Why Solo Builders Should Sell to Enterprises, Not Just Consumers | Joseph Enochs, Chief AI Officer

I think as a solo founder, I wouldn't go after enterprise to start.

Why not?

How? Tell me how.

My kids are creatives. So I said, well, hey kids, what do you think about AI and video generation? And they actually did not want me to do it.

You still went ahead and did it.

I really had to do a lot of soul searching about the place for creative people in the world of AI. Do I always wanna try to build a bridge versus try to build a wall? My son was like, if you could create a digital paintbrush that would allow me to have the agency of my creativity while not taking that away. Those are things that would build a bridge between the creative community and the technical communities.

Are you seeing certain roles disappearing because of AI?

Microsoft, in their customer service, they tried to use AI for customer service. Well, they ended up saving $500 million. It wasn't about how many people they could let go. It was about how much they could empower the people that they have.

What do you think is left with them that's not replaceable by AI? Does something like that even exist? Hello everyone. Today I have Joseph Innes. He just published the O'Reilly book, Video Generation with AI. Oh great, you have it. I can tell you that I skimmed through it in the past few days and I have to say it's really comprehensive. Basically everything related to video generation is there and you can pretty much start from scratch if you want to do anything related to video generation. Welcome, Joseph.

Hey. Thanks, Angelina. Thank you for having me. I'm so excited to be here. And thank you for the kind words about the book. It was a long time in coming. And I have really tried to make it accessible. The team at O'Reilly and myself have really tried to do everything we could to make this more of like an entry level for folks who really want to understand the underpinnings of what happens with generative video.

Yeah, it's a very interesting topic. But before we dive into more about your book, I'm curious, is there any debate in your corner of AI that you observe is heating up right now? Any trends you're seeing

Oh my gosh. So many crazy things. And with recently with everything that has happened with Hugging Face and OpenAI, and today it was announced that NVIDIA actually bought Hugging Face, which was amazing. So I think a lot of folks are excited about the capabilities of AI and AI agents. And a lot of people are concerned about the automating— automation of defense against AI. that could— frontier models that could potentially do harm.

What's your view on this? Are we trending towards the right direction, or it's concerning to you?

It's concerning. I think that there are valid concerns out there. I think we know with this last use case that was talked about at Black Hat by OpenAI, we know that the sort of the threat actors, the people who are offensively trying to go after and exploit vulnerabilities, have an automated loop to do so. However, in large organizations and in businesses, the loop to defend the environment, the loop to test those threats and identify and validate those threats, that loop goes through a standard sort of change process, which I think the mean time to remediate some of these things is around 43 days, whereas the exploits can happen within a few days. There's some, some challenges that are out there. There are actually some government mandates that just recently came about that are talking about how organizations do have to do something better with their patching and vulnerability strategy, but the regulations, I think, don't fit any capable manual process. We're going to have to figure out how to use AI to support and defend our systems against these automated adversaries.

Yeah, that is definitely alarming at the moment. On the one hand, I'm talking with a lot of founders who are automating a lot of things like self-driving money, AI agents taking actions on our behalf. There's tremendous amount of value in that.

I think there, there is a tremendous amount of value. I think the, if we can create boundaries Bounded autonomy, meaning autonomy that can ride within our own guardrails and actions that can be, let's say, undone if they're done the wrong way, right? Being able to identify when things are after they're like, we validate, was it supposed to do that? It didn't be able to remediate those things. So I think there's a lot of excitement and capability of the systems. We think of these agents out there like miniature robots. And we've got to prepare for the time when they are in the physical world. Right now they're obviously in sort of the hyperspace, but— At some point, they will be physically walking around with us. There's some robots out there, but like real robots that you can delegate to do tasks and work and will actually fold your laundry and do things like that.

Yeah, those are nice. But even just within the digital world, it's already taking actions on our behalf, which can cause a lot of damage. As a consumer, do you have any suggestion like how we can protect ourselves?

I think right now, just be prepared with what you— whatever autonomy that you give it within some sort of— we call them bounded loops. Like these things, I'm okay if the AI does these things. And again, if you can find a way to make it so that it identifies and validates those things that it does and remediates them, that would be the key. And then for other things, I think you want to have those, maybe those things done in batches, right? Where there's small batches of work that the AI agents do for you. And then you can validate that those small batches of work that are done for you. And then things that are critical for you, some of those things will need to be human in the loop, probably from now on, right? It'll do things and you'll need to validate.

I think you're still speaking for the developers or the more of the builder technical audience. For regular consumers, they wouldn't know any of these or anything like that, right? I think maybe the caution is what we can do is to be careful what you're delegating when you're chatting with AI, and maybe the products out there that's claiming this is autonomous driving for some parts of your pipeline or your life, be careful with that.

