with like Claude code and Codex. And then I feel like the next version of this is maybe. Actually I don't know what the next version of this looks like, but AI employee is starting to pop up now a lot more and I think because the quote unquote agents are easier to visualize or build or create. Is that what you're seeing right now?
[13:23]
B
Yeah, yeah. I mean it's two, there's two sides to it is definitely the Claude code Codex, like the Harness agent agentic tools for slightly more technical people. But then the AI virtual employees are really, really easy to adopt and understand for less technical people. Because if you give it a Persona and you drop it in slack, then people are able to pick it up easier. So I'm seeing that a lot, especially with like the Grokbot launch that came today. Like there's a lot of these companies that are doing really well. I think it's a combination of the two that to work though, like I don't think one works without the other. You have to connect the two together.
[14:03]
A
Yeah. And how does like a, a company when they are getting started on all of this? Right. Like I felt like with my own AI setup I spent definitely a month or two intentionally trying to create knowledge files or, or you know, basically ways of, of where I would be different. So like if the prompt was you know, build a wireframe for a landing page, my personal one would be way different than like if you just ask generic chat GPT. How does that whole process of like creating and downloading context and sharing context work in companies that are, you know, like they have multiple employees and they're doing things in different platforms.
[14:45]
B
Yeah. In order to, so you have to give people a workspace where they can work without friction. Basically. Like let's just say you're creating a task like you're creating A landing page or something like that, you need to give people a workspace where they can do that end to end. Because when you try and tell people write a perfect prompt to generate a landing page, it's very rare that someone can do that. It's basically impossible that someone can just do that. Right? But if you give them a way for them to generate that landing page end to end, and then iterate on it and iterate back and forth with the agent, then what you get at the end of that, if you just say, note down all of the adjustments that I made and all of the edits that I made and turn this into a skill, then you actually do end up with something useful. And they also don't need to understand the concept of what a skill is up front. Or like they get stuck in this decision paralysis because they don't know what to do.
[15:39]
A
I just learned what a skill is last week.
[15:41]
B
Yeah, exactly. Like you don't really need to know, right? Like you don't. All you need to know is like, the agent knows. The agent knows what to do, or the harness knows what to do. All you need to do is do the task and give it feedback. So the best way that I like to do it is connect all the systems for people up front so that they can at least do work end to end, unencumbered, give them that workspace and then just tell them to do the work. And when the work is done, then turn it into something reproducible or at least do it in a place where I or someone else has access to it or it's saved down somewhere so that you basically have that as training data and then someone else can aggregate that into a repeatable workflow. But these things fail very quickly when someone comes in like an AI consultant, pretend they understand your business, interviews a couple people and tries to create a skill to do something which is hard. Like to create a great landing page is hard. It's many years of experience.
[16:34]
A
Right. Especially maybe you have like certain documents that, you know, have legal constraints or how you can word things and. Yeah, totally.
[16:43]
B
And if I told you to just write that down up front, you would miss things. You would be. You would forget that you have your brand guidelines in some random Google Doc or notion.
[16:50]
A
Yeah.
[16:51]
B
Somewhere, you know, but if you did it and you created and you were like, oh, this doesn't look like my brand, you'd be like, oh, wait, I actually, I usually share this with the creative strategist. And that's how it comes together.
[17:01]
A
Yeah. And are most People like, like in this process of creating skills, is this something that like HQ is doing on its own for as you're using the product, or is this something that people are setting time aside and saying, okay, like these are all the skills I need to build out now let me go and build out these workflows and try to train HQ on
[17:22]
B
happens in two ways. So one is when you set up hq, it will ask you what are the things you want to create? What are the repeatable workflows you want to create? And it will go through that process with you. The other way that it does it is the one that I just described where it at least guarantees that all work is captured and recorded somewhere. And then sometimes people will go back and create those skills or those workflows automatically. But the interesting cool thing about HQ is because that's all stored somewhere, you're kind of just creating reproducible workflows without even having to really think about it. Because I can just say, do that thing that I did last week and HQ will be able to find it for me. And now it's become a reproducible workflow in a way. So people, they get lazy. Like I'm, I'm like a victim of this too, where I'm not even like creating the skill always. I'm just saying like create that deck the same way that I did it last time. And now I have like created a meta workflow almost just because I have the training data somewhere.
[18:21]
A
If you've ever wondered how the top brands actually run tv, here's Tick pick on why they use Tatari. Tatari has really been like an extension of our team. They make it extremely easy to buy tv, which if it's your first time, can seem a bit scary. The client services team is incredible. The platform is super easy to navigate, tons of great insights and Atari team is always bringing you deals where they think, you know, your audience might be. So our favorite feature in the Tatari platform is the reporting and the ability to see performance at the network and even the show level. It's something that our executive team loves to see and then it helps us optimize towards, you know, better media buying in the future. That's what real TV performance looks like. To learn more, go to Nick co Tatari. Do you have a lot of any stories here about like where somebody has done something on HQ or created some kind of like a skill or, or process and then the whole company has been able to benefit from it?
[19:25]
B
Yeah, we have hundreds of examples in our company. I can also talk about it for other companies too. But I mean a good example for our company, something I'm experiencing right now is I'm like an analytical person, but creatively I'm not quite there, you know, like I can't, I can't create a beautiful deck. Like I'm not going to put beautiful assets together. But we have someone on our team who is incredible at that. And I created this one pager for something we were trying to show and it like, it looked terrible, like it had all the right numbers there, but it looked absolutely terrible. And then she made one and it looked amazing. It looked great. And now. So she made this one deck and now this is the template for all decks that we create inside of hq. And I just took that one pager I created, I told Claude, I said reference the deck that she made and just combine them together. So that's the new aesthetic. And now all of these assets that we're putting together, like partnership decks and sales decks and investor decks and everything like that, they all look amazing because they're created through her artistic eye. But everyone else gets to benefit from that. So that's one example. And that extends out to all different decisions or things created in the company. On like. Yeah, all the engineering things that we do are based on what the guidance was of our best engineers basically. And it's not just the code itself, but it's like the rules and policies that they put in place for our code base that live within hq or the way that I analyze data, which is like a little bit different. Like it's HQ is a quite complicated system and so the way that I analyze data now everyone else can adopt and look at through the skills and workflows that I've created to make sure they're looking at the right data. So it just, that just extends in many, many different ways to everything we do in the business.
[21:15]
A
Yeah. How do you see different roles within a company? Like leveraging AI in their own ways to be advantageous in what kind of
[21:25]
B
business and like our kind of business or like.
[21:29]
A
Yeah, yeah, I think like more on the brand side, I guess.
[21:36]
B
So a lot of it is around. There's like two sides to. And it almost depends if you are revenue generating or like a proper generating role or not. Like what I see for a lot of marketing teams is that they're able to generate a lot more. They're able to basically extend themselves outwards and generate more and do more revenue generating activities or people are able to do that with smaller teams. They're able to maintain the same amount and smaller teams. So it's basically more per team member to the extent that I know single people who are excellent marketers doing like two or three or four roles themselves just with AI and creating agents and that's really, really cool. So these people are doing more with less and it's resulting in growth for their company, like more ads, more landing pages, more emails, or at least being able to handle it themselves. And then when it comes to other roles, like backend roles, finance roles, stuff like that, I'm seeing of course, like it's just more efficient. There's a lot, there's a lot of efficiencies. People are getting more disciplined about data and about infrastructure. Like everyone wants a data warehouse now because they know they can have access to do all these things. But I'm seeing more sophistication on the level of analysis that they're doing. Like before someone may have just had, they were just like creating some kind of inventory forecast or basic or putting their P and L together. But now with some really young brands without a ton of resources, I'm seeing them put together super sophisticated like LTV to CAC analysis per sku, per subscription cadence because, and they're able to run these analyses that are really, that are like significantly impacting the trajectory of their brand because they have so much access to data and because if they just use Fable 5 and Claude, it's like a state of the art AI, like Data Analyst, you know, so more, it's resulting in people able to do more with less and a lot more sophistication in what they're doing as well.
[23:42]
A
Yeah, I feel like the, like there's the notion that AI is going to replace people's jobs, but AI either on its own or AI being run by somebody who doesn't have domain expertise is never going to compare to like, like you're saying these, these brands probably have somebody who's like a sharpshooter at what they do. And then being, you know, being able to use AI now they have all the leverage in their own pocket.
[24:05]
B
Yeah. And I mean I never agreed that AI was going to take a bunch of jobs like I, to me that always assumed that brands were operating at capacity already. Like everyone was just operating at 100 capacity with 100 efficiency and there was no room for growth, but I never saw that in the company that ever worked with. So I feel like these people are just doing more. They're just doing more and better. Like the silly mistakes that I Used to see people making like, putting a ton of ad spend behind products that are out of stock or like really, like really messy tags inside of Shopify. And like skews, like these mistakes aren't happening as much because the ability for AI to just clean it up for you is very, very easy. And so the sophistication that they're running at is more and so they're just doing better. But these people haven't gone anywhere. The people that can use AI, they haven't. They're not getting fired.
[24:58]
A
Yeah. Have you seen that one chart too? That's, it's got like all the dots and it's like, this is how big the world is and then the last dot is just like, this is the people who've just discovered AI.
[25:08]
B
Yeah.
[25:08]
A
It's like, yeah, we've still got a long ways to go.
[25:11]
B
We do, yeah. And I like, I think companies are just going to do very, very well. Like, I think, I think there's a lot of room to grow for all companies with the people that they have.
[25:19]
A
Yeah. What do you think? Like brands that are looking to be, you know, like, they're not looking to be in a spot next year where they feel like now they're behind on AI, you know, but maybe they're not fully AI pilled yet. Like, where do you recommend they go or look into or like, where should they even start?
[25:40]
B
I mean people, most people who ask me this question, I say to them, are you using it every day? And they say no. And it's like, well, that's, that's what you have to do. You have to open Claude code or open Codex or create a Hermes agent or an HQ agent. They should be definitely be using hq, but just like use this thing every single day and try and do everything with AI. Like, I can't imagine almost any task that I do in my day to day that I wouldn't go to one of these AI tools for first. And so it just blows my mind that people are like that. People don't know what to do or where they can work. And I've worked in, I've worked with brands, I've worked with agencies, I've worked with software companies. I've worked in like the biggest and smallest companies in the world. And there's everything that everyone does. If you set up the right systems and you understand how to use them, Cord code can do it just as well as you. And so if you are the leader of your company, you have to be doing that every single day. Once you Have a good grasp of how to use it. You should sign up for HQ and then get your whole team on HQ so you can roll this out to the rest of your team. But if you're not doing it every single day, if you're not training yourself, then there's like, there's nothing else you can really do. Yeah, it's not going to magically come to you. I think people assume that it's just going to magically come to them.
[26:58]
A
Yeah. I heard this head of HR for a pretty large consumer brand recently, basically be like, you know, we can't discriminate against people when we hire, but like, if you're not using AI and this day and age, you're basically just not fit for a role at any, you know, anymore.
[27:18]
B
Yeah.
[27:19]
A
Like, that's almost just a given.
[27:21]
B
Yeah.
[27:22]
A
But at the same time, I feel like there's so many companies I talk to that are just still, you know, so far away from any sort of real adoption other than, okay, let me, you know, take a CSV, dump it in here and ask some questions.
[27:35]
B
Yeah, yeah. I mean, even if that's where they start, like, even if that's just where they start and then they just do a little bit more every single day, they'll get good eventually. Like, I wish I could shake people and just tell like if I took the average employee and I told if you used Claude code every day for the next two months, you would probably be the most valuable person in your company. And, and I see that every single day, time and time again and people just don't do it. I think it's hard. It requires a different level of thinking. People are busy. But I wish people realized that.
[28:08]
A
Yeah. Okay, couple questions I have. So like now going back to, you know, HQ and having kind of your brands or your company second brain, you know, agentic and ready to access, how do you make sure that this thing is always staying up to date? How do you retire Stale context, You know, how do you make sure that if, like there's one thing that's wrong in there, it gets caught or fixed so that the whole company is now not operating off of this? Like how do you think about that or how does that work?
[28:40]
B
The best way to deal with that is to use it every day. Like the more people you have using it every day, the quicker you would uncover and pick up and fix incorrect context. Like if I'm doing something inside of HQ and it says something wrong that I see related to my job, I picked that up very quickly and now I can remove it. And it's kind of the same with every. It's like this, it's the same with all work. Like it's not very different to pre AI work. Like if I was creating an SOP and the SOP had a wrong step in it and then I did it and it was wrong, then I should go and fix it. And so that same kind of responsibility is on people, although that's not always a good enough answer. And so we have a couple ways of automatically dealing with it. We have something called a gardening agent. And the gardening agent basically will automatically go into hq, look for it, look, it can look within the actual message sessions themselves and try and determine this. Like if someone said something was wrong or incorrect and it wasn't solved, then it will automatically remove that. It will look for contradictory context and try and remove that. And then it will also age context based on date. And so if something is old, it can expire it or out of date, it can expire it. And it can do this with the full context of your company. Like for example, a really good example is, let's just say we work together and then you leave the company and it's like, and then it says you should go talk to Nick about this. But HQ should know that you left the company because somewhere inside of hq, somewhere inside of a call transcript, somewhere inside of a slack message or an off boarding session, there should be something that says Nick left the company. And so HQ should be able to find that and then fix that context. And that's what the gardening agent is supposed to do.
[30:30]
A
Yeah, that's pretty cool. I'm curious, like, where do you see the future of HQ going? Like, we were talking a little bit earlier today about some of the other companies in the space, but yeah, where do you see HQ's future?
[30:44]
B
The future of HQ is there's two pieces to it. One is that it should be the default option for collaborative AI work. So when we work together, when we do something with AI, there's no really great way for us to truly collaborate. Like to be truly multiplayer in that, working on something at the same time, working on a project at the same time, right? There's like, it just doesn't, it's not built for that. And so HQ should be the way to do that across multiple different agents, across multiple different people, across different teams and vendors. If anything requires any kind of collaborative work, HQ should be the default solution to that. And then the other one is that what we are collecting inside of HQ is all the Important context that you need to do your business to run your business. It has context on every single thing that you do in order to run your business. And so what we should be able to do is tell you proactively, once we recognize it, what are the best models that you should be using, what are the areas of opportunity for automation, what are you missing that you could be doing, what are agents that you could be creating and then also let you train your own models to be able to do that. We see a future where companies themselves can have their own models that are hypertuned to how they run their own language, their own workflows. And we should give you the ability to do that so you can get your edge based on using hq. And you should be able to basically create this recursive self improving loop where you do work, the model improves, you do more work, the model improves. And we should be able to give that all to you through hq.
[32:26]
A
Yeah, and I think too the other piece there is like the, or one thing to call out is it's not really the model that improves. Right. Because the model stays the same, but it's like your HQ operating system is what continues to improve and get smarter and smarter.
[32:41]
B
Well, we should be able to give you a model like we like there's going to be a future probably in about two years from now where it's going to be cheap and affordable enough that rather than having a Hermes agent running your Mac Mini, I can actually ship you a computer where a true model runs for your company, like inside of your office. Right. Like this is an actual AI model that is yours and we should be the ones giving that to you because we have all the training data to make that possible based on what we have inside of hq. And now, like this is your ip. It's like AI is more than just a tool. It's the intelligence of your company. Like, it's truly the, it's the differentiator of your company. Like, it's, it's a reflection of you and your intelligence basically as a team. And so that's important. And I think people are going to be ambivalent to give that all to the model companies eventually. And so we should be giving that, we should be allowing you to truly own that, like truly physically own that for your company.
[33:38]
A
Yeah. Wow, that's pretty cool. What kind of like AI habits or use cases do you think are going to be looking ridiculous by next year that people do right now?
[33:52]
B
What's going to be ridiculous? I mean, I think Claude, Chad and Cowork are going to go away. Like, I think anyone saying they were using cowork instead of code, like, that's just all going to collapse and go away. I think any company that's stuck to any single LLM provider, like we have companies that we work with that are just, I'm just a Claude teams company, those companies are going to get in a lot of trouble and that's going to look silly. I think that in the future we're going to see more efficient model routing and I think it's just going to happen automatically. And I think the whole thing of like, you know, if you go into Claude code or in Codex and you like select the model, you select the effort mode, you select the like plan thinking like all of that, I think that just goes away and all of that complexity goes away and it just happens automatically. So I think all these automatic toggles that we have, but I think in general, like the whole prompt engineering thing, like that was, that was like a little bit premature and looks silly now, but I think we're at a more sophisticated level now where we're not doing too much crazy stuff. We're basically just talking to the models now. We're not having to do any crazy, like, hacky things. And so I don't see anything like that disappearing.
[35:05]
A
Yeah, I feel like prompt engineering, like, I remember, I think maybe last summer I talked to Billy Howell. Have you ever talked to Billy?
[35:14]
B
No.
[35:15]
A
Billy Howell's like, he was vibe coding a bunch of apps last summer and like he was teaching me what a PRD is and how you have to get a prompt to make a prd and then the PRD goes into replit or whatever. Now it's just like you just rip a whisper flow for four minutes and like that's. That's about all you need to do.
[35:33]
B
Yeah, yeah, 100%. Well, I actually think something that might just be disappearing is all the. I think Vibe coding is going to stay and I think people are going to be creating internal apps, but I keep seeing people creating and shipping SaaS, companies that don't need to exist. And I think that's like a moment in time, like they do something cool and totally.
[35:52]
A
Also there's going to be like a hundred thousand SaaS, companies with no customers.
[35:56]
B
Yeah, exactly. Goes away. And I think people realize that there's more alpha and more value in sticking to the businesses that they're good at and just try to apply AI to that.
[36:06]
A
Totally cool. Well, Jacob, thank you for coming on limited supply and talking about hq. We should definitely we should figure out something else to do with hq. Because again, like I was telling you earlier, I feel like there's so much that it enables you to do that I still even haven't uncovered or understood yet.
[36:26]
B
Definitely should do some working sessions.
[36:29]
A
100% cool. And where can people find you?
[36:33]
B
You can find hqforwork.com and you can find me on Twitter. Jacob Posel POSCL thank you for coming on. Thank you.
[36:48]
A
Thanks for listening. We'll be back next time to cut through the noise on CPG retail and E commerce. If you enjoyed this episode, why not share it with a friend Friend. And be sure to subscribe wherever you listen so you don't miss the next one.

Summary8 min read

Podcast Summary: Limited Supply – S17 E6

Episode Title: Why Your Company's AI Chats Are a Total Mess (And How to Fix It)

Host: Nik Sharma
Guest: Jacob Posol, Co-founder of HQ
Release Date: August 12, 2026

Episode Overview

This episode of Limited Supply dives into the realities and challenges of deploying AI tools—like ChatGPT and Claude—across entire organizations. Nik Sharma welcomes AI expert Jacob Posol, co-founder of HQ, a platform aiming to make AI adoption within companies more collaborative and effective. They discuss why so many companies’ AI chat deployments are fragmented and messy, how to fix these issues, and what the future of AI in business looks like.

Key Discussion Points & Insights

1. The Fragmentation Problem with Company AI Adoption

AI Rollout Issues:
- Companies often simply give employees access to AI tools (e.g., Claude, ChatGPT), resulting in individualized, siloed workflows.
- Lack of a unified sharing mechanism leads to duplicated effort, disjointed knowledge, and inconsistent efficiency gains.
- Some employees embrace AI deeply, while others fall behind, causing "FOMO" and organizational divides.
- Jacob (03:34): "You'll have the person ... build something cool, and they'll have no way to be able to share that with their team. That often results in a huge amount of duplication of workflows and of work, which is kind of the opposite of the efficiency that you hope to get with AI."
Root Cause:
- Not an AI problem, but an infrastructure problem—current tools don't support easy collaboration or knowledge sharing.

2. Is “AI-Native” Really Happening?

AI adoption is generally top-down, championed by founders or CEOs, rarely by entire teams.
Outside the Twitter/builder sphere, AI uptake across companies remains surprisingly low.
There’s still resistance or skepticism among “normies,” creating further barriers to widespread adoption.
- Jacob (05:52): "I'm always surprised by the lack of AI proficiency that I still see ... Most of my circles ... it's just the people at the top that are really trying to push it because they're the ones who experience the ROI from it."

3. Jacob’s Backstory & The Birth of HQ

Started in machine learning, moved into e-commerce logistics, saw ChatGPT’s arrival as a game-changer, and instantly pivoted to build in AI.
Early efforts to build AI for logistics fizzled due to tech limitations—became possible only after later model upgrades (Claude Opus 4.8, etc).
Co-founder Corey Epstein first built HQ as a personal tool, then teamed up with Jacob to turn it into a collaborative AI OS for organizations.
- Jacob (07:38): "He had actually created this personal tool called HQ ... I was like, Corey, this is sick, but this needs to be a different tool. This needs to be for companies."

4. How HQ Works (Explained Simply)

HQ connects all company AI tools into a shared environment—like a "Dropbox sync" (Nik, 11:19) for AI, ensuring everyone has access to the same workflows and organizational knowledge.
Enables "company-wide memory" using vector memory—context isn’t just stored in files but is deeply searchable and retrievable.
- Jacob (10:27): "HQ basically connects all of our different AI tools together so that we're operating on top of shared context."
Makes sharing workflows and skills seamless, so lagging team members can leverage work done by power users.

5. Transitioning from Chatbots to “Agents” and “AI Employees”

Two main directions in the space:
1. More technical, customizable code-based agents (Codex, Claude Code, etc.)
2. AI "employees" or virtual team members embedded in everyday tools (e.g., Slack bots with personas)
Hybridization will be key.
- Jacob (13:23): "The AI virtual employees are really, really easy to adopt and understand for less technical people ... you have to connect the two together."

6. Building and Sharing Skills/Workflows

The most effective organizational AI adoption involves capturing actual work as “training data”—not asking everyone to write perfect prompts up front.
Workspaces should be frictionless: people do their tasks, AI records workflows, and these become “skills” or reproducible processes.
Overly consultant-led, top-down skill creation often fails. The value comes from real, evolving work patterns.
- Jacob (15:39): "The best way that I like to do it is connect all the systems for people up front so that they can at least do work end to end, unencumbered..."
- Jacob (17:22): "The cool thing about HQ is ... you’re kind of just creating reproducible workflows without even having to really think about it."

7. Real-World Examples

HQ enables sharing of best practices across roles—e.g., a creatively skilled team member's deck template becomes the company default, immediately leveraged by all (19:25).
Engineering and analytics best practices, as well as business processes, can be propagated and standardized company-wide.

8. How Different Roles Benefit

Marketing: AI allows smaller teams to produce more (emails, landing pages, ads), and single “superusers” to cover multiple roles.
Finance & Ops: Increased data discipline and more sophisticated analyses by leveraging AI tools and shared data warehouses.
- Jacob (21:35): "Single people who are excellent marketers doing like two or three or four roles themselves just with AI... It’s resulting in people able to do more with less and a lot more sophistication."
Domain Expertise Still Matters: AI multiplies the impact of sharp operators; it doesn’t substitute for domain knowledge.

9. Practical Steps for Teams New to AI

Start using AI daily: It’s a habit and skill that compounds quickly.
Even "dumping a CSV and asking questions" is a valid entry—just keep going deeper.
- Jacob (25:39): "Most people who ask me this question, I say to them, are you using it every day? And they say no... You have to open Claude code or ... HQ agent ... and just, like, use this thing every single day and try and do everything with AI."

10. Keeping Company Knowledge Accurate & Fresh

Daily usage is the best safeguard against stale or incorrect context.
HQ automates context hygiene via a "gardening agent"—removes outdated or contradictory information and can even note HR changes (e.g., offboarding).
- Jacob (28:39): "The gardening agent ... will automatically ... try and determine ... if someone said something was wrong ... it will automatically remove that ... and [expire] out-of-date context."

11. The Future of HQ and AI in Organizations

HQ as the "default option for collaborative AI work"—a multiplayer, project-based experience.
HQ aims to provide companies with their own locally running, hypertuned models—and the ability to own their AI “intelligence” and keep their own IP.
- Jacob (32:40): "...we should be able to give you a model ... where a true model runs for your company, like inside of your office... AI is more than just a tool. It’s the intelligence of your company."

12. What Will Seem Outdated Next Year?

Using a single LLM provider for everything will be obsolete—more efficient, model-agnostic routing will be standard.
Manual model selection, prompt engineering, and toggle-choosing will disappear—much will be automated.
Proliferation of SaaS for “mini-apps” will fade; more energy will shift to using AI to supercharge core business, not spinning up throwaway tools.
- Jacob (35:55): "I think people realize that there’s more alpha and more value in sticking to the businesses that they're good at and just try to apply AI to that."

Notable Quotes & Memorable Moments

On Siloed AI Use:
- "You'll have the person ... build something cool, and they'll have no way to be able to share that with their team ... which is kind of the opposite of the efficiency that you hope to get with AI." — Jacob (03:34)
Adoption Gaps:
- "I'm always surprised by the lack of AI proficiency that I still see ... it's just the people at the top that are really trying to push it." — Jacob (05:52)
On Replacing Jobs:
- "AI either on its own or AI being run by somebody who doesn't have domain expertise is never going to compare to ... these brands [with] somebody who's like a sharpshooter at what they do." — Nik (23:41)
On Getting Started:
- "If you used Claude code every day for the next two months, you would probably be the most valuable person in your company." — Jacob (27:35)
On the Future of Organizational AI:
- "AI is more than just a tool. It’s the intelligence of your company. It’s ... the differentiator of your company ... a reflection of you and your intelligence as a team." — Jacob (32:40)
AI Hygiene via Gardening Agent:
- "The gardening agent ... will automatically ... try and determine ... if someone said something was wrong ... it will automatically remove that ... and [expire] out-of-date context." — Jacob (28:39)

Key Timestamps

Main Interview Begins / Jacob Joins: 02:23
Fragmentation of AI Use & Sharing Issues: 03:34
Adoption Gaps and "AI-Nativeness": 05:52
Jacob’s Path & HQ Origin Story: 07:38
How HQ Works (Simple Explanation): 10:27
The Power of Company-Wide Memory: 12:13
Emergence of AI “Agents” & Virtual Employees: 13:23
Capturing & Sharing Skills/Workflows: 14:45
Real Examples of Know-How Propagation: 19:25
Role-Specific AI Benefits: 21:35
Advice for Teams Just Starting Out: 25:39
Keeping Company Knowledge Fresh (Gardening Agent): 28:39
Future Vision for HQ & Collaborative AI: 30:44, 32:40
Predictions: What Will Be Outdated Next Year: 33:51
Wrap-Up & Where to Find Jacob: 36:32

Final Thoughts

Companies are still in the early innings of AI adoption, with most efficiency and competitive advantage limited by internal infrastructure, cultural buy-in, and willingness to experiment.
HQ aims to be the connective tissue for organizations, making AI collaborative, shareable, and persistent—unlocking much more than piecemeal adoption.
Core advice? Start using AI every day, deeply and broadly, and you’ll be miles ahead of the curve—and more valuable to your company than ever.

Connect with Jacob:

hqforwork.com
Twitter: @JacobPosel

Host: @mrsharma

If you found this episode insightful, share it with a friend or colleague ready to scale their organization’s AI maturity, and subscribe to Limited Supply for ongoing tactical wisdom in DTC, CPG, and e-commerce.

Loading summary

Transcript69 lines

[00:01]
A
Welcome back to Limited Supply, the podcast where we get deep into the tactical and strategic side of e commerce, digital marketing and building consumer brands. I'm your host, Nick Sharma. I've spent the last nine years building, scaling and investing in brands. And through this show and my weekly newsletter at Nick Co Email, I'm here to share everything I've learned. The wins, the losses, the experiments, the tactics and the insights. All so you can unlock your next hundred thousand dollars in revenue. Today's episode is a good one, but before we dive in, let me tell you about our channel chosen sponsor for this week's episode. If you're a DTC operator and TV feels like a black box, that's because most platforms give you 20% of the actual available inventory and none of the measurement to see if it even works. On the other hand, with Tatari, you can buy linear and streaming directly, plus you can measure it like a digital channel. That's the reason over 400 brands like Manscaped Toqovas and Chime run performance TV using Tatari. See how they leverage the platform at Nik Co Tatari. That's Nik Co Tatri. Welcome back to another episode of Limited Supply. I'm your host, Nick Sharma and today I've got a very special guest with me and he himself is an AI expert and somebody that I've learned a bunch from when it comes to AI. And you know, one of my goals here is to bring constantly whatever I'm learning, bring it to you guys and in a way that's easier, easy to digest and easy to understand. So today I brought on a friend of mine, Jacob Pozol, who I met recently and have actually learned a lot from him and the company that he is building, which is really around creating kind of an operating system, an AI operating system for your brand or your company, your organization. And so we talk all about, you know, how to get on with AI, how to start using it as a company where there's onboarding failures or what succeeds when you're onboarding and just really like, how are different people within the org also using it to their advantage? So check out today's episode, let me know what you think. Shoot me a DM on Twitter. You can also DM Jacob if you got any questions for him. And I'll see you in just a moment. All right, Jacob, welcome to Limited Supply. I'm excited to have you here today. You've been all over the Twitter feeds. Honestly, we. We never really connected until recently either.
[02:24]
B
Yeah.
[02:25]
A
But we've been Twitter friends and Obviously, you know, we've seen all this, everything on the timeline, talking about hq. So I thought, you know, why not just bring you in and let's talk about hq is also somebody who's now been a user for a couple months and started to learn a little bit more about what you guys are doing. I thought it would be really cool to share what you're doing and how, you know, other brands and other operators can leverage it.
[02:47]
B
Yeah, I'm excited to be here. Thank you for having me.
[02:50]
A
Of course. So why don't we start with, you know, kind of like the one thing that I think will get people excited about Hearing more about HQ2 is what, what happens, you know, when a whole company starts using Claude or ChatGPT, and what doesn't happen, you know, like I was at, I was at a dinner recently and somebody was telling me about this new logistics tool that she built in the. In Claude, and unfortunately, her direct reports can't get access to it because she built it in Claude. So I want to start with that and then I want to go into more of HQ in your background and then let's talk about how different functions are using AI from what you're seeing as, as their clients of hq.
[03:35]
B
Yeah, yeah. I mean, I've seen. I've seen the insides of lots of businesses as they try roll out these tools, and it always results in this extremely weird fragmentation, which is obviously what we're trying to solve. But there is this extreme gap in certain individuals in their ability to pick up the tools. And so you can give 50 people Claude or 20 people Claude, and the distribution of how they would actually end up using that tool is completely different. And so you'll have the person like you just described there. They'll build something great, they'll build something cool, and they'll have no way to be able to share that with their team. That often results in a huge amount of duplication of workflows and of work, which is kind of the opposite of the efficiency that you hope to get with AI. You also have this, like, weird FOMO that happens where people start to get scared and they, like, see other people accelerating past them, they can't quite catch up. They get scared of the tool when it's not really necessary. And because it's so new and you kind of have to go to the edges to figure out how to use these things, you just. The differences are, like, are very wide. And so what I've seen happens is you try and you roll these out within a company, but you Never really get that nice compounding benefit that you expect. Like, you get people being more efficient and more effective, but you don't see it moving the needle for the business. But I think that's an infrastructure problem. It's not an AI problem. And so that's obviously what we have tried to solve with hq.
[05:14]
A
Yeah. And are you seeing companies becoming AI native? I actually, you know, like, obviously I feel like the answer is yes, like when. Whenever I ask that. But this. This past weekend, I met up with a friend of mine who's a YouTuber and I just randomly brought up AI to him and he's telling me that, like, from what he sees, you know, among. Call it normies, not us, you know. Yeah, podcast listeners or people, I guess, whatever. He said that more normies are like very anti AI, and so I wonder if that's trickling its way into the business world outside of our bubble. Like, I'm curious what you're seeing.
[05:53]
B
Yeah, I haven't seen businesses that are fully AI native and enabled unless they're very recently founder businesses or you have. You can have majority AI enablement, but that has to come from an extremely strong CEO founder figure. I'm always surprised by the lack of AI proficiency that I still see. The tools have gotten easier and gotten more powerful, so it's easier to be good with AI, but you don't see. I'm so surprised. Most of my circles, I try. When I try. When I socialize with friends, I try not to talk about AI. It's my only one time that I can, but it's obviously not possible. But yeah, I haven't seen that same penetration outside of the, like, Twitter sphere that I live within.
[06:40]
A
Yeah.
[06:40]
B
But again, like, the. That sphere is mostly founder, CEO types, you know, like, even within their companies. It's not the same. It's not. It's not. It's just the people at the top that are really trying to push it because they're the ones who experience the ROI from it.
[06:56]
A
Yeah, I feel like too, like, that is. Yeah, that's actually probably what I've seen too. There's usually a few people in the org who are super AI pilled and they're trying to bring everybody else with them and kind of get the org set up. Okay, I want to go into your backstory a little bit. So you were in engineering and then you sort of made your way into the marketing sphere or the kind of like the tech sphere then, you know, and I feel like your journey with AI probably evolved. I'm curious like what, what did you think you were getting into at the beginning versus, like, where did it evolve to? And like, how did you guys get to hq?
[07:39]
B
I mean, my AI journey started very early. I was. I was doing a. Basically an AI machine learning master's degree when, well, I graduated right before ChatGPT was launched. And so I was already in the machine learning space. I got a job as a software engineer and data scientist at an E comm logistics company. And I was helping them build basic forecasting models, like sales forecasting models to help forecast demand within the warehouses. And then ChatGPT was launched and I basically just quit my job instantly. I was like. Because I think I saw the writing on the wall a bit sooner because I saw how powerful this was. And so I literally instantly quit my job and try to start an AI company. And I was trying to build an AI logistics company for E commerce, which is very different to where I am now. But it just wasn't that powerful back then. And I think you could always see where it was going, but it wasn't quite there. And then in a roundabout way, I got connected to Common Thread Collective with Taylor Holiday. He kind of took me and took me under his wing and I worked with them for a while. And again, like, we always had these super ambitious visions of what we could do and what we were trying to do. And it felt like we were always waiting for the next model to release. Like, okay, now we're going to be able to do it. Now we're going to be able to do it. And then that actually happened, like around December last year, that actually happened with like the Opus 4. 7, Opus 4. 8 launch. And then we had this amazing moment. We were able to just do everything we wanted to do, build what we wanted to build. And I kind of just went crazy just doing as much as I possibly could. The way that HQ came about was Corey Epstein, who, who I was connected with again also through the same, like, Twitter world. Him and I had been in touch for a while. He had actually created this personal tool called hq. At the time, it wasn't a team tool, it wasn't a company tool. It was his way of being able to manage AI personally more efficiently and more effectively. And he had started converting all of his friends onto it. So that's how he got me onto hq. And I was like, corey, this is sick, but this needs to be a different tool. This needs to be for companies. This would be so much more powerful if all those context could be shared across companies. And that's how we teamed up, co founded this company and now that's how we have hq. And now it just feels like we're able to do whatever we want. Like it's just, it's this unbelievable feeling. And now we're building infrastructure because we've experienced so much across so many different businesses. I've seen so much across so many businesses through that journey. So now we're just trying to build the best infrastructure for companies to run on.
[10:19]
A
Yeah, that's pretty amazing. And can you explain like to a fifth grader level person how the product works?
[10:27]
B
Yeah. So when you have a whole company using Claw, using chatgpt or using AI, what happens on my chatgpt is just on my computer, it's in my personal account. What happens in yours is just on yours. And we have no clean and easy way to share that with one another. And HQ basically connects all of our different AI tools together so that we're operating on top of shared context. So if I do something, I can share it with you. If I build a workflow, I can share it with you. If I connect to a new system, I can share it with you. And so that allows us to work together, that allows us to compound what we're doing across an organization and it allows us to get those people who are lagging behind to be on the same level as us. Because if they just use ChatGPT connected to HQ, they're automatically inheriting everything else that we've already put in there from the rest of the company.
[11:20]
A
Yeah. I also describe it as like Dropbox sync across the company for all things AI related.
[11:27]
B
Yeah, yeah, that's a good one. If you. There's a good way to compare it to other companies. It's like Dropbox for agents, 1Password for agents and then like agents for agents kind of thing as well. But yeah, those are like a good, good ways to think about it.
[11:44]
A
Yeah, I use it to basically just stay up to date across all devices and all team members across all projects. It's pretty amazing. And the other thing I think is cool and not talked about enough is your memory system is like the way it's built. Is it called vector memory? Is that right? Yeah, yeah, vector memory. So it's not just reading markdown files, it's like a whole different way to recall memory, which just makes all the context and makes the output so much better.
[12:13]
B
Yeah, like it's really important that. Let's just say you built that logistics tool and I started building a logistics tool. It's really important that Claude is able to recognize that you built it first and to find it and, and to do that we had to give this, create this company wide memory which uses some really cool searching and retrieval techniques that we've built.
[12:33]
A
Yeah, very cool. Now I feel like a lot of companies right now are popping up with their or you know, like this. Everything started in the chat window and then, and then a bunch of other softwares have integrated models within what they already do. So like you know, influencer searching software now has a model to help find better creators maybe. And then it's going toward more like open source software where you plug your models in and you, you know, have your thing work however, which is probably where it sits with like Claude code and Codex. And then I feel like the next version of this is maybe. Actually I don't know what the next version of this looks like, but AI employee is starting to pop up now a lot more and I think because the quote unquote agents are easier to visualize or build or create. Is that what you're seeing right now?
[13:23]
B
Yeah, yeah. I mean it's two, there's two sides to it is definitely the Claude code Codex, like the Harness agent agentic tools for slightly more technical people. But then the AI virtual employees are really, really easy to adopt and understand for less technical people. Because if you give it a Persona and you drop it in slack, then people are able to pick it up easier. So I'm seeing that a lot, especially with like the Grokbot launch that came today. Like there's a lot of these companies that are doing really well. I think it's a combination of the two that to work though, like I don't think one works without the other. You have to connect the two together.
[14:03]
A
Yeah. And how does like a, a company when they are getting started on all of this? Right. Like I felt like with my own AI setup I spent definitely a month or two intentionally trying to create knowledge files or, or you know, basically ways of, of where I would be different. So like if the prompt was you know, build a wireframe for a landing page, my personal one would be way different than like if you just ask generic chat GPT. How does that whole process of like creating and downloading context and sharing context work in companies that are, you know, like they have multiple employees and they're doing things in different platforms.
[14:45]
B
Yeah. In order to, so you have to give people a workspace where they can work without friction. Basically. Like let's just say you're creating a task like you're creating A landing page or something like that, you need to give people a workspace where they can do that end to end. Because when you try and tell people write a perfect prompt to generate a landing page, it's very rare that someone can do that. It's basically impossible that someone can just do that. Right? But if you give them a way for them to generate that landing page end to end, and then iterate on it and iterate back and forth with the agent, then what you get at the end of that, if you just say, note down all of the adjustments that I made and all of the edits that I made and turn this into a skill, then you actually do end up with something useful. And they also don't need to understand the concept of what a skill is up front. Or like they get stuck in this decision paralysis because they don't know what to do.
[15:39]
A
I just learned what a skill is last week.
[15:41]
B
Yeah, exactly. Like you don't really need to know, right? Like you don't. All you need to know is like, the agent knows. The agent knows what to do, or the harness knows what to do. All you need to do is do the task and give it feedback. So the best way that I like to do it is connect all the systems for people up front so that they can at least do work end to end, unencumbered, give them that workspace and then just tell them to do the work. And when the work is done, then turn it into something reproducible or at least do it in a place where I or someone else has access to it or it's saved down somewhere so that you basically have that as training data and then someone else can aggregate that into a repeatable workflow. But these things fail very quickly when someone comes in like an AI consultant, pretend they understand your business, interviews a couple people and tries to create a skill to do something which is hard. Like to create a great landing page is hard. It's many years of experience.
[16:34]
A
Right. Especially maybe you have like certain documents that, you know, have legal constraints or how you can word things and. Yeah, totally.
[16:43]
B
And if I told you to just write that down up front, you would miss things. You would be. You would forget that you have your brand guidelines in some random Google Doc or notion.
[16:50]
A
Yeah.
[16:51]
B
Somewhere, you know, but if you did it and you created and you were like, oh, this doesn't look like my brand, you'd be like, oh, wait, I actually, I usually share this with the creative strategist. And that's how it comes together.
[17:01]
A
Yeah. And are most People like, like in this process of creating skills, is this something that like HQ is doing on its own for as you're using the product, or is this something that people are setting time aside and saying, okay, like these are all the skills I need to build out now let me go and build out these workflows and try to train HQ on
[17:22]
B
happens in two ways. So one is when you set up hq, it will ask you what are the things you want to create? What are the repeatable workflows you want to create? And it will go through that process with you. The other way that it does it is the one that I just described where it at least guarantees that all work is captured and recorded somewhere. And then sometimes people will go back and create those skills or those workflows automatically. But the interesting cool thing about HQ is because that's all stored somewhere, you're kind of just creating reproducible workflows without even having to really think about it. Because I can just say, do that thing that I did last week and HQ will be able to find it for me. And now it's become a reproducible workflow in a way. So people, they get lazy. Like I'm, I'm like a victim of this too, where I'm not even like creating the skill always. I'm just saying like create that deck the same way that I did it last time. And now I have like created a meta workflow almost just because I have the training data somewhere.
[18:21]
A
If you've ever wondered how the top brands actually run tv, here's Tick pick on why they use Tatari. Tatari has really been like an extension of our team. They make it extremely easy to buy tv, which if it's your first time, can seem a bit scary. The client services team is incredible. The platform is super easy to navigate, tons of great insights and Atari team is always bringing you deals where they think, you know, your audience might be. So our favorite feature in the Tatari platform is the reporting and the ability to see performance at the network and even the show level. It's something that our executive team loves to see and then it helps us optimize towards, you know, better media buying in the future. That's what real TV performance looks like. To learn more, go to Nick co Tatari. Do you have a lot of any stories here about like where somebody has done something on HQ or created some kind of like a skill or, or process and then the whole company has been able to benefit from it?
[19:25]
B
Yeah, we have hundreds of examples in our company. I can also talk about it for other companies too. But I mean a good example for our company, something I'm experiencing right now is I'm like an analytical person, but creatively I'm not quite there, you know, like I can't, I can't create a beautiful deck. Like I'm not going to put beautiful assets together. But we have someone on our team who is incredible at that. And I created this one pager for something we were trying to show and it like, it looked terrible, like it had all the right numbers there, but it looked absolutely terrible. And then she made one and it looked amazing. It looked great. And now. So she made this one deck and now this is the template for all decks that we create inside of hq. And I just took that one pager I created, I told Claude, I said reference the deck that she made and just combine them together. So that's the new aesthetic. And now all of these assets that we're putting together, like partnership decks and sales decks and investor decks and everything like that, they all look amazing because they're created through her artistic eye. But everyone else gets to benefit from that. So that's one example. And that extends out to all different decisions or things created in the company. On like. Yeah, all the engineering things that we do are based on what the guidance was of our best engineers basically. And it's not just the code itself, but it's like the rules and policies that they put in place for our code base that live within hq or the way that I analyze data, which is like a little bit different. Like it's HQ is a quite complicated system and so the way that I analyze data now everyone else can adopt and look at through the skills and workflows that I've created to make sure they're looking at the right data. So it just, that just extends in many, many different ways to everything we do in the business.
[21:15]
A
Yeah. How do you see different roles within a company? Like leveraging AI in their own ways to be advantageous in what kind of
[21:25]
B
business and like our kind of business or like.
[21:29]
A
Yeah, yeah, I think like more on the brand side, I guess.
[21:36]
B
So a lot of it is around. There's like two sides to. And it almost depends if you are revenue generating or like a proper generating role or not. Like what I see for a lot of marketing teams is that they're able to generate a lot more. They're able to basically extend themselves outwards and generate more and do more revenue generating activities or people are able to do that with smaller teams. They're able to maintain the same amount and smaller teams. So it's basically more per team member to the extent that I know single people who are excellent marketers doing like two or three or four roles themselves just with AI and creating agents and that's really, really cool. So these people are doing more with less and it's resulting in growth for their company, like more ads, more landing pages, more emails, or at least being able to handle it themselves. And then when it comes to other roles, like backend roles, finance roles, stuff like that, I'm seeing of course, like it's just more efficient. There's a lot, there's a lot of efficiencies. People are getting more disciplined about data and about infrastructure. Like everyone wants a data warehouse now because they know they can have access to do all these things. But I'm seeing more sophistication on the level of analysis that they're doing. Like before someone may have just had, they were just like creating some kind of inventory forecast or basic or putting their P and L together. But now with some really young brands without a ton of resources, I'm seeing them put together super sophisticated like LTV to CAC analysis per sku, per subscription cadence because, and they're able to run these analyses that are really, that are like significantly impacting the trajectory of their brand because they have so much access to data and because if they just use Fable 5 and Claude, it's like a state of the art AI, like Data Analyst, you know, so more, it's resulting in people able to do more with less and a lot more sophistication in what they're doing as well.
[23:42]
A
Yeah, I feel like the, like there's the notion that AI is going to replace people's jobs, but AI either on its own or AI being run by somebody who doesn't have domain expertise is never going to compare to like, like you're saying these, these brands probably have somebody who's like a sharpshooter at what they do. And then being, you know, being able to use AI now they have all the leverage in their own pocket.
[24:05]
B
Yeah. And I mean I never agreed that AI was going to take a bunch of jobs like I, to me that always assumed that brands were operating at capacity already. Like everyone was just operating at 100 capacity with 100 efficiency and there was no room for growth, but I never saw that in the company that ever worked with. So I feel like these people are just doing more. They're just doing more and better. Like the silly mistakes that I Used to see people making like, putting a ton of ad spend behind products that are out of stock or like really, like really messy tags inside of Shopify. And like skews, like these mistakes aren't happening as much because the ability for AI to just clean it up for you is very, very easy. And so the sophistication that they're running at is more and so they're just doing better. But these people haven't gone anywhere. The people that can use AI, they haven't. They're not getting fired.
