Yohei Nakajima: How AI Agents and Building in Public Are Changing Venture Capital | Glasp Talk #57
This is the fifty-seventh session of Glasp Talk.
Glasp Talk delves into intimate interviews with luminaries from various fields, unraveling their genuine emotions, experiences, and the stories behind them.
Today’s guest is Yohei Nakajima, General Partner at Untapped Capital and the visionary creator of BabyAGI. This autonomous agent framework sparked global attention in the AI community and was featured at TED AI.
As both a venture capitalist and a hands-on builder, Yohei is pioneering a new way of creating startups using autonomous agents to rapidly prototype, validate, and scale ideas.
In this conversation, Yohei shares his vision for how AI agents will reshape startup prototyping. He explains how multi-agent systems, generative workflows, and automation can accelerate early-stage innovation. He also discusses lessons from building over 100 projects, how founders can use tools like Replit and ChatGPT to uncover untapped opportunities, and why "building with agents" may become the new normal for venture creation.
From rethinking what it means to be a founder in the AI era to designing systems that think and build alongside humans, this episode offers a glimpse into the future of entrepreneurship. Whether you're a founder, investor, engineer, or simply curious about the intersection of startups and AI, this conversation will inspire you to explore what’s possible when you build with agents.
Transcripts
Glasp: Hi, everyone. Welcome back to another episode of Glasp Talk. Today, we are excited to have Yohei Nakajima with us. Yohei is an innovator, artist, and a venture capitalist. He is a General Partner at Untapped Capital, an early-stage venture capital firm that focuses on uncovering hidden gems in the startup world. With 15 years of experience supporting early-stage startups, Yohei has collaborated with global giants like The Walt Disney Company and Nintendo during his tenure at Techstars and Scrum Ventures. And he is renowned for his "build in public" philosophy, leveraging cutting-edge technology in transparent ways to accelerate learning, engage with founders, and fuel innovation. And among his most celebrated creations is BabyAGI, an autonomous task-planning agent that went viral in 2023, earning widespread recognition, GitHub stars, academic citations, and a stage at TED AI. Today, we will dive into Yohei's journey through venture capital, his groundbreaking projects in AI, and his vision for the future of innovation and investment. Thank you for joining us today, Yohei.
Yohei: Thank you. That was a great intro.
Glasp: Thank you. Thanks so much. So first of all, you are a GP and a cofounder and a GP at Untapped Capital, and you started the fund in 2020. But we wonder what gap or opportunity you saw that inspired you to start your own fund. It also has the investment thesis submission changed over time, especially after, like, a pre- and post-AI era?
Yohei: It is a good question. One of the opportunities that inspired Untapped Capital was an exercise I did in reflecting on what strengths I had. During my time at Techstars, especially when I was the director of pipeline, I was leading outbound efforts on early-stage pre-accelerator startups. I continued doing a lot of outbound at Scrum Ventures, where I was proactively reaching out to startups. As I thought about where I had a unique strength, doing outbound was something that I did not see pre-seed venture funds doing. You know, outbound is a very well-accepted sourcing strategy in every asset class, including late-stage venture. But in a pre-seed venture, a lot of VCs are still very network and relationship-driven. So that was the initial idea around it. In trying to identify it, kind of create an identity or figure out what the benefit of doing outbound was, you know, we realized, you know, I realized that one of the benefits is that we can connect with strong founders that are outside of typical networks. Right? Well, you know, a lot of venture capitalists, because they invest in their network, their networks overlap, so you have these, you know, handful of well-networked founders who are, you know, who are getting a lot of the attention. But there are a lot of great founders who do not have that network. And so our Fund One thesis was really around tapping into these founders outside of typical networks. The Fund has evolved. You mentioned build in public. That was a that, you know, that is something that evolved and became part of our strategy throughout the fund. Most recently, I would say, you know, this, this outbound strategy is now just one of the tactics we use in a more overarching strategy around being very top down at a pre CPC. Again, so it is going against the kind of classic invest in your network, you know, focus on building your network strategy, but more so, really focusing on building a process to come up with kind of conviction and hypotheses around the future, and then proactively reach out to and market to startups that that fit that future. So that is kind of the evolution of untapped in our philosophy.
Glasp: I see. And I was always curious. Why did you name Untapped Capital? What does the definition for you means untapped?