I think that's good advice. I think anything that you think would be risky, that would not be easily able to be validated and undone, right? If it would, could fix it in 2 minutes or 5 minutes, that's probably something that you could be okay with an AI doing for you. If you do something that takes you a day or so to remediate, that's probably something you want to reconsider. And if it's something that's going to be obviously detrimental, those are things that you don't want to delegate to the AI, right? And just like normal people, right? So I think it's more about, in that sense, the things that if it does go wrong, how quickly would it be for you to be able to go and fix it?

And for developers and builders, we're so excited to be able to build so many things on the one hand, right? I'm curious, did you talk about setting up guardrails in your book about video generation?

There's no— the book doesn't go specifically into guardrails. The sort of the hero's journey that it goes through is about— it starts out with really people excited about video generation. So right out of the gate, people with very little, like basically free, if you have access to the internet, you can click on like a Google Colab, which is some service that's free from Google. You can go in there and you can prompt and have it generate some videos for you. Now, these are the smaller models, so they're completely free and the outputs are not gonna be the highest and highest of quality, but it's gonna get folks understanding how those things work. And then in the, that first portion, we start talking about the prompts and how they impact and some of the settings that impact output and duration of videos. And then the second chapter kind of steps back in that hero's journey to really talk about ethics and talk about sourcing of data and talk about sort of the provenance of all of the information that you should get and derive for models that you're training in. And then we deconstruct all of the various pieces from the diffusion transformers, which is like one of the biggest underpinnings of video generation. We go through that whole cycle. You basically build those things from scratch. And then when you get into the later chapters, it's talking about making AIC, so vision language models, where these, the AI can actually interpret what is seen. And then chapter 7, we go into audio and video. And then chapter 8 is about building sort of safe products around the model. It's not just about the model. It's about if you wanted to develop a product. So in this particular section of the book, we're really talking about those particular things.

What kind of safety did you talk about in the last chapter?

The most thing, it's more, again, this is, this part is for like folks with developers, but really the models that you choose, you'll want to choose models that have been like proven not to have been made with copyrighted information. One of the organizations that we worked with and the CEO of the company, actually, it's an open source project called LTX, and they have really strict standards on the provenance of where they get their data. So what we really want is folks to use models that have provenance on the data that they use to generate the video so that we're not taking IP from other people. And so in this scenario, it's really choose the models that have been validated, especially if you want to make this into a product in the future or make this into something that you want to have copyrighted.

So is the trigger of copyright choosing which model to use or choosing what data to use for video generation? 'Cause I imagine I can give you some seed images to generate something.

Yeah, so you really, it's like a little bit of a combination of both, right? So when you choose a model to generate, you want a model that has not been built from utilizing copyrighted materials. And then the models, the images that you're using to generate as a keyframe, you want those to be your own content as well, or to have permission from the artist who created that piece of art to generate the images or videos from that particular keyframe. So you want to pay close attention to it. If you're just having fun and things of that nature, probably less of a concern. However, if you're trying to do this professionally, which many creative people are, right? Those things are, you want to make sure that you don't get into any questions about the chain of rights associated with the IP that you're creating.

Which models are IP safe right now?

I think the biggest one for open source, I would say in my mind, is we work with is LTX. I think they're the primary one that A lot of people have validated the training materials are more— are safer for folks to be able to generate videos from.

So I guess this is a great segue for you to tell us a little bit about what is your day job and how did you end up writing a book on video AI?

Yeah, so I'm the Chief AI Officer for an organization called Enterprise Vision Technologies, and we work with large enterprises from all industries, from healthcare, media and entertainment, large retailers to banks. And so for me, I— my background is in applied machine learning, so I've been basically doing data science for quite a while. During COVID I worked with the county out here in Los Angeles to do the COVID tracking. So all of the things that are associated with outbreaks and things of that nature. So I've had a legacy in machine learning. In early 2023, I actually wrote a book about large language models. And then when I wrote that book, a lot of books about large language models came out. And for me, I was like, you know what would be the next sort of wave of technologies that people would want to learn about? And the interesting thing is that my kids are creatives. And I had done a lot with image generation, things of that nature, and just for the utility company, drones and cameras and things to protect against wildfires, right? So a lot with images and things of that nature. So I said, well, hey kids, what do you think about AI and video generation? And they actually did not want me to do

it. You still went ahead and did it.

Because I really had to do a lot of soul searching about the place for creative people in the world of AI. And it's important to me. We have our most important people in our lives are our friends, family, and loved ones, right? And for them, it was important to me to really learn from them and build the bridge. I always want to try to build a bridge versus try to build a wall. And it was to build a bridge to really understand where they were coming from and where I was coming from in my work and try to build that bridge. And one of the things I found along that journey is that one of the big things about creative people is that they did not want to lose agency. They did not want the AI to take that agency away from them, that creative freedom away from them. And they wanted to be able to create pixel-perfect imagery or videos or things of that nature, just like they did with the traditional tools, the traditional, let's say, CGI-type tools. And they likened it to like a paintbrush. My son was like, if you could create a digital paintbrush that would allow me to have the agency of my creativity while not taking that away, those are things that would build a bridge between the creative community and the technical community. So I set out on that journey to write the book and I originally called it Making Machines Dream because it was like the concept that you were putting in your dreams into these machines and the output was actually them dreaming. So that's where I first started with it. And then O'Reilly told me, I think there's a technical book in there somewhere. So we broke it out into the components to teach people the underpinnings of the video generation, vision language models, and building products around those, those vision language models and video generation models.

So you were already building the bridges before you actually started writing for O'Reilly on this book.

I did. We had did some work again with the utility, the electric utility, a lot with vision, helicopters and drones footage with OpenCV and other projects and doing blur detection and things of that nature to try to help prevent the next wildfire. So that was early days. And obviously computer vision has been used in radiology and things of that nature in medicine. And we believe that this next phase, and it's actually already been happening, when I was writing the book, I believe that the next phase would be multimodal models, right? It's not just going to be language and grammar. It's going to be a tokenization of the physical world. And vision was really the first piece of that. That's one of the reasons why, if you look at the bottom of the book, it says Diffusion Transformers and Multimodal Learning, because it's really the first phase of moving outside of grammar and into vision and into audio. But we will be tokenizing many other things, right? We'll be tokenizing, and we already are, tokenizing proteins and tokenizing chemistry and doing things for clinical trials, toxicity, manufacturability of drugs. Those are all things that are coming. And I'm happy to announce that O'Reilly has accepted my second book on physical AI.

What are you going to write about?

There's a growing number of books about robots, right? And robotics. But there's a gap about the whole components encircling all those things, like creating a digital twin of a physical environment in a virtual environment. So like open world models and visual twins. So it'd be a lot about those things and about bringing the technology into the physical world. I'm going to save some of it, but it's all about all those components that will make AI move out of just the cyberspace into the physical world.

So that version of the digital twin has a physical component to it, meaning robotics.

Just imagine, could be anything. It wouldn't have to just be robotics. It could be clinical trial, right? In a clinical trial, they're trying to find compounds that can address an ailment Like a cancer, right? And so they're looking for some compounds that can address that. One of the first things that they have to do in a clinical trial is figure out: Is it toxic and is it manufacturable, right? So chemistry and toxicity and all of those components, those things have to be trained into a model. So those are physical components that need to be tokenized. And what that will allow scientists to do is if they're going after some sort of drug to cure cancer, then they'll be able to have. Have these compounds, and then they'll be able to say, are these compounds non-toxic or toxic? And the models will be able to say, hey, this group, this subset has a high likelihood of being non-toxic. And then the next component that they have to do to say, can these components be manufactured? Because even if it's non-toxic, if you can't manufacture it with known manufacturing processes and chemistry and things of that nature, then it's not going to be able to make it capable for you to manufacture it. So those are kind of like the first component of a clinical trial. And then at some point, they'll move into different phases of clinical trials and think about creating a digital twin of a living organism. I'm not saying that we're there yet, but they call that the wet lab. And that in that scenario, you could have a digital twin of a mouse model and you could be able to virtualize that thing, create a digital twin of those scenarios so that now you could potentially, is it non-toxic? Is it manufacturable? And is it potentially have efficacy against this particular ailment? So over time, we will continue to create these digital twins, right? So that we can essentially simulate things much faster. And if we could cut the cycle just in the pharmaceuticals, if we cut that in half, man, that's a significant amount of good that we can do for the world. And that's just one example of how bringing AI into the physical world can help people.

So you are making digital twins of, let's say, the mice and then simulating a wet lab reaction to a potential chemical compound and whether that can have an effective curement of some type of ailment.

That's definitely one of the long-term objectives, right? But there are many micro objectives to getting to that point. But the The capability, the theory is there. We're already tokenizing protein folding. You've heard of these open-source models for protein folding, models that for, again, for radiology, some of the best models that are out there for radiology are AI models. But if you just extrapolate that out into the future, that's the reality of things that are coming to help people.

That's crazy. If you can create a digital version of my health data or anybody, any consumer's health data, and We can do a digital trial of whether a medicine can work on me or something like that without me actually trying it out.

As long as that can be done in a safe and privacy-preserving way, it could help a tremendous amount of people on the planet. And democratizing that is, I think, is a fantastic goal. I know you see people wanting to fly to Mars and populate Mars, which is a great goal too. Not to— no judgment. This is judgment-free zone. But I think there's a lot of good that can be done here, environments. And those are good causes that I think we owe it to our communities to try to go after. This book on video generation, I know it might seem odd and a stretch, but that's the long-term vision in my mind, is that video was the easiest on-ramp for that multimodality, right? And audio, right? But next you're moving into these other areas like chemistry and other things.

Speaking of video generation, let's go back to your book. What's the biggest challenge today versus last year in video generation?

I think right now it's getting super exciting. The technology is getting very good. I think we still have a lot of problems with consistency. It has gotten a lot better. Before, you could create models that were a certain length. So now this year you can create models that are much longer, right? But there's still issues with consistency. There was a famous commercial. I don't want to talk too much about it, but it was with a soft drink company and they created a commercial and there was inconsistencies from frame sequences to frame sequences and everybody looked at it, right? of people looked at it and said, man, dude, could you— you guys are a large company. You obviously have a budget to do marketing. You're saying that you couldn't make sure that this was at least closer to pixel perfect? You use these low-grade models to do so? But I think the biggest thing is still consistency. And from last year to this year, the duration of videos that you can create is much better and the quality is much better too. But consistency is still something that we are improving.

Can you explain how this consistency problem is being resolved, the design thinking behind the model. I think the idea is generating one image is okay, and then a video is hundreds of thousands of frames of those images. And then what we want to achieve is consistency of the same person, same subject, or the background, right? So that's the detail that the creative people can tell versus me as a regular consumer. I may not have the eyes, but it's probably a sore eyesore for those creative people.

Pixel-perfect people creators are to do that, right? So it's very important, the lighting And things of that nature. It's very important to them, and I and I'm more from the technical side of it trying to the creative side their dreams come true if you will. And it really is important to them that that look that illusion that they're trying to create is super important. And that scene box like the scene sixty-degree scene box, the plants, the lighting, the person, the all of those components are things that we are trying to be able to tokenize and train against. So that when we basically want to generate something or reimagine something that's already been created in another form, that it's consistent throughout those frame sequences. It is still a big challenge and we're overcoming it with fine-tuning and the training loops and agents and things of that nature. It's getting a lot easier to do these things. The hardware is becoming a lot more capable so that you can do a lot of these things without paying a significant amount of expense. And we'll get there. We will definitely get there.

You said something called scene box, right? Is it one frame of the image in a video?

When I say a scene box, I mean, imagine that you're in, um, an episode of Friends. They're recording that on set and everything that you see in the camera is 2D, right? But there's really a 3D scene box around that scene. And so everything that's inside of that scene box is something that creative people want to maintain consistency on frame by frame. And so with video generation. That's one challenge to do, a challenge that we're trying to overcome. And Gaussian splatting and open-world models, these things are reimagining these things in 3D, which will make it much easier for us to then compress them down to a 2D image you're seeing here, or video like us talking with each other.

This is so interesting. So you're saying that from image to video, it's from 2D into a temporal 2D. However, you're reconstructing the scene to 3D Because in reality, we're imagining that 3D scene behind the whole idea. And with the reconstruction, you're tokenizing all the components as well. So everything can be trained. It's not just red, blue, green pixels.

No, it's like your— the shape of your glasses, right? That color. And yes, exactly. The perfect shape. And when you turn your head and you look up and down, that has to be maintained. And all of the elements in your background, right? From scene to scene, every one of those components is super important for consistency. The lighting, right? It's a lot more than I think people recognize. The beauty of it is too, is that like you said, from a 2D image or multiple 2D images, now you can create a 3D scene bot and then you can tokenize those objects into 3D. Animators are actually doing that a lot now where they can take video and then take that video again and turn these into 3D objects. And then those 3D objects can be used for animation.

How are the creative people receiving something like this, this trend of the duration's longer, generation is more consistent, it looks better?

I think there's still a lot of angst from the creative community about the future direction of what the technology's gonna be used for. I do feel like a lot more are utilizing the technology in their workflows. So things that have been traditionally very cumbersome, a lot of different tools that are used, like when you're making videos, there might be like a movie, there might be 15 different tools that are used, right? Probably even in your workflow, there's probably a lot of different tools that you have in your workflow. So I think a lot of creative people are recognizing that if they can utilize these tools to reduce friction in their creativity, to help them stay in that creative zone, right? Keeping that sort of that state of creativity going, right? Keeping them in that zone. I think those are people who are excited about those things. Again, there's still a large group that has a lot of angst about the technology.

Among the creators, what do you think is left with them that's not replaceable by AI? Does something like that even exist?

Everything. There's no fear. There should be no fear. We need creative people. I have a tremendous amount of respect for— when I was a kid, I was painting, I was doing all types of art things. But where I grew up, I joined the Marine Corps and went into the military. That was something that where I lived at, and my father was in the military, my grandfather was in the military. So it was just something that my family kind of wanted me to do. And for me, it drifted away as a, not as a passion, but as something that was reality just because of the logistics. But I recognize that today in this world, that creativity and creating something and having an idea and having a dream and having this technology, this is the time right now that there's no better time where your ideas and dreams can come true and the technology can assist you to getting there. So I think we always need creative people. We're going to need creative people for creative solutions and ideas. I think should have no fear.

So creator who's anxious about AI replacing their jobs where they couldn't find paid gigs anymore, what'd you say to them to convince them that, you know, embrace the technology?

Yeah, I think I would just say there's digging your heels in against it, probably the worst thing that you can do. Finding ways that it can assist you And it's interesting because in— I may actually release the original manuscript for Making Machines Dream. But the way that I took that, and I, again, I have my son and I have a couple daughters, right? And older and younger. And I, when I researched sort of the history of Disney and all of the histories, because I went— a lot of research went into the book when I wrote it. I found out everybody's heard of Pixar. And one of the founders of Pixar was a gentleman by the name of John Lasseter. And John Lasseter's first job at Disney was he was on the Jungle Cruise ride. He was a captain. at the Jungle Cruise ride. So he was literally working at the theme park as a young creative person, and he ended up helping create Pixar. And for a time of his career, he ended up running Disney Pictures and Animation. And then if you look at Ed Catmull, it's probably a name a lot of people don't— haven't heard of, but Ed Catmull was— his first job at Disney was selling popcorn and balloons. And it was an interesting thing because when balloons came out, so he was selling popcorn and it was great, but when the balloons came out, There were 50 balloons and you had to count the balloons and tie the balloons and blow the balloons up and pull a balloon out and give it to people. It was too overwhelming for him. And the interesting thing about Catmull was like Disneyland and Disney World and Epcot Center, but several of the theme parks around the world, theme parks, Ed Catmull built those. So he couldn't sell popcorn and balloons. That was too out of his zone, but he could build the park from scratch.

How did that happen? Like from selling popcorns and balloons, from balloons to become this, the creator and designer of Disney and Pixar.

It's a fantastic story. There's a biography about him that you guys should read, but talks about like when he went to see The Wizard of Oz, people were mesmerized by the story, but he was mesmerized by the mechanics of what was actually going on, right? What did they do to actually create that? So he did theater and he also did construction. And so the combination of those things, and he tried many times to actually get into Disney, and at some point he did, right? And he was able to work his way up to running all of Disney Parks and Experiences. And so the book, the objective was to take the journey of these young people who were creative, the popcorn and balloon salespeople and the captains of the rides, and really tell them, like, to give them some excitement that this is possible. Don't be so concerned. Follow your passion. Follow your dreams. Big things happen from, like, small first steps. So it was really talking to young people and then people who were in the, the middle part of their career, which is very challenging. You've been doing something for a decade or even longer, right? And now things are changing. But the advice in the book was, is really about, there are things that you can potentially do now that would've been out of your reach. Maybe you were doing certain components of a film. Maybe one day you and the other folks in the studio could actually do a screening your own idea, obviously based on the vision of the director, but you could potentially be able to create on screening day your own interpretation with this technology. So there are things that you can do with it that would— doors that could be unlocked for you that you might not have thought were possible with, but this technology can help you enable it. And then the final section was about with executives who were once again selling popcorn and balloons that are now executives to have empathy, 'cause these are challenging times for people, for young creative people, for people who have been in their careers for a while, to really have empathy about that scenario and really understand where you were once in this position as well. Enable this pathway, clear pathway for creatives who are going through some challenges with the advancements in technology.

You should really publish the first book too of your O'Reilly book. O'Reilly doesn't support us. Maybe we should call Tim.

Well, Tim, if you're listening, I'm sure there's a way. I would, I think there's a, put a lot of thought into it because like I said, that your, our families are our most important, our friends and our families and people that our community are the most important people around us, right? And we should take care of them and shepherd that. That's an important social contract that I believe in.

If I were an AI engineer and I quit my job yesterday and I'm thinking about making a video generation product, what's the minimum stack that I should use or should

I think NVIDIA and AMD both have good technology right now. So they both have these small boxes. Here, maybe I'll show you one of them. So this is a little AMD supercomputer. A box like this, it has, it's a little pricey, right? But a little box like this has 128 gigs of shared memory in it. This box can run pretty much any video generation model. So with something as small as this, you can run a lot of this sitting on your desktop. I have a bunch of these and I have ones from NVIDIA as well. I have over a terabyte of VRAM here in my laboratory. Something as small as that stack and you can create some amazing things. You could take that, you can read through this book and you can figure out all the internals of these video generation models and you could design and build a product. In fact, chapter 8 goes through ComfyUI, it goes through LTX Desktop and it goes through Invoke AI, which actually was acquired by Adobe, and it breaks down the primitive components and the important components and pieces on how to build an AI-based product on video. So you can do it for very little.

And let's say if I listen to your advice and start building this, what use case should I build for? What's your suggestion?