[24:58]
A
Yeah. Have you seen that one chart too? That's, it's got like all the dots and it's like, this is how big the world is and then the last dot is just like, this is the people who've just discovered AI.
[25:08]
B
Yeah.
[25:08]
A
It's like, yeah, we've still got a long ways to go.
[25:11]
B
We do, yeah. And I like, I think companies are just going to do very, very well. Like, I think, I think there's a lot of room to grow for all companies with the people that they have.
[25:19]
A
Yeah. What do you think? Like brands that are looking to be, you know, like, they're not looking to be in a spot next year where they feel like now they're behind on AI, you know, but maybe they're not fully AI pilled yet. Like, where do you recommend they go or look into or like, where should they even start?
[25:40]
B
I mean people, most people who ask me this question, I say to them, are you using it every day? And they say no. And it's like, well, that's, that's what you have to do. You have to open Claude code or open Codex or create a Hermes agent or an HQ agent. They should be definitely be using hq, but just like use this thing every single day and try and do everything with AI. Like, I can't imagine almost any task that I do in my day to day that I wouldn't go to one of these AI tools for first. And so it just blows my mind that people are like that. People don't know what to do or where they can work. And I've worked in, I've worked with brands, I've worked with agencies, I've worked with software companies. I've worked in like the biggest and smallest companies in the world. And there's everything that everyone does. If you set up the right systems and you understand how to use them, Cord code can do it just as well as you. And so if you are the leader of your company, you have to be doing that every single day. Once you Have a good grasp of how to use it. You should sign up for HQ and then get your whole team on HQ so you can roll this out to the rest of your team. But if you're not doing it every single day, if you're not training yourself, then there's like, there's nothing else you can really do. Yeah, it's not going to magically come to you. I think people assume that it's just going to magically come to them.
[26:58]
A
Yeah. I heard this head of HR for a pretty large consumer brand recently, basically be like, you know, we can't discriminate against people when we hire, but like, if you're not using AI and this day and age, you're basically just not fit for a role at any, you know, anymore.
[27:18]
B
Yeah.
[27:19]
A
Like, that's almost just a given.
[27:21]
B
Yeah.
[27:22]
A
But at the same time, I feel like there's so many companies I talk to that are just still, you know, so far away from any sort of real adoption other than, okay, let me, you know, take a CSV, dump it in here and ask some questions.
[27:35]
B
Yeah, yeah. I mean, even if that's where they start, like, even if that's just where they start and then they just do a little bit more every single day, they'll get good eventually. Like, I wish I could shake people and just tell like if I took the average employee and I told if you used Claude code every day for the next two months, you would probably be the most valuable person in your company. And, and I see that every single day, time and time again and people just don't do it. I think it's hard. It requires a different level of thinking. People are busy. But I wish people realized that.
[28:08]
A
Yeah. Okay, couple questions I have. So like now going back to, you know, HQ and having kind of your brands or your company second brain, you know, agentic and ready to access, how do you make sure that this thing is always staying up to date? How do you retire Stale context, You know, how do you make sure that if, like there's one thing that's wrong in there, it gets caught or fixed so that the whole company is now not operating off of this? Like how do you think about that or how does that work?
[28:40]
B
The best way to deal with that is to use it every day. Like the more people you have using it every day, the quicker you would uncover and pick up and fix incorrect context. Like if I'm doing something inside of HQ and it says something wrong that I see related to my job, I picked that up very quickly and now I can remove it. And it's kind of the same with every. It's like this, it's the same with all work. Like it's not very different to pre AI work. Like if I was creating an SOP and the SOP had a wrong step in it and then I did it and it was wrong, then I should go and fix it. And so that same kind of responsibility is on people, although that's not always a good enough answer. And so we have a couple ways of automatically dealing with it. We have something called a gardening agent. And the gardening agent basically will automatically go into hq, look for it, look, it can look within the actual message sessions themselves and try and determine this. Like if someone said something was wrong or incorrect and it wasn't solved, then it will automatically remove that. It will look for contradictory context and try and remove that. And then it will also age context based on date. And so if something is old, it can expire it or out of date, it can expire it. And it can do this with the full context of your company. Like for example, a really good example is, let's just say we work together and then you leave the company and it's like, and then it says you should go talk to Nick about this. But HQ should know that you left the company because somewhere inside of hq, somewhere inside of a call transcript, somewhere inside of a slack message or an off boarding session, there should be something that says Nick left the company. And so HQ should be able to find that and then fix that context. And that's what the gardening agent is supposed to do.
[30:30]
A
Yeah, that's pretty cool. I'm curious, like, where do you see the future of HQ going? Like, we were talking a little bit earlier today about some of the other companies in the space, but yeah, where do you see HQ's future?
[30:44]
B
The future of HQ is there's two pieces to it. One is that it should be the default option for collaborative AI work. So when we work together, when we do something with AI, there's no really great way for us to truly collaborate. Like to be truly multiplayer in that, working on something at the same time, working on a project at the same time, right? There's like, it just doesn't, it's not built for that. And so HQ should be the way to do that across multiple different agents, across multiple different people, across different teams and vendors. If anything requires any kind of collaborative work, HQ should be the default solution to that. And then the other one is that what we are collecting inside of HQ is all the Important context that you need to do your business to run your business. It has context on every single thing that you do in order to run your business. And so what we should be able to do is tell you proactively, once we recognize it, what are the best models that you should be using, what are the areas of opportunity for automation, what are you missing that you could be doing, what are agents that you could be creating and then also let you train your own models to be able to do that. We see a future where companies themselves can have their own models that are hypertuned to how they run their own language, their own workflows. And we should give you the ability to do that so you can get your edge based on using hq. And you should be able to basically create this recursive self improving loop where you do work, the model improves, you do more work, the model improves. And we should be able to give that all to you through hq.
[32:26]
A
Yeah, and I think too the other piece there is like the, or one thing to call out is it's not really the model that improves. Right. Because the model stays the same, but it's like your HQ operating system is what continues to improve and get smarter and smarter.
[32:41]
B
Well, we should be able to give you a model like we like there's going to be a future probably in about two years from now where it's going to be cheap and affordable enough that rather than having a Hermes agent running your Mac Mini, I can actually ship you a computer where a true model runs for your company, like inside of your office. Right. Like this is an actual AI model that is yours and we should be the ones giving that to you because we have all the training data to make that possible based on what we have inside of hq. And now, like this is your ip. It's like AI is more than just a tool. It's the intelligence of your company. Like, it's truly the, it's the differentiator of your company. Like, it's, it's a reflection of you and your intelligence basically as a team. And so that's important. And I think people are going to be ambivalent to give that all to the model companies eventually. And so we should be giving that, we should be allowing you to truly own that, like truly physically own that for your company.
[33:38]
A
Yeah. Wow, that's pretty cool. What kind of like AI habits or use cases do you think are going to be looking ridiculous by next year that people do right now?
[33:52]
B
What's going to be ridiculous? I mean, I think Claude, Chad and Cowork are going to go away. Like, I think anyone saying they were using cowork instead of code, like, that's just all going to collapse and go away. I think any company that's stuck to any single LLM provider, like we have companies that we work with that are just, I'm just a Claude teams company, those companies are going to get in a lot of trouble and that's going to look silly. I think that in the future we're going to see more efficient model routing and I think it's just going to happen automatically. And I think the whole thing of like, you know, if you go into Claude code or in Codex and you like select the model, you select the effort mode, you select the like plan thinking like all of that, I think that just goes away and all of that complexity goes away and it just happens automatically. So I think all these automatic toggles that we have, but I think in general, like the whole prompt engineering thing, like that was, that was like a little bit premature and looks silly now, but I think we're at a more sophisticated level now where we're not doing too much crazy stuff. We're basically just talking to the models now. We're not having to do any crazy, like, hacky things. And so I don't see anything like that disappearing.
[35:05]
A
Yeah, I feel like prompt engineering, like, I remember, I think maybe last summer I talked to Billy Howell. Have you ever talked to Billy?
[35:14]
B
No.
[35:15]
A
Billy Howell's like, he was vibe coding a bunch of apps last summer and like he was teaching me what a PRD is and how you have to get a prompt to make a prd and then the PRD goes into replit or whatever. Now it's just like you just rip a whisper flow for four minutes and like that's. That's about all you need to do.
[35:33]
B
Yeah, yeah, 100%. Well, I actually think something that might just be disappearing is all the. I think Vibe coding is going to stay and I think people are going to be creating internal apps, but I keep seeing people creating and shipping SaaS, companies that don't need to exist. And I think that's like a moment in time, like they do something cool and totally.
[35:52]
A
Also there's going to be like a hundred thousand SaaS, companies with no customers.
[35:56]
B
Yeah, exactly. Goes away. And I think people realize that there's more alpha and more value in sticking to the businesses that they're good at and just try to apply AI to that.
[36:06]
A
Totally cool. Well, Jacob, thank you for coming on limited supply and talking about hq. We should definitely we should figure out something else to do with hq. Because again, like I was telling you earlier, I feel like there's so much that it enables you to do that I still even haven't uncovered or understood yet.
[36:26]
B
Definitely should do some working sessions.
[36:29]
A
100% cool. And where can people find you?
[36:33]
B
You can find hqforwork.com and you can find me on Twitter. Jacob Posel POSCL thank you for coming on. Thank you.
[36:48]
A
Thanks for listening. We'll be back next time to cut through the noise on CPG retail and E commerce. If you enjoyed this episode, why not share it with a friend Friend. And be sure to subscribe wherever you listen so you don't miss the next one.

S17 E6: Why Your Company's AI Chats Are a Total Mess (And How to Fix It)

S17 E6: Why Your Company's AI Chats Are a Total Mess (And How to Fix It)

Podcast Summary: Limited Supply – S17 E6

Episode Title: Why Your Company's AI Chats Are a Total Mess (And How to Fix It)

Episode Overview

Key Discussion Points & Insights

1. The Fragmentation Problem with Company AI Adoption

2. Is “AI-Native” Really Happening?

3. Jacob’s Backstory & The Birth of HQ

4. How HQ Works (Explained Simply)

5. Transitioning from Chatbots to “Agents” and “AI Employees”

6. Building and Sharing Skills/Workflows

7. Real-World Examples

8. How Different Roles Benefit

9. Practical Steps for Teams New to AI

10. Keeping Company Knowledge Accurate & Fresh

11. The Future of HQ and AI in Organizations

12. What Will Seem Outdated Next Year?

Notable Quotes & Memorable Moments

Key Timestamps

Final Thoughts

Podcast Summary: Limited Supply – S17 E6

Episode Title: Why Your Company's AI Chats Are a Total Mess (And How to Fix It)

Episode Overview

Key Discussion Points & Insights

1. The Fragmentation Problem with Company AI Adoption

2. Is “AI-Native” Really Happening?

3. Jacob’s Backstory & The Birth of HQ

4. How HQ Works (Explained Simply)

5. Transitioning from Chatbots to “Agents” and “AI Employees”

6. Building and Sharing Skills/Workflows

7. Real-World Examples

8. How Different Roles Benefit

9. Practical Steps for Teams New to AI

10. Keeping Company Knowledge Accurate & Fresh

11. The Future of HQ and AI in Organizations

12. What Will Seem Outdated Next Year?

Notable Quotes & Memorable Moments

Key Timestamps

Final Thoughts