Yohei: I liked the word, well, we, as I mentioned, want to invest in, you know, founders outside of typical networks. Now we want to invest in ideas that are out, you know, that other people are not thinking about. I think the word untapped is associated with, kind of, you know, opportunities that other people are not seeing. Right? Untapped markets, untapped potential, untapped talent. And so untapped, I feel like, encapsulates a lot of, like, what we look for and how we try to work as a VC. And the name actually was, my wife came up with it, and I felt like it was the perfect name for capturing what we were trying to do.
Glasp: Interesting. Cool. Yeah. I like the name. Yeah. I agree. And since you are now famous for Baby I mean, you are building so many AI projects and and one of the famous projects is BabyAGI and mini Yohei, you know, ask anything or something and and also knowledge graph, and so on. But has that built-in public changed the way you get the like deal flow, or when you look for tracing founders and also, like, operations in your VC?
Yohei: Yeah. You know, since early on, a lot of the stuff I built was so that we could use it as a VC firm, or that maybe founders could use it, to help their business. And then, as some of the projects we build are more experimental, where BabyAGI falls into it, that is kind of the three buckets. Kind of investor tools, founder tools, and experiments are the three buckets we build in. I would say absolutely. I think I fell into this by accident, but what I found is that building publicly is such a great way to connect with founders and builders generally. Right? I feel like it is fun to just jump on a call with other builders and just immediately nerd out on, like, the building side of things, and I feel like it creates a unique relationship with founders that is not just as kind of a venture capitalist. And so what I love, especially, you know, through open source and through X or Twitter, I feel like I have built, you know, many relationships, kind of, just online through sharing projects with each other. And even today, I continuously, you know, would jump on a call for the first time with somebody where both of us know each other, have known each other for over years. And it is such a fun way to kind of build relationships in today's age.
Glasp: Yeah. Also, does that help you find a piece when you start a new fund, like a fund one, fund two, fund three?
Yohei: I mean, at this point, BabyAGI has definitely helped my name be more recognized. And when it comes to fundraising, of course, all that momentum and recognition and credibility are very, very helpful. And, of course, I feel like at this point, most of the people I meet, you know, are in some way, you know, many of the people I meet are in some way someone I met because I was introduced to someone who knows me because of BabyAGI. And, so BabyAGI is the gift that keeps on giving, is what I say.
Glasp: I see. And I heard that you made around 80 to 100 projects before making BabyAGI. So, not sure if it is true or not, but how do you come up with those ideas?
Yohei: Good question. I think maybe AGI was about project number 70, but I think at this point, we have over 100 projects. You know, I think I mean, I am not, you know, I am an idea guy. Right? I come up with ideas all the time, you know, playing with AI. I think that through playing with AI, there are a lot of ideas that, you know, I wish existed, or maybe, I am using ChatGPT in a certain way, and I think, hey, this would actually be a fun tool if this were broken out by it. And and I think everyone comes up with ideas. I think really the difference is that when, you know, I just have gotten into the habit of when I have an idea to just ask AI to build it for me and then see if it works. But a lot of it is, yeah. It is inspired by what I see online. Some of the ideas are, you know, from other people. I think Instagram, which was one of my first knowledge graph projects was I was not, you know, was that I do not I have a couple followers who were telling me that I would like knowledge graphs after they saw me do BabyAGI, and I I, you know, read up on what knowledge graphs were, and then I just immediately built a project that weekend with it.
Glasp: Do you when you come up with some ideas, do you keep the idea somewhere, or do you just go to ChatGPT or or Replit and then ask put the prompt to build try to build?
Yohei: It depends. Some ideas, if I have an idea, I might just open up Replit on my phone and just, like, type it in and just ask it to build it. If it is a more complex idea, sometimes I will go to ChatGPT first, or if I am, like, unsure if I have, like, a fuzzy idea, I might be like, hey. I feel like we could do this, and then I will brainstorm it a little bit with ChatGPT, and then I will just take it to Replit.
Glasp: So in that sense, has your tech stack changed over time? I know nowadays you use Replit, and I saw your recent in in your recent project, you use Replit, Resend for email, and so many tools. But has that changed over time? And also, what tech stack do you prefer nowadays?
Yohei: Yes. It has changed over time. I mean, before I started building with AI, I was, you know, I could do a little coding, but I was mostly using no-code tools. So my stack was Zapier, Airtable, and a code tool software was on top, sitting on top of Airtable. So it was there was a no-code stack. I started building with AI just around the time that ChatGPