Open vs. Closed: Why AI Belongs to Everyone

0:00Every week, it seems like the new tools that drop are just mind blowing. Yeah. And it's just a nonstop now. That rate of acceleration, you can feel it. Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Page, and we're here to demystify AI for you so you can get some value from it. Let's talk some AI. Today, we've got a special guest, Robert Ranson, director and founding member of the Agentix Foundation, also founding partner of Human Race, an agentic engineering firm. Amy's an author too, the AI dividend preparing for a post-laver economy, all kinds of, you're a renaissance man, Robert.

0:41But for those who aren't familiar with the Agentics Foundation, it's an open foundation dedicated to building scalable, intuitive, and human-centered Agentic AI systems and infrastructure. And it was originally founded by Reuven Cohen, who hands down is my favorite person to follow in the AI space right now. Robert, first of all, welcome to the show. I think you'd agree there on the Reuven side. The dude's just insane. Yes, yes, absolutely. Thank you. I'm happy to be here, Matt. And it's always a lot of fun to have these conversations. And Reuven Cohen is his own singularity. That's probably the best way to describe it right there.

1:19And so Robert and I, we were both kind of early members in this small, random community that Reuven had started. and it began as like just this Friday meetup online called the AI hacker space. And essentially Roof would jump on, blow our minds with something new he'd built that week and, or maybe just had AI building while he sleeps, which seems to be the standard now. And then he would open the floor and then others would share their projects. And it was just this really cool community that started. And what kind of started is like this casual gathering has grown like wildfire so much so that it's like an official nonprofit organization.

1:58And we're going to deep dive more into the community. Hey, I think it's the best one anybody out there that's interested in AI can be a part of. But I also want to get into this central topic around open source first closed source, right? Because it's one of the core principles behind the agentics foundation. At the core of the discussion is that, and I want to start Robert, just give me your, what is your definition of open source versus closed source and your stance on it? Let's start super high level and then we'll rabbit hole, dive down some interesting parts. Sure. I mean, at its core, I think it's not trying to retain control over ideas.

2:36And that the tools that we build out around the concepts are meant to be tools freely available to society to run with. And I agree with you, Matt, that's from the very beginning. And actually next week or on the 26th of April is our one year or 52nd live event. Is it really? Yeah. So that's the actual calendar year. I'm trying to remember back when I, how I found, I think it was maybe just some random like LinkedIn post. I wasn't even connected with Ruv and I just randomly saw it and I was like, oh, that seems interesting. And I jumped in it. It's just, it's evolved so much. Actually, let's stay there.

3:19I'm curious, like, what is your story of getting involved in what was the AI hackerspace and now the Agentix Foundation? Like, how did you meet Reuven? Kind of give me your story, your perspective as this thing's grown into this, like, how many people are in it? Huge. Yeah, there's 1 ,200 members in the WhatsApp space, and that's capped. So we can't invite any more folks in. Occasionally, somebody will leave and somebody else will come in. So folks, there's a little bit of rotation, but it's capped. That's why we've also now opened up the Discord side. There will be a place for the community to grow.

3:53There's over 85 ,000, I think, followers on LinkedIn. There is, or no, that's 120 ,000. And there's at least 65 ,000 in the Reddit over several Reddits. So it's an audience that's pretty much closing in on a couple hundred thousand. And then there's a huge long tail audience for those live events. So every week, Ruben does a coding session on Thursdays. And then we have the open forum demo and new tool drop integration conversations. that's every Friday. And then those get recorded. Our friends over at Kaltura have that archive available. So you can go back over the last year and look all those different coding sessions.

4:34But for me - That'd actually be a fun exercise, Robert, is go back to the first episode just as like the throwback. Yeah. And see what we were talking about because I'm sure it'd blow our minds just how far things have progressed. It does undoubtedly because every week it seems like the new tools that drop are just mind blowing. and it's just a non-stop now that that rate of acceleration you can feel it like today for instance i was eating lunch so i jumped on a little bit late but rufin was building something what was it it was like being able to find rare earth metals or something i don't know what it was but it's something insane leveraging ai today was one of those moments matt where even his buddies are like wait i think you're crazy quick break in the pod if you're listening to this podcast, chances are you've been thinking about how to actually use AI inside your business.

5:26And that's exactly why we built the AI Opportunity Finder. It's a free tool that helps you uncover high impact, tailored AI use cases based on your business, your goals, your pain points, and your industry. No fluff, no generic use cases, just real ideas that fit your business and the rank by ROI potential. It takes about three minutes to run and it's like having your own personal AI strategist for free. If you want to try it for free, check out the link in the show notes or go to hatchworks.com backslash AI dash opportunity dash finder. You know what I mean? Yeah. But where a concept idea runs through deep research, you came across an article, which is a lot of scientific work has been going into quantum signals of magnetic.

6:11It's a, how do you even describe it? Magnometer or magnetic fields with the science being that there are fingerprints of each layer of material and item and concentrations of metals and all that kind of those all have fingerprints it's just that we have not looked through the right set of tools at the to find those fingerprints much it took us a long time to figure out gravitational waves and this is it speaks just so potently to the era that we're in because now an incredibly talented engineer like proven with these tools can utilize a recursive, like a chain of deep research to really explore the entire footprint of that scientific body of work and then hypothesize about what a code base that utilizes the science and tests for it would look like, and then build it and then run demos within 24 hours.

7:10So it's crazy. It is crazy. It means that frontier scientific work around the earth, around the entire planet can be identified and have tools built around it very quickly. That's acceleration. That's what it feels like in civilization. When those kinds of advances can be made, discovered, tested, validated, and pushed forward so quickly. Yeah. You mentioned something and folks want to check out, like you mentioned, the recordings are out there and you can join the community as well. We'll get into some of that later but you mentioned the thursday sessions those kind of came later this is like that's right reuven's call it vibe coding sessions where you literally you can just be a fly on the wall and see him vibe coding and you can see his how he thinks about his stream of consciousness as he approaches it and he said something today that i think really resonated with me he talked about like the idea of vibe almost being like on a spectrum because there's one spectrum where it's Like you're literally just starting from nothing and you're just like, it truly is like vibes based.

8:15You're letting AI guide you. And then to the other end of the spectrum truly are having a detailed plan of what you want to do and build. And there's varying degrees of where that is. And then he's also built the Spark framework too, which I don't know how much you've played with that, but that's essentially his methodology for using AI to build. Yeah, specification, pseudocode architecture revisions, and then convention. So to work with AI coding assistants in order to give them an overview of what it is they're about to build. And it's, yeah, it's really, it's wonderful too, because I think Reuben first really formulated that back about eight months ago.

9:01And it's been adopted now. Now, when you look up RUMO Boomerang or custom RUMOs, it's coming back for Plexiglas, it's coming back with that, with the Spark framework. So really hats off to the work that Reuben's been doing in this space. Agenic engineering. Define that for us, because I don't think a lot of people either know what that is or even the term agentic. I don't know about you, but spellcheck still says it's not a word for me. Right. I'm like, easy. They'll catch up. So, agentic engineering is not the process of building agents. Agentic engineering is the process of building systems that build things.

9:38Right? Yeah. So, agentic means it can interact with information and then, based on a goal set, make decisions about how it's going to respond to that information. So, it takes action. It is agentic. It has agency. The key point is that there's a touchpoint with information that's incoming in order for it to formulate it. It's not an AB. It's not a binary decision. that we can utilize large language models to give it a space of logic to be much, much more powerful. But genetic engineering is literally building systems that build things, right? And that's what we've been doing for the last year is really exploring the space of what engineering looks like with artificial intelligence and what kinds of things that you can do with that.

10:23And you still hear an awful lot online and LinkedIn and everything. Oh, it's not there. This is really important. And Matt, you put your finger right on it. The vibe coding to eugenic engineering and that spectrum, what that looks like, is really whether or not you understand what's being built and you've planned for what's being built and when it's going off the rails or not. A genic engineer is somebody who's learned the space, the architecture, or is a previous computer science engineer, a developer. And they're not just having the coding assistant go from scratch on its own figured as you go.

11:03That's the vibe coding space, which is beautiful because literally millions and millions of people around the world and growing every single day are diving back into. software and what they can create and discovering this new tool. And the ease of developing software is coming closer and closer to everybody to make it universally accessible. You can build great stuff. We're not there yet, but it's coming closer every day and people are learning more quickly. So they're approaching the technology every day. And the impact of that is more creativity, more discovery, faster iteration. and that impact is a huge boom to society within the right guardrails.

11:47And to your point, there is this like knowledge gap right now. It's almost like this arbitrage over knowledge of where the technology actually is. And I think two key points. One is even if AI LLMs did not progress one iota from today, there's still so much untapped potential. And I think that's a big thing. And then B, y 'all go check out for the listeners, the episode I did with Michael Lue from Stripe. The key phrase he said that sticks with me almost every day is like, this is as bad as AI and LLMs are ever going to be. Right. It's only going to keep getting better. You put those two things together, it's insane.

12:24And I love how you define the agentic side. It's about, it's very much a systems thinking you're building the systems to build the systems and whatnot. And we're a similar type of firm, HatchRex AI. So it's the way I like to describe it in simple terms is we build AI native solutions and we use AI to build them. Yeah. On that side, it's like you do those two things, but the key thing is you're using AI to build in essence. The other cool thing too, we mentioned the Spark framework earlier. Reuven open sources everything he does. And we'll put that in the show notes if anybody wants to go deep into that.

12:55Like it's almost absurd how much stuff he open sources and just puts out there. but that's like his foundational view and mindset is things should be open this shouldn't be gatekept but like back to the original kind of closed versus open discussion this is like a very interesting debate to me because there's i see points on both sides that make sense right there's the point on one side around just the transparency and trust and you know the safety side of it right Cause some will say, oh, it needs to be closed cause this is so powerful and it can't get into the hands of nefarious actors and people with ill intent.

13:35And on the other side, it should be completely democratized and open and that will most benefit society versus a small number of folks at the corporate level dictating how this grows. In like both points, I'm like, yes, makes sense. What's your take there? Especially like in the wrong hands. I think that's the scariest point for people is like when it gets in the wrong hands. And can you stop that anyways, I think is the point. But I'm just curious, like going into some of those points around the debate. Yeah, it's entirely all those concerns are completely legitimate. So they are things that we need to constructively and creatively solve for.

14:14I think most people would agree it would be uncomfortable to have a singular entity or organization that had the sole access to near-infinite, cheap, and abundant intelligence. And that was not, to some degree, ubiquitously understood and accessible. accessible. And that's the heart of it for the Agenix Foundation and Agenix Starwood, is that these tools are so important that the awareness of how to use them, what they can do, how to build them, how to develop them responsibly, deploy them responsibly, responsibly, all of that needs to be open. Now, transparency, accountability, traceability, and how we identify and respond to the potential of bad actors, that's a tool set and a number of tool sets that a number of organizations are working on.

15:10And also, let's be practical, let's be realistic. Governments and leading institutions are going to retain the most powerful models that they've been able to build for a period of time before they're public. So there will always be a leading edge of governance. And obviously as a civilization, you want to be able to have trust and faith in the transparency and accountability of your governance, but also ensure that the ability to protect all of us collectively. Yeah, there's a whole bunch of things there. The base level, like you said before, of what we've already got is so incredibly powerful.

15:48And we've only scratched the surface in terms of utilization and where that can be applied that we've got lots to work with already. Yeah. My view is I like to believe that there's more good actors out in the world than bad actors. When something's truly open source, that there's almost this, these two forces combating each other. And ultimately, I think the good will win. At least I like to believe that. But there is power in the decentralization of this like insanely powerful tool and innovation, right? Yes. Yeah, absolutely. And the collective input of human creativity, because we have such great context of the entirety of our lives and our worldviews, uniquely still leading in terms of side by side of being able to direct our creativity anyways.

16:36That's why I love the Friday Live events, because you get to hear from so many. And what Riven's been doing is he's been showing what's possible in the hands of an amazing engineer. Right? Taking everything to its logical conclusion. And so via Genics.org, we want to make sure that those tools, the understanding, the repos the members are working on, all of that is available. And it's really exciting. We've got, I think there's about 80 chapter requests around the world. And one of the first in Asia will be in Singapore. So there's lots of really neat work there. Because at the ground level, people want to know in their communities who else is involved, who else is engaged, what are the leading tool sets.

17:20And so those local chapters will be a great asset to the community. So I think the other point with open source first, closed source is how do you define it? Because you hear like companies and model makers coming out saying they're open source, but they've just essentially given the weights to their model. What's your view on that? Do you view that as fully open source or is there a spectrum of open versus closed source as well? Yeah, for sure. Yeah, the weights of the models, that's an open source model. Otherwise, yeah, it's just a variation on the theme if you're not exposing those weights.

17:54The ability now for lighter open source models to train up on leading closed source models means that the economics of that intelligence gap really significantly change, right? Since the Quinn, which led to the DeepSeq, which... Which Quinn was Alibaba's model, and then was it DeepSeq that leveraged that? So you're almost saying they're building on the backs of these other models in a sense during, for the, as a, an approach for their training of the new model in a sense. Yeah, that's right. That's right. Cause they can just tap right in, directly interact with the leading thinking models and then benefit from that logic exchange.

18:40I think the repo that Ru was talking about was DPRO, which allows you to, and it's not expensive either. It's not, it's not what it was six months ago. We talked a little bit about Reuben's Spark framework that he leverages, but like on the AI IDE coding front, curious your take on the different tools, which ones you like, which ones you use for different purposes. And then this may be old news by the time this launches, but I know you saw a couple of days ago when AI is talking about acquiring Windsurf. And what does that start to do? But maybe start on just the different AI IDEs that are out there.

19:20What's your go-to? Are there ones you use in different scenarios? I've been using Klein since it first came out, and that has been my preference from the very beginning. And now Rue, Rue Code. Which is built on it, right? Yeah, which is a fork off of Klein. That's right. That community, they're doing just great work. And so, yeah. So Windsor, Cursor, really for public-facing, small projects, both. Levelable, you can get pretty far now, especially that you can integrate Superbase and work with that. But as far as in the IDE, for me, it is completely through code with custom modes. Boomerang, being able to create subtasks and custom modes and the creativity of what we can do with those custom modes is just a, it's a wonderful combination.

20:12And that is, you do that well with the Spark framework, which means all the lifting comes in the planning and the articulation of what it is that you want the software to do. And user stories and functional specs and everything else builds up from there. That's where the work is. That with root codes, custom modes and the Spark framework, that gets us there. So I want you to go deeper on those two things, these custom modes in the boomerang, but just for people as well, because I feel like a lot of people either aren't using these tools or they're new to them, but like root code, for instance, you can be in just a regular old VS code instance and just, it's almost, it's like a plugin, right?

20:52So you kind of have the chat window there. So that's the way you can access that. Technically you can do that in cursor as well, but may do need to, I guess is the question, but go deeper on those two things, custom modes, boomerang, what are those? How should you think about them? for folks listening. Okay, great. Yeah, so root code, VSC extension, open source. You can set and select the models that you're going to work with, that you want it to work with. And it lives on the left side of your IDE. And you interact with it. And it can read, write, edit, do all kinds of, create the code for you.

21:31Or edit a review, or you can keep it in ask mode and you just have conversations about your code. That's an underutilized thing, by the way, is like collaborating, aligning with your, whatever tool you're using just to plan together. Right. Yep. That's it. That's it. Because in, in that dialogue, all kinds of really neat insights and creative inspirations come. So key is to have that fun and do that exploration before you really get into say, okay, now this is what we're going to build. And so you've got that mapped out. So with root codes, now you can think about it for a second. I want to turn off the ability to edit, leave the ability to read, and we'll call that ask mode.

22:12So then I said, great, go through my project documentation. Let's talk about how we're going to, what kind of ideas we have for the UI, right? And have a conversation back and forth. Maybe what are some interesting visualizations we could pull from the data? That's ask mode. architect mode, let's say, you can give it a specific set of custom instructions whenever it's in architect mode. You describe how it is that you want that documentation and those plans and the specs to be created. How do you want them written? What's the form? What's the flavor of it? Your general approach to writing those kinds of documents.

22:52So different modes can have different custom instructions so that when you enter that mode, it's got a, it's wearing a specific hat. And Rue code enables us to create custom modes. So we can create any mode that we want in a template in the remotes and, and now we can move between them. Okay. There's one really important mode, just like in any multi-agent system, interestingly enough, because this sort of transitions root code from a coding assistant into more like a multi-agent framework, specifically for coding. And that most important mode is the orchestrator mode. So that's the mode that it's going to be in when it's thinking about the project or the part of the project that it is working on.

23:39And it's the one that has to, is going to determine what the next task is and which mode it's going to assign it to. Fascinating. That's the boomerang. It's almost like having another human in the loop in like a weird way. I get more into this like mode of as I'm working with AI, I intentionally pretend like I'm working with another human in a sense. Yep. Yep. It is fascinating where you can go when you get down the rabbit hole of coding, when you can, we have a whole team of specialists in the development environment, right? One of the first things I did when I saw the room modes, I said, great, let me create a custom researcher mode.

24:29And I just laid out, used the OpenAI API connection and GPT-4O search preview model. and anytime you need to look up API documents, just use that. And I've got my key and my M and it works. So now when the coding team is building a project and it arises to a problem with an API setup, it can just look it up and continue on. And then Chris Royce morphed that on Revisuggest and morphed it into a perplexity MCP server. So now you can do deep research while coding. Phenomenal. Earlier, you mentioned lovable, which we glazed over that. But I got to say, that's almost my favorite go-to tool for just call it vibe code here, whatever.

25:12And the beauty of it is you mentioned the connection with Supabase. So I've got into this nice little method that I work, especially for like niche, like use cases and things where I'll almost have it define, you align with it on what you want to build, but I'll have it like build out the landing page of the thing first. It's something about that. Does this nice like alignment on the positioning and what we're trying to build the value prop and then connect into Supabase, have it do authentication, set up the backend, and then you just start building feature by feature. But it's pretty insane of how capable it is when you have it integrated with Supabase.

25:49And then the design is just very impressive in terms of what it can do from a design perspective. Haven't they done a wonderful job on there? It's really nice that it puts out in the sandbox environment so that as you're asking, for example, At the end of the day, I'm a sculptor. I wholesale... Did you sculpt that behind you? The thing behind you? Yes. Who is that? That is no one in particular. That's just a clay study. I didn't know if it was like Aristotle. I was listening. There's a beautiful clay sculpture. What do you call it? A bust behind Robert Hickman? Yeah, that's right. Thank you, Matt.

26:23Yeah, I was actually... I thought I was retired from the business world until Jenny... Now you're back in. AI brought you back in. I was completely immersed in the studio and the studio work. And when Gen AI first was publicly available, 3.5 came out, it was because the lifetime before the Clay studio, I was an operations exec. The impact on workflows, operations, organizations was just so obvious. It hit me like a two by four, and I've been immersed ever since. But I do still keep the Clay work for my Zen counterbalance. Nice. There you go. There you go. That's, you do the vibe sculpting. It's the next thing.

27:05That's the thing that's caught on. It's like there's vibe everything, which is funny how that's become like a moniker. Lovable is Anton and the crew. They, I think they've done it just, obviously they've done a phenomenal job and they've, and the work that they do under the hood is, it's just a wonderful user experience. So you can chat in natural language. You've got dev mode. I don't know that's actually trademarkable or copyrightable, but. Yeah, they got into issues with Figma. The response back to Figma was funny. It's some back and forth. Oh, right. I didn't see the response yet. I'm sure it would.

27:38Yeah, that'll be good. But Lovable itself is a wonderful program. And what this means is, yeah, demos, value props. You can create a little app just for a meeting. It makes sense now. And for non-developers, there is this side of it that's like you and Reuven and folks like that. They're like deep into it. But anybody can now build, have that experience. And I feel like it's such a good inroad to then say, oh, how does this work? How does that work? Then you leverage AI to teach you. And there's nothing stopping you from progressing to wear a Reuven using AI as your teacher. That is the uncomfortable truth.