I always say follow your passion first. That's what I tell my children, right? It's like, find your passion, follow your passion. I've always been supportive of that because things are going to be difficult no matter what you do, right? You're always going to have challenges, but But when you're passionate about something, that passion is gonna see you through challenging times, right? When the times get hard, it's not gonna be stress because you're following your dream, following your passion. So the first step is to follow your passion, right? Which I think is common advice. But the second thing is to find your community. And that's whether that can be in a video game or that could be in sports, or that could be in some craft or something. Find your community, right? And then once you find your community, find a need in that community and try to help and support and fill that need in the community. And there's a lot of examples of like the folks who created Cal AI, like the app to track your calories. So I think that's, you don't, now you don't need an app that makes $1 million a month like Cal AI, but think about it. If you get 10,000, 20,000, 30,000 subscribers and followers supporting you on your journey and your community, that can support you, that can support you and your family. So I think that's something that's going to be more common because of what's happening in the job markets and things of that nature, I think there will be some ability for creative folks to be able to create their own way, right? And to follow their passion. So those would be a few components and advice that I would give.

Thank you. I'm building this channel towards builders who actually are in a position to build something that one day they would like to sell. So I'm curious, with the AMD hard drives you mentioned and your book, I can start building, but then the next question becomes, how do I sell? How do I find my community? Can you share some of your experience at EVT? You must experience the early days, right? You were there for how long? 13 years?

I've been there for a while. Yeah. There were 9 employees when I started, but now, and it was, I think we were around, let's call it $30 million and now, yeah, approaching a billion. Yeah. It's, it's come a long way. So I think the audience that we are talking towards are more like enterprise clients, right? From Fortune 100 down to Fortune 300, right? pretty much the largest companies. And that community, stability and safety and reliability are very important, right? And they are very risk-averse because of regulations. And so in that scenario, in that community, when you build something, you really want to make sure that it aligns with the standards of those organizations. So security standards, I don't want to throw out too many of them, but there's like the ISO 42001, which is like the AI standard, right? The NIST AI Risk Management Framework, Right? So you would want to align and understand that you're building something that is going to be consumed by these organizations and really try to understand how their governing bodies are going to look at bringing your product or service into their organization. And that starts with safety and trust in addition to capability. So while you're building things, it's very important to keep those things in mind.

I think as a solo founder, I wouldn't go after enterprise to

Why not?

How? Tell me how.

Enterprises have problems. Think about this problem we were just talking today. They have problems with this, like patch management. It's a major problem. And so in that scenario, I think there's a lot of companies. I think, what was it? Opus lasted 10 weeks before they were acquired. They created one of the first autonomous patching systems for, you know, remediating vulnerabilities, right? Which is, by the way, it's like a mundane thing. You click on your Windows thing and you say update, right? And it updates. But imagine doing that for 5,000 machines, right? So they created a series of agents that did patch management and they sold their company. I think it was 10 weeks they were able to get acquired. So it's really trying to understand the business challenges and the business problems and then creating solutions to address those challenges. And there are so many challenges that these organizations are facing, especially because of AI. Like these frontier models have enterprises concerned, very concerned. I sit on the board of the HDF Group, which is an open source project for large datasets, like for NASA data. And so as a member of that board, like as the governing body, these people are, I'm like the newest entrant to this. And these people are very skilled. My other peers on the board are very skilled about risk and governance and things of that nature. And so the top executives in these enterprises are concerned about these things. And these are brilliant people and they They are becoming and have to become very aware of and more sophisticated on what this technology can do to protect the organizations that they serve because they, they're part of the governing body. So there are problems out there that young founders can come and solve and they're, they get up to the board level. I know, 'cause I talk about it monthly and how we can protect the business. And it's a, an obligation for, for the board members to do. Don't shy away from it 'cause it's a big organization. They've got problems and they need good founders to come and help them solve these challenges.

How can a founder get to know these problems? As you know, some of the cases where you have to work for a company who's facing these problems, and even if you work for a company, you may not have the visibility to really understand these problems inside out to be able to think of a holistic solution or understanding in depth enough.

I think we all do have the tools that we need. If you look at, I always go back to, there's a great author by the name of Gene Kim. I don't know if Gene Kim, he wrote a book called The Phoenix Project. It's in technology books. It's one of the— it's a fantastic book and it's just got a long legacy in technology, but it's written such a way that it's a narrative way. It's like a rewrite of a book called The Goal by Eliyahu Goldratt. And Eliyahu Goldratt studied under a professor and then his— a guy named Russ Akoff. And then Russ Akoff studied under a guy named W. Edward Deming. But the point of saying that is Deming and Akoff And Eli Goldratt and Gene Kim and all these folks talk about like the value streaming, like figuring out the problem, the end state challenge, and deconstructing that step by step into those value-added components, right? And finding out where process— think about this as a relay race. And if you're passing off the baton and the baton falls every time, that's what happens in most of these enterprises. So finding those challenges, right? And timing it just right. There's a lot of timing that goes into it, right? Because that company Opus that got acquired, that situation, was there was a gap in the market and these organizations needed to have a company because next quarter, right? Because if they didn't have something on the books by 1 or 2 quarters out, then there was going to be a lot of pressure from the street. So in that scenario, there's a lot of timing, there's a lot of challenges that need to be resolved. But in that scenario, I think we all know problems and we all, we can deconstruct at least in the sphere that we have to be able find solutions to those problems. And so I think everybody has it, is just really trying to, again, be creative, think about how, what's the timing of these things, anticipate how these things are going to be needed, and then just really start building.

