How to Build an AI-Native Company in 2026

0:00 Hello everyone. Today I'm joined by Amir, co-founder of Humalytics, an AI-native analytics and conversion optimization platform. He's had two exits already and now focuses on building and helping companies become truly AI-native. In this conversation, we break down how to actually run a business with AI from AI-powered go-to-market and ad campaigns to building a Slack-based co-founder. To getting teams to adopt tools like Claude and also what the future of AI-native enterprises looks like. So, how are you doing today, Amir?

0:34 Good. Good. Thanks so much for having me on. I'll I'm excited I saw kind of your previous lineup and you had a great guests. I'm excited to see kind of a little different different twists we can get a little bit technical and also theoretical talk about the landscape and show you some examples as well on kind of how we're using AI at Humalytics to be a AI-native company. So, I'm really excited for today. Thanks, Amir. Yeah, I mean as I mentioned to you when I reached out, typically we we talk to public company CEOs, but where we are today, I mean everyone has heard about AI and but I think there's a bit of a disconnect between what people actually think it can do and you know, what it's actually doing today by the people that are leading in in that area, which is going to affect all enterprises. So, it's really topical and trying to raise awareness of that basically. And so, as someone who's you know, deep into that, we we wanted to reach out to you.

1:23 where where are you calling from by the way? I'm actually based out of Toronto, but I spend my time between Southeast Asia and like I I I I move around a little bit in Asia, but primarily in Thailand and I pretty much escape the cold in the winter in Canada and and only come back here when when it's just just just warm enough for me to be able to go outside. So, yeah, spend my time between the two.

1:46 I I slow mad, I would like to say. Yeah. Is is what I am. >> >> so, I thought we could start with a big sort of question. Where are we with AI right now? Like what uncover the misconceptions for us and like is the hype real? and can you build a company with AI agents now? Absolutely. I think you know, I'll give you a little bit maybe a little bit of my background where we started and kind of my my history in terms of like building companies and kind of where we are today. I would say we you know, I I come from a background of like building products in the healthcare space. We essentially when I first started my career we're looking at how do we build interoperable systems between the medical records and the pharmacy systems and bring you know, electronic prescribing and you know, and better interoperability in the Canadian healthcare system. That was quite a while ago and the landscape around how we build products, build teams was quite different versus what it is today.

2:46 and afterwards I you know, did a little bit of consulting and then end up building up you know, a reasonable size business in the healthcare technology space. There was a huge gap in being able to have deep industry expertise in healthcare and technology. And you know, our at that at that time our biggest leverage in building a great business was both industry expertise and human capital. that has changed quite a lot over the past couple of years with OpenAI and Anthropic and a lot of these tools or models have come out. It's become powerful enough where as the models get better or improve, you should almost see a correlation of the output of your company and the enterprise value as well, like a 10x value of that. So, you know, you know, a good example of that is you look at a lot of the existing tools today that like Replit or Cursor or Anthropic, as the models get bigger and the companies that are using these models, they also exponentially getting better better or bigger 10x that.

3:50 So, the past couple of years after kind of building the companies and going through some exits, my my focus was on how do I build what does an AI-native company look like and how do I build a company that is very lean, efficient and effective using the tools that we have today. And my co-founder and that you know, and I that have been kind of working together and building businesses for the past decade decided that this time around we want to build a business that is really focused on efficiency, being lean and having the same level of impact that we had for the previous businesses.

4:23 So, you know, how do you get to a million a million revenue or two million revenue, but instead of or or more, whatever the number is, with as few people as possible and leveraging a lot of AI tools. You know, previously we had a business like you know, I think it was like 15 people 17 people doing about like three to five million in revenue every single year and I look back and I think you know, we're on track to hit you know, to to to hit those targets with a much smaller team.

4:49 So, the premise was how do we do the same thing with the tools that we have today. Mhm. Yeah, that's really interesting. and obviously a lot of people started using AI tools, but it's a lot more than just talking to an LLM ChatGPT. So, it's it's evolved a lot. so, it'd be great if you just just take us through these things, show us what you're doing today, you know, how do you how do you use AI as a co-founder? What does your day-to-day look like?

5:15 Absolutely. Yeah, so so I think it's important to have a little bit of premise in terms of like how where we are today and how it's evolved over time, right? We when when when when the first tools came out, it was really around okay, like how do you just leverage existing tools to augment the existing work that you have. So, maybe I'll just kind of give a a little walk-through of kind of where we the the evolution of AI itself. So, when ChatGPT first came out, you're really looking at having you know, ChatGPT or Claude being helper, right? It's like an AI or a tool that is helping you with general purpose tasks without any sort of customization. And now we see an evolution of being able to automate things for you, right? So, you give a set of instructions and context, access to external tools, especially when we started seeing tool calling and MCPs, where now these agents can access third-party tools to then do a specific task for you. So, previously you would you know, give ChatGPT a bunch of information and say, hey like you know, break down what this means for me. So, for example, you would upload your Stripe reports and say, what does my churn look like or can you forecast a churn model?

6:31 The evolution of that was now automation where saying now I want you to you know, if I every time I upload a Stripe report or connect to Stripe, I want you to do XYZ. Now we're at the part of augmentation and agency where AI is fully embedded into the core workflow with a human still in the loop and it can serve multiple functions. So, you have things you know, Claude co-work or Claude code that is very much embedded within a specific function whether it's go-to-market, design, sales, marketing, whatever the whatever the function is and it's able to work across multiple different tasks, pull data from different platforms and is very workflow-specific. And now I think we're at the latter stage and this is parallel with how would I how would classify a company or teams within a companies where they go through this evolution or this like framework of helper, automation, augmentation, agency where now you have AIs that is fully autonomous, something like open claw that is able to execute on its own.

7:33 and you know, all of this I got to give credit to Anthropic. They really kind of built out this framework and they have a lot of great resources for this. and and and that's kind of where we are today and and you know, the the framework we try to both adopt at our companies and companies that we consult with is you know, what does that look like for you? Are you able to delegate tasks to it?

7:55 Are you able to define what the output of that looks like? Is the AI the model able to actually be aligned with what you're looking for? So, is there discernment? And is the information accurate? Are you able to take full responsibility for the output that's coming out of it? and and over time, you know, we we had to kind of understand what is this evolution of how we work with these tools look like. First it started with prompting, having the right prompting techniques. Then it was around are you giving the right amount of context? And then are you giving the the constraints that's required, the guardrails, the output and then access to tools.

8:27 And I would say Anthropic is probably a leader in the space where they are solving for problems that we didn't know existed. So, a good example of that is before we would have to give a lot of context, system instructions, access to tooling or load up a lot of tools to get the output that we need and then you know, they they've been very very diligent with how efficient these models are. So, for example, one specific example is progressive tool calling.

8:57 Previously you would you would connect a tool that would load all of this information into the model for to have context on doing specific tasks for you and now you know, it progressively exposes how it accesses these tools and how it accesses information. So, that's just an example of kind of how we think through AI at a core and the the principles behind it. Mhm. And can you take us through some examples, you know, Sure. What what are the what are the high-value examples of of how it is your co-founder today? Like what what automations Yeah. Yeah. you're using that probably most other people aren't.

9:37 Yeah, so I think part of it is for me specifically. so, I work out of I work out of like a Obsidian. It's you know, we are now our like in a file system, you know, markdown-based. >> Just for the people that don't know. Yeah. Yeah, absolutely. So, Obsidian is basically it's it's a it's a note-taking app, but it's file based. So, you know, typically with Apple notes, it's within the Apple ecosystem. You're required to, you know, it's locked in, right? Like you don't you don't really own the files. But with Obsidian, it's just an interface that sits on top of markdown files. So, you're able to create markdown files and view it in an interface that is, you know, readable.

10:18 and you own the files, so you can take it anywhere. It's on your MacBook or on your Windows, whatever the case is. And this really works quite well because it's, I guess a common denominator or language between a lot of the models where they interface with JSON, text files, and markdown files. And for me, I primarily use Obsidian for notes even before, you know, LLMs were a thing. and use Claude Code on top of it to, to kind of enhance my work. Mhm.

10:45 So, yeah, one example is, I I shared an example, a while back on Twitter was using Claude Code and Obsidian and third-party tools to be able to automate, paid marketing campaigns and outbound email campaigns. So, an example of that is, being able to connect to third-party tools like Clay, Hunter, Apollo, whatever it is, to pull lead data, enrich the lead data, build out campaigns specifically in there, and then have follow-up emails. So, and a good example of that is, >> >> we we set up to create an email campaign for, for reaching, you know, for Humbletics.

11:29 So, we're AI analytics and CRO tool. We want to partner with agencies. We want to go after marketing teams and and enterprises and startups. And have an email outbound campaign to build awareness around it. So, what I would do typically is essentially load up Claude Code within this workspace specifically. So, I'll show you an example of what that looks like. So, we're going to, essentially open up the the file here and load it within Claude Code. So, this is a workspace that's going to open up in Claude Code.

11:57 One second. Cool. So, what I've done now is essentially I have Claude I'm going to I've opened up a terminal that's going to give me access to Claude Code within my Obsidian workspace. And what I can do is, manipulate, access third-party tools, and add information to my existing Obsidian folder. So, what I'm going to do is say, you know, study this workspace, outline exactly what the email folder has. So, to give you, to give you an idea while that's happening is I'm going to go back to this Obsidian folder right here and show you kind of what I previously have done so you have an idea of what is possible. So, we wanted to essentially build out a lead list of different agencies that are currently focused on building Webflow products.

12:53 So, basically like, our analytics tool works really well with Webflow. It's a third-party Webflow, website builder. And I connected to Hunter to pull lead data from Hunter. It's a third-party tool that, does lead enrichment to actually find specific agencies and other companies and that use Webflow. And it was able to pull their specific emails, their positions, their names, and the companies that they're at using this specific tool. And so, how did you know that they were using Webflow?

13:27 So, you can actually use, well, a couple things. So, Hunter has built-in, capabilities that's able to detect tools. So, what they're looking at is a lot of like third-party, website builders actually have, like a code snippet inside of it that says built by Webflow or built with Framer. So, that's what it's typically looking for. Okay. Yeah, so and then basically, what I did is Claude Code will actually pull all the leads and contacts, at these specific companies, and then put together a strategy on how to actually target these people. So, who are the personas? What does the competitive landscape look like? And then what is the email sequence actually set up?

14:08 what should it look like in terms of reaching out to them? So, this is all done through Claude Code directly connected to my Obsidian. So, what I typically do is use Claude Code to build out a strategy and then from there hook it, have it mapped out in Obsidian, and then go and execute it in in actual in actually in, in hunter.io. So, I'm going to give you a I'm going to show you a sandbox of kind of what that actually looks like so that you can see how, Claude Code builds out the the the sequences essentially. So, this is what we have so far. We have the leads and contacts. We have the strategy and the emails. And then I would typically take this and connect it to, to the sequences in Hunter. So, we're going to go in here. And what we've done essentially is I haven't launched this yet. this is what I've been just set up. But what we've done essentially is we're going after agency owners.

15:00 And it built out, built out, a sequence, a set of like, emails that it followed up. And this is all done programmatically, connected directly to the the emails, to to hunter.io through the API, and, sent out the emails. So, maybe this is not a good example. Let's let's find a better one. So, yeah, so Hunter, has the capability to email. That's how I expected. Yeah, yeah, are you are you familiar with with Hunter or not? I'm familiar with Hunter. I've used it for, finding emails for people, but I didn't realize that they've they've gone into sort of, sending the emails as well.

15:35 Yeah, yeah, absolutely. So, you can you can find, specific lead data as well, but also send out email sequences. Now, this is where, you know, I actually haven't sent this out yet. And I've I've ran previous campaigns. This is a example one I just had as a draft. Mhm. Where I would This is where it's important to have the human in the loop process, right? Because it's like you don't want to just get AI to run this end to end and to end where it's finding the lead data, building out the sequences, and then sending this directly. You want to actually be involved in reviewing this. Like I personally I don't think I would send an email style like this. This This is probably not the best way to frame this. so I would go in and actually edit it and then make sure that, you know, as we're targeting these people, it's it's relevant to the audience. Like it's the the context is personalized. Because outbound emails is hard as it is with or without AI. And if you're automating this whole thing without really having full understanding of it and like no visibility, I I I I highly doubt, it'll be very successful. We We've typically ran campaigns.

16:33 Yeah, we're not there yet. Yeah, yeah. Okay. Yeah, so this is just one component of it. And really for this one, I have less expectations on, like on the enterprise side, I would probably go through LinkedIn and build relationships with the key decision makers at like a senior level. this is more so for visibility for top of brand where it's like I want them to have visibility and then I will probably run a middle funnel campaign where I'm running paid ads on Meta and not on on Google so that when they're searching for Humbletics, you know, they're being targeted through a Google ad campaign. And I'll show you what that looks like as well. Yeah, that'd be good.

17:10 And just just going back to so the Obsidian What was in Obsidian, the note taking tool? Has that got your target market, details about the business? Yeah. Exactly. So, all all that context. So, it's basically using that as the base for creating the campaigns. Exactly. So, I I have a folder called a contacts library. And basically how it works is it has a breakdown of everything pertaining to my different like projects and businesses that I'm working on. So, what I do is actually use AI to build out, a set of contacts files that, I would use within Claude Code to say reference this contacts folder so you have a full understanding of our business and what we do. It's very similar to ChatGPT projects and ChatGPT or Claude projects where you have both system instructions and contacts files that give you all the relevant information you need or you want to provide to your agent to be able to actually, yeah, like, to be able to ensure that it's giving you a good output. It's aligned. MD files is just it's very easy for Claude Code to read, basically.

18:12 Exactly. Yeah, it's, text files, markdown files, JSON is a very good readable format for these LLMs to be able to actually understand the data. Yeah. Cool. So, we've covered We've covered, what the email side looks like. And I'll show you an example of how we also set up our, paid campaigns. So, again, this is These are just all sandboxes. These are just, for like I want to give you an idea of like how we map it out. but like we'll use also Claude Code to connect to Meta and Google and map out exactly kind of what the campaign launch strategy should look like. So, this is just, again, these are very early stages of everything that's happening. So, I I would take it with a grain of salt for anyone that is saying they're very successful with the these kind of tactics. My my purpose or my intent of this is to show what is possible where we're seeing a convergence of both technical and non-technical people in different functions and how they can use Claude Code as the key interface for work in the terminal to build out all of this.

19:18 So, I don't want to, hype it up and say, yeah, this this is this is driving outcomes or this is the way. I just want to show what is in the world of possibilities. And you as the expert, whether you're a go-to-market expert, a paid media expert, can use these tools however you see fit to enhance and augment what you what you exist what you already do. So, for example, we use Claude Code specifically to map out campaigns for, Meta. And what each campaign would be kind of essentially targeting in the audience and then even created specifically like like a audience library as well. So this is all done programmatically through Obsidian and then we will connect to meta directly to build out these campaigns without having to actually manually go and set it up. So we're programmatically doing this in a way and you can even use other tools like Manas to pull information. So this is just an example of like how the campaigns were built out in terms of strategy and then we would have a prompt library.

20:19 So are you just you're asking Claude code to build the campaign in Obsidian in an MD file? Is that >> Exactly. Exactly. So the steps were >> can be sent via API to Manas or a connection. >> Exactly. Exactly. So how how it works is basically you can use meta APIs and Google APIs to connect directly to Claude code through the MCP and programmatically set up those campaigns. So so you can essentially create both programmatic assets, programmatic ad strategy campaigns and everything in there and as a human in the loop you're reviewing the work and then actually making it go live. Yeah.

21:00 So I'll give you an example of like for example how we created programmatic ad images for our for our meta campaigns. Now this is something that for example >> >> you would have a multiple team members creating or being involved in process for where they would actually have to create the image assets, create the copy and then test through it. So we're doing this programmatically where we're saying here's the context around our business, here's the campaign, here's all the assets that you have. So these are like UI images and use these to now create an internal tool for us. So this is more of a internal tool that we have that creates different ads based on the like the ad vertical categories. So like case study, retargeting, competitor, product demo, pain points, social proof. So these are all programmatically created all through Claude code using as existing assets that we have instead of having to have a designer design it copywriter write it. So you can see exactly the headline, the primary text, the description that we can then programmatically load into meta.

22:09 And this is this is a tool you've built yourself essentially. Claude code did. I mean I just basically how we prompted is Claude code, here's context about the business, here's all the assets, build me an ad library tool where I can view all my ad assets in different categories. Mhm. So you just yeah you just add images some campaign copy and it's just pulled them all together into different formats relevant for Exactly. So you know maybe maybe we'll show an example of what that will look like. So we can go into Claude code here and say I just need to reference the file. So in the meta folder we have ad assets and ad viewers.

22:55 Review the existing campaign strategy and content, create five more ads. And basically what I'll do is I'll reference all the existing context within this workspace folder across the prompt library, context of what the business is the different existing assets that we have and go ahead and create any existing assets. So I think we we previously had about 38 ads. So while that's happening we'll go back and see Actually let's see what it works through. So what it's doing right now is going through all the paid existing paid meta assets.

23:33 So we have 38 ads in there already right now and it's going to now most likely create five more ads in different categories within this ad viewer. And how does how does Claude view an ad as an image and understand what the image is? Is that it built into Claude or are you using something else or So these are all programmatically created with code. So the images are actually HTML HTML code and CSS which you're able to download as a PNG. So it's actually it's HTML structure. So it's understanding it from a programmatically from a code standpoint.

24:10 Right. Yeah. So we have about you know we have the ad library in here. So it has a complete understanding of the design system, the ad structure and the view registration and it's going to create five new ads. So let's switch over to the ad library and see what it creates. So we have about 30 ads so far. We should probably see a couple more down here afterwards. It's still working through it. While that's happening do you have any questions for me?

24:39 yeah. So well one is actually about the slack bot and how you use that but maybe we'll do that after this. We'll do that one afterwards. Yeah. Yeah the the the scraping like research you do that's probably a big a big thing that people can save time doing. I know there's a lot possible now with Apify and Firecrawl stuff like this. Exactly. What sort of scraping do you do websites? What's the most what's the most valuable sort of things that people can use it for?

25:08 Yeah. So Apify and Firecrawl are both great tools to kind of extend existing agent capabilities for scraping. And what that means is essentially like if you're ever if you don't want to ever leave your existing Claude code interface and want to just get additional information whether it's brand assets scraping context and even for example what I showed earlier with the context library. So building out the LLM.text files which is an overview of a business. I would use Firecrawl to actually go to a website and scrape content. So for example if I wanted to build an overview or a context library for my existing product I would say hey go to this website using Firecrawl, scrape all this content for me and then create an LLM.text file, create product messaging text file for me.

25:59 So that's kind of kind of the the value add when it comes to Apify and Firecrawl. And competitive intelligence things like this I'm assuming you do this. Exactly. Exactly. So right now we're we'll go back to showing what Claude code is doing. So right now it's actually creating a bunch of new ads. So it's now building it out using code. So it's not actually generating images but using HTML to create the images that we will then install we will then download later on.

26:31 So it's essentially yeah creating ad 42. It's doing a competitive one. yeah it's it's looking at different categories and then programmatically creating these ads which we then take and load into into meta. And just going back to the scraping and research. What what do you think if you have what's the most valuable way of using it if there's one thing people could start using it for? for for enrichment, right? So or gathering lead data. So for example you want to do research on you know find me the top 50 for us specifically it's like find me the top 50 agencies that are using tools like Waffle and Framer that is within our target ICP. The the agent is smart enough to know where to scrape, what to look for and then actually find that and and and and like find that information for us. Mhm.

27:28 Cool. So we now have So we had 38 ads before we have now 43. It went ahead and programmatically created these ads for us. Now like I would again as a human in the loop I would work on refining some of this, right? There's a lot of text on here. I would then maybe also serve other example of high performing ads, you know give it a library of ads that like here's really good meta ads. He uses reference to refine these existing assets. So we've now just programmatically created five new ads that would have taken you know a day or two from across multiple teams to to create.

28:04 It's amazing. And so with yeah we're talking about human in the loop still. Do you think almost everything still needs that touch point just to especially just to improve I think but probably systems that you're putting in place? 100% 100%. I would I would definitely always be involved in every single process because again it comes down to accountability, right? Like I'm responsible for the output here. I'm not going to have an agent fully go in and do everything end to end without me being involved because I want to take accountability for what I'm doing and be I'm still the expert here when it comes to some of these things. So it's like I want to be involved in that process and give it the refinement it needs.

28:46 So like today basically we're saying it's removing lower end admin sort of or you know lower end junior roles that doing things but still need that strategic you know Yeah. Exactly. You're you're you're still the you're still the like expert and it's really augmenting your work, right? Like it's it's a product it's a productivity multiplier. It's interesting like I probably work more now and have a higher output than I did before like with these AI tools. It's it was supposed to save me time but it's actually I'm working more and but I have a higher output for it. You know what I mean?

29:23 and so if execution is not is no longer a I mean if we just like pull this out over a few few six months or whatever a year. If execution is no longer a problem especially with the companies like you that are using it to the you know to the top of its potential at the moment or at least like not as much of a problem as it used to be. >> >> you can run campaigns, generate creatives, you know optimize in real time Like that you can get those feedback loops going as well.

29:50 What is actually limiting What is limiting performance, you reckon, in in the our systems today? At least at the companies using us to to the top their ability. Yeah, so this one is that's a really good question. So I think actually it comes down to company culture and enablement, right? which is, you know, I'll reframe the question better. It's like how does a company become AI native or how do they get the best uses out of these tools? If I was, you know, considering your audience, I would say at a senior level, I would if I was sponsoring something like this, I wouldn't really focus on, "Oh, how many chat GPT licenses do we have?" Or what does token output usage look like? It's more around are we giving people the right tools and resources to be able to learn how to actually use this? To stay top, you know, to stay ahead of all the the landscape is changing so fast. So it's like there's always new tools, new resources, but are we enabling them to one learn and the sandbox and the environment to actually play around with this? And then two is as as a leader, am I championing am I a good champion to say like, "Listen, we should use these tools and we're not going to get we're not going to let company culture or bureaucracy get in the way to slow you down." You know, I I I I I worked for a venture lab for huge huge multinational financial services insurance company and this was at the peak of early of like chat GPT and AI and I remember I had helped build a website for them really really quickly that would have taken months and a quarter million dollar budget to do.

31:24 And despite that, they still had to go through the steps and processes to get her approved, which I understand, you know, that's the case here. But it's like you really want to take away the friction to be able to give your team enable your team members your your colleagues and your team to actually use these tools without having to go through the the red tape. Mhm. Yeah, but there's definitely it's almost like an anti-AI movement, I think, that stops people want They don't If they don't want to adopt it, then, you know, it's going to hold you back from from from seeing the value in it.