You know what, I was talking with another friend of mine and we were talking about young founders or whoever, domain experts don't understand different domain problems. And there's a gap there. So people need to, as I said, meet in the middle or somehow find out the problem before you devise a solution. And in the startup world, in most cases, people chat their way into this. So you just talk with enough people and then you feel your way through this value stream chain and being able to find ways to decompose the problem and find the problem that worth solving. So I think that makes sense. I have an idea. How about we start a problem bounty or something like that, a board that we just list all these problems and then have Like my audience to come and pick some of them and work on them. And you might have—

Let's do it. I'm in. I'm in. I'm in.

So why not? Because I feel the community, the developer community in AI, I think we should talk more. I'm hearing a lot of conversations. So after interviewing almost like 100 AI founders and researchers, by now you are maybe 90, number 90.

Oh, awesome.

Right. So I noticed that people do have offline talks here and there. I'm worried about my multi-agent system's reliability. What are you doing that this VP of AI is talking with another VP of AI? We just started adapting or adopting AI, right? Building agentic systems. What's your lesson and learn? And people are asking these questions. So I feel maybe we can make it more publicly available so that the conversation can happen more organically.

I think that'd be fantastic. I agree with you, Angelina. I think that's fantastic. I think it, I think people would get a tremendous amount of value and I think it would help people fine-tune sort of the next steps for them, right? And to your point, generate more conversations as we're all navigating these things together.

Awesome. A partnership formed. Yeah, I'm so happy. So setting that aside, I also want to ask you, for me, I grew up being told to think long-term. I wonder if you still agree that long-term thinking still matters today. As you said, there was a company you mentioned that's fixing the patch problem was acquired within 10 weeks, right? And a company like Menus, they're well known. I think they almost get acquired in about 3 years, right? Almost. I think it didn't happen, but it's almost the holy grail of being a startup founder in Silicon Valley, isn't it? That's what people are shooting for. People are not thinking of going public or like how your company has been doing that 15 years or 20 years. No, people are thinking, what if I can sell this in 2 years?

I think you have to be understanding that the technology is moving much faster and Now, I think long-term is important, but think of long-term relative to the speed of the foundation models. The speed of the frontier, if you're wanting to be a founder and create a business, the time gap from you having an idea and developing something, and then that having this moat, the time is being compressed and it's being more compressed every day, right? So today is like the most amount of freeway that you're going to have to find the solution and solve it, right? Because tomorrow the technology is going to get better. More ideas are going to be shared, right? So I think there, it is important for long-term thinking. In fact, like I mentioned, Dr. Russ Akoff, he is one of the well-known for systems theory, and that's exactly what he espouses is, is that systematic thinking, which means that you should take the problem that you're looking at and find out what it's a part of, like the larger community of problems that it's a part of. And then you should find the larger community outside and outside. And once you have a full scope, end to end of what that solution space actually is, then and only then should you drill down and become analytical and really try to solve and pinpoint into your one specific problem. So I think that is super important. But again, the temporal component of it, right? Time is all deals, they say.

What do you think AI has changed about your team structure and your organization? Is your organization using AI?

I think pretty much everybody I know, everybody in every pretty much is using AI to some degree, even if they're using it at home, right? For recipes or for asking about something, taking a picture of something and asking, what is this? Or asking for ideas for vacation ideas. What should I do in this location? I, everybody, I don't know that I know anybody that's not using it right now.

How are you using AI?

I use it constantly. I'm, I use it all the time. I use obviously Google and Gemini, right? So like Google searching, it's interesting. Used to, you just did a Google search, but now you have the chat screen right there based on your Google search. So the results come up and you're like, hey, I'm looking for this. The results come up and now it's a more of an interactive experience, right? And so something as simple as that, right? Looking for things, finding different recipes, restaurants, finding good things to do, fun places to do, creative ways whenever there's a problem, right? Simple things like, what should I do about this? Man, AI is just, it's, I think it's very empowering and very, can help people. And And just so many situations.

Are you seeing certain roles disappearing because of AI, like within your org, your type of organization, like a company consulting tech?