31:54 >> Yeah. And there's definitely a sort of at least in the last 6 months, it's almost like a lot of people are saying, "Oh, maybe the hype's not It's all hype. It's not actually doing what people are saying it's doing." But a lot of people don't understand how, you know, the people at the top of the game are using it. yeah. Yeah, I think a byproduct of like sharing our journey like building AI native company has led to a lot of companies coming to us and say, "Hey, like, you know, how can we work with you to like consulting for AI native?" And I and I work with a lot of existing companies today. And when I said sit down with a lot of these leaders at like at different functions, I say, "Listen, one, we shouldn't do this company-wide across the board.

32:32 Let's find a very specific team like usually we do these like AI fluency benchmarks where we look at a team that is going to be the most open to trying something new like this and say, "Let's focus on a team like your go-to-market team, your marketing team build a mandate around what AI will look like, assess where they are on the fluency scale of like how they use currently use AI, the problems they have, and how they want AI to solve for it.

32:58 And then let's just upskill them, enable them to understand what is in the world of possibilities. So a lot of consulting I do today is not around going in and implementing an automation or building out a specific agentic workflow for them, but just showing here's what you can do. Here's how you can use it. Here's what matters. Here's why prompting matters. Here's why context matters. Here's how MCPs and tools actually work. Let's build something together and then give them the the the sandbox and the environment to go and build something afterwards. So success is around for them to be able to do something without my direct involvement. And and then most importantly at a senior level, give them the sponsorship and support they need to do something like this without any red tape.

33:41 Do you think it's fair to say in the world of today, the biggest competitive advantage would be using these tools to the top of their ability in whatever firm you're at? Yeah, absolutely. Yeah. Yeah. Yeah, absolutely. I think yeah, it's it's both internal and external facing. So maybe I'll give an example of what that is and like what we're seeing. So we're we're really trying to look at everything that we're like our what we're building with our companies is a test bed for what we think is going to be the future. So I'll maybe I'll just talk about kind of what I'm working on right now a little bit to give you an idea of kind of where we think the direction where things are headed. So So a couple things. One is you know, we we actually have built our website using first Webflow and then we migrate to Framer. Like a you know, for for those that don't know, it's like a website builder. And then one of the challenges we ran into was we wanted to programmatically create pages based on the campaigns that we were running to speak to our audience. So meaning that we wanted to be able to say, "Okay, like, let's have a dedicated page for the director of marketing at these at this specific at this specific industry and have the messaging on the hero section tied to that."

34:55 It became a bottleneck for us to be able to create these pages really quickly in Framer because one, we had a dependency on a Framer developer and just the turnaround time to be able to experiment and run these experiments. So we ended up actually migrating to a custom code site and we said, "Okay, the biggest challenge we have is CMS. The most that a lot of these website builders have is obviously the dependency that's removed on developers, but also the CMS where they can just create content at scale."

35:21 With the rise of agents now and the fact that they can be accessed through Claude, through Slack, through Cursor, your agent is the new CMS. The whole thesis is that moving forward, you can programmatically use agents to create pages and manage your content all through as your CMS. So like this is a service we we actually launched through one of our like through our holding company, which was you know, being able to actually migrate your existing site from Webflow and Framer to a custom code site so that you're one, removing the dependency on developers completely and using the agent as your CMS so that you can quickly and programmatically add content, update content. This is like perfect for a lot of founders actually.

36:05 Like we work with a lot of founders that come to us and say, "Hey, like, can we quickly change the the heading text on this like website?" They don't even still want to go into Webflow and Framer. They just want to be able to talk to an agent to say, "Hey, change this for me." Right? And also like we're seeing now, you know, just yesterday Stripe announced MPP, mission payments protocol, where agents are now actually interacting with third-party sites and apps, creating accounts, paying for services. Like the future is probably going to be agents interacting with services and websites just as much as humans. Mhm. Getting information like Firecrawl, using Firecrawl to scrape data, or having an agent go and sign up for a product on your behalf like we saw with Open Claude. So we're seeing a lot of infrastructure for authentication, payments, and email. Like we use something called Agent Mail to have an where our agent actually has access to its own inbox.

37:00 you know, you want to you you're going to want to create websites that in the future that are agent-friendly. So for example, we've talked about how agents are really good with seeing markdown files where behind the scenes, you know, we're going to have new endpoints where agents can just go to a URL, hit a very specific endpoint that gives them a complete breakdown of here's what the product is, here's how you can create an account, here's how you can access your API, here's what the schema looks like without having to like crawl every single page or try to find where to go. You can essentially have a a markdown that has access to everything that you need. And we think that's the future of websites.

37:36 Mhm. So yeah. Yeah, I mean most people are like it used to be the gatekeeper, well, is either a CMS or a developer. and yeah, it's just far simpler just to ask it to do something, isn't it? For anyone who's who's used Claude code for coding, they know that this is just a you know, you you can literally just ask it in the terminal and it will it's smart enough to know where you can just drop in a photo of where you're, you know, where in the code or the on the website that our website you want to do it and it will smart enough to know that you the heading is there and it won't, you know, you'll to change the copy on there and takes seconds rather than, you know, minutes, whatever.

38:16 Exactly. So it's like what is the what is the new UX or UI for that look like? How do you build, you know, it kind of goes back to like early in my career. It's like how do you build into operability between all these different systems to have a readable format between them to be able to exchange information. And this is a a very similar parallel to that. It's like how do you create a new common language, you know, and you're seeing this with Cloudflare. They they came out with a new crawl endpoint where you hit, you know, you can give an API endpoint to hit to crawl everything relevant about your website. And that's kind of that I think the future where you're going to have a where we're already there right now is agents will be quickly access this information.

38:54 And then yeah, another example of that is, you know, you talked about kind of internal Slack agents. I can't specifically show a demo of like our agent because it's like private client information that it's going to access, but I'll give you an example of what it looks like, which is, you know, we needed one of the biggest gaps I always had was reporting. Like being able to get a daily report of how much money did we make? What does our funnel look like? You know, are we losing money? Are we making money? What does reactivation expansion look like? So we ended up building a custom agent at that time before Claude code came out where it would pull information from Stripe, Chargebee, all these different tools and then give us a report. And we ended up actually productizing that cuz we had other companies that came to us and said, "Hey, we need something like this as well." Where you can essentially have a chief of staff in your Slack. we use Slack, but it's like our primary tool, but it's essentially an internal agent built on top of like cloud code where it accesses all the different tools within your organization and you can connect it to Stripe, email, Notion, perplexity, everything within your existing stack and have it be as your chief of staff.

40:05 So, whether it's pulling reporting, being able to do like automations on scheduling like here's three tickets you should know about, being able to create content for like let's say you know, investor meetings for example and you know, having your stand-ups set up across your different teams or even find information like around leads when you're within your existing pipeline. We use the tool religious religiously inside Slack. We even have a dedicated channel where we've created sub agents where you have multiple Slack agents actually working together and then and mapping it out. So, now programmatically we have agents that are creating pages for us on our website. This goes back to my earlier point about having custom code built sites where now it's going in directly to our repository looking at our top of funnel data within our analytics tool and then creating a page based on the pipeline of like users that are coming in and making it more personalized.

40:59 Yeah, it's really Yeah. Interesting to see how how things have moved along even over 6 months. things have become >> Yeah. possible that wasn't before weren't before. Where if we just look in the future what does the enterprise of the future look like? If I'm say and then cuz of things are moving just so fast in a year's time what how how are we are do we still need a human in a loop in every step or are some tasks you know, we can trust AI and after to do it autonomously?

41:32 where the where the the LLMs going to advance to and what do you see as the future looking like? I actually think we're going to pay a premium for a lot of expertise in the space, right? I would happily love to like have team members that are you know, we have fractional team members that work with us, but they're experts in what they do and we you know, that's the value or the premium we're paying for, right? Where it's like you're the human in the loop. We're giving you the tools you need to get the work done to save time where you don't need to be wasting, but we still are looking to your expertise and your industry knowledge to stamp your name on it and say, "Yeah, I you know, this I'm accountable for this for this output and this outcome here, right?" We're seeing a shift where these tools made us more productive and now we want it to be outcome oriented. I no longer want to just s- you know, be more efficient or have save time, but I want it to actually drive outcomes. And I think that's the the the North Star a lot of enterprises or teams should look to is how can these tools actually be outcome focused? It's cool. I mean, it's cool that I can create an email campaign or an ad campaign in Obsidian and use cloud code and save all this time.

42:41 But is it actually working? Is it saving me time? Yes, but is it making me more money? We're still we're still figuring that out. And I you know, that's why even earlier I said listen like we're doing this, but I don't even you know, I'm still experimenting. but I want to show what's the what's what's in the the world of possibilities, right? So, for enterprises you know, I I come from enterprise background. I worked in big companies. I would say silo teams that have no red tape and bureaucracy and a good champion to get them to do these things will do really really well and you start small and then you build you know, you you bottom-up approach. And then on the flip side where I am now, I truly think you can build 5 million, 10 million, 20 million, even 50 100 million dollar companies with being fully AI native with as few people as possible using these existing tools.

43:32 Yeah, wow. And you will start to see those sort of stories over the next 6 12 months come out probably. We're we're already seeing it. We're seeing a lot of consumer-facing apps that are solo founders or just a small fractional team, no funding even now >> Yeah. they're selling for millions of dollars, yeah. Yeah, that's what I mean. It's incredible really. and what what's your opinion on in the future who's who's going to win the foundation models or the applications or is it going to be more of a mixed?

44:04 Yeah, that's a really good question. People talk about it all the time, right? And it's it's really hard to to to know and it's just be good to get your insights on it. I mean, just for people who don't know the foundational models we're talking about Claude, we're talking about Open AI etc. and application layer are things like Humble and Analytics Yeah. Yeah, so yeah, so I think the yeah, from an application layer standpoint the companies that are building on top of this I mentioned earlier like as the models get better better by 5x 10x, the applications should as well. Like a good example of that is Replit. I think Replit you know, for people that aren't familiar Replit used to be like a a code editor and it had some really like multi-collaborative tool tooling and features. When Open AI came out and Claude came out, they added AI agent capabilities. The company's revenue like 5x 10x 20x as the models got better, their revenue you know, went up. They went from like 5 million to 400 million ARR for example. So, you're going to see that across the board with a lot of the application layers. To your earlier question around who wins in this yeah, I think I think there there is no moat. The moat is speed, execution.

45:20 speed and execution I would say and but I I do think that personally Claude is winning. They they they Anthropic is winning. They focused on enterprise from the get-go. I I personally used all the tools and I I always advocated for Claude within the workspace just because of how focused and the fine-tuned it was for getting work done. and and they seem to just they they figured it out. They really did from both coding application standpoint, from work, everything around it. It's less consumer focused and more work oriented which is really what matters here for for me at least. Yeah.

45:56 And do you think there'll be a time where the models cuz I mean, feasibility there's no limit to the intelligence that they can get to cuz I think in a moment they're sort of genius level on specific tasks at least when you you know, when you do these tests. but it they're not like human obviously they can get smarter. So, in a year's time they're going to be a lot smarter. How what on earth in I just don't understand in the future how any human could be smarter in a specific area or how they how do they maintain an advantage if some of these models have access to an insane amount of sort of data, right? In a specific area but let's say marketing.

46:34 Mhm. And they're smarter at synthesizing it and coming up with ideas. Is that is that a future that exists or is or like I don't know, it's crazy to think about. Yeah, I I I don't think I I I have an answer to that question yet to be honest. I'm trying to that's a that's a question I think about every day as well. Like it's like ex- ex- ex- ex- ex- existential crisis for myself where it's like am I going to be am I going to be relevant today or tomorrow and you know, maybe I have like a 2 3 year window to just make as much money as possible and then go live in a yacht. I don't I don't know. that's a really good question.

47:12 I think for now we're safe probably and then we might we might revisit this in a year from now, yeah. Yeah. Yeah. We'll check in. Check in in a year. >> check in. Yeah, yeah. cool man. Yeah, I mean, this has been great. Thanks so much for taking us through you know, all these different areas that people can use today to get more out from their work and like you said they obviously need to dig into it themselves and see what works best for them.

47:40 maybe before we wrap it up if you could just tell us like what what should someone work on first or build first to to try and become AI native? Yeah, I'm listen, I I would say again, I I work with a lot of companies on AI consulting with a lot of C-level executives that had no technical prior knowledge. I've tried to teach them cloud code maybe 6 months ago, 7 months ago and they were still very much hesitant to use it like opening up a terminal.

48:14 But now that I actually had lunch with someone a couple days ago a CEO at a pretty big company and he's like, "I built my own app. I built my own Slack agent." You know, and I was so fascinated. I was like, "What changed over time?" And it was the fact that the models have gotten so smart to be able to debug and just get them going really really quickly. It's removed >> distance of like it went wrong and everyone's just like, " can't bother with it anymore." Exactly. Exactly. Like the friction point of a terminal was so scary to them, but now it's not anymore.

48:41 So, to your question, I would just say the best thing to do is just download cloud code, come up with a specific idea or or problem that you're trying to solve for and then work backwards and use cloud code to learn and understand how it's approaching solving the problem and just build something. That is the best way to do it by doing and you learn by doing is what I would advise. And is the terminal the future?

49:06 Or do you think so? Yeah. Yeah, yeah, absolutely. You know, I I had a thing before I used to say cursor cursor was the interface for work. I would I take that back now. I think the terminal is the interface for work. Okay. Cool man. Yeah, thanks so much. I really appreciate it. I don't know if you want to say anything before we wrap up. Yeah, I know. I mean, if if anyone's curious about what I do, you can just find me online Amir MXT a m i r mxt.

49:31 Cheers, man. And have a great day. Thanks for the opportunity.