I don't think that's— I don't, not for a while. I think that, how do I say this? I have to really calibrate my speech here for a moment. Maybe I can just give a public use case. Microsoft. Oh, and Meta. I just thought of 2. Microsoft in their customer service, they tried to, you know, use AI for customer They ended up saving $500 million and the billions, billions that they spent to do so made them recognize that it wasn't about how many people they could let go. It was about how much they could empower the people that they have. Meta ran into the same thing with the community notes. They tried to AI automate community notes. And what did they find out? They found out that they really needed people. So it's really about empowering the people that you have, right? And using the technology to outmaneuver the competition, finding the new ideas, right? Being able to create versus trying to do so much more with less. 'Cause it just, actual evidence of use cases that have been, I think, on the large scale have shown that you cannot change the hiring curve so much. You should really think about how to empower your people and, and make your people work more collectively together as a team and to be more effective in the market.

What about the roles definitions? Are these roles, are roles being redefined? Like frontend, used to see like frontend developer, backend, PMs. Are they redefined in your organization?

I think in enterprise, when you talk about, let's say roles, let's, let's talk about like operations or even talking about financial and banking or like in healthcare or things of that nature, there's regulations that they have in pharmaceutical It's called GXP regulations. And in banking, it's the FFIEC regulations, right? Those regulations, like if there's change in an organization, like something changes, like you make a change, like a patch, like we talked about, but something changes, a new feature changes, it, that requires that because of these regulations that you have people, you have these controls and those controls have responsible parties and those responsible parties are responsible for those activities. So those regulations still exist. Now with the technology, you may be able to do more roles, right? You may be able to do more of those jobs and cover more of those use cases. But until those regulations and audit and separation of duties change, it can only go so far, right? Now you can do these things much faster. You can do them without rework. You can do them much safer. And you can compress it again into the reduction of, let's say, multiple roles into a smaller one to take 3 roles and combine them into 1 person because of the technology. In large enterprise, many of these roles are because of regulation and audit and separation of duties. So it can only go down so far. Now, I think there are some things that deserve to be more efficient, and I think those things will be more efficient. But I think for the most part, I think it's more about making people efficient and empowering people versus Versus reducing headcount?

I think it's more empowering if I wear more hats. It used to be like, I'm a data scientist, now I can do data ops, MLOps, and I can do data engineering all together. It used to take 3 people to do one thing, and now I can do all of it. And then in the meantime, put on a PM hat as well, in the— when you were doing the data analysis. So it feels more empowering, actually.

100%, Angelina, 100%. Like, to your point, technical People have spent tremendous amount of time in their technical components. But now, to your point, now I think there's a bridge for them to do more administration, to do some more of the business components, right? In addition to maintaining their acumen as technology people. So I think it is super empowering. And that's the same scenario that we were talking about for creative people. That's why I'm trying to encourage them not to get discouraged because of the structure that they find themselves in. today, this can empower them to be able to do their own, to follow their passions like they wanted to do. They went into creativity for a reason, right? It's the same exact thing that we're talking about for data scientists and developers.

Yeah. I feel I'm empowered to actually do more soul searching of what I truly want to do because of this level of empowerment. Otherwise, I would have to spend how many years to really understand a discipline to be able to tell, oh, I don't like it, or I I like it. Now I feel like I can just try it.

And you can get guidance on the steps to try it, right? Whereas before, when you were searching for, I remember when I was in the military, it was all on paper, right? We had library books of paper and we were drawing things on paper with pencils and it was, and now you've got this amazing resource that you can really find the best way, find the signal through the noise to figure out how to do something the right Yeah, I couldn't agree with you more. It's fantastic.

Can you give me one piece of advice to make me a better founder?

Think about the good things that are going to come from what you create and what you provide. And I believe that good people will seek you out. Good people who are financiers, good people who are, have mandates for acquisitions of companies will seek out good founders who are trying to really help the community. I believe that. I know there's probably a lot of people that think I'm naive for that. But I think if you believe that you'll meet good people and attract good people, that even when you're working out cap sheets and things of that nature, that if what you're building is really, if there's good energy behind it and it is timely and solving a need, I think I would say stick to your ethics and stick to your morals because people will want to have good people around you. And even if this endeavor that you have right now is not landing on fertile ground, the things that you've done today, And the next time that you go out and try, maybe there will be more opportunities for you to find fertile ground. So I think it's really about doing the right thing for people and for your community and really finding those business ethics to be able to solve those problems for people.

Thank you so much for today's conversation. I think TwoSet.ai should make this announcement now that we're going to partner with Joseph to have this problem bounty board on TwoSet.ai going forward. And so any builders and developers who are willing to solve a real industry problem, shoot us an email and let us know what you wanna work on.

Absolutely. I'm looking forward to it.

Yeah. Thank you so much.

Thank you. And thank you to the community. Thanks for the invite. And I'm looking forward to seeing all the, the business challenges that people provide and all the problems that we're gonna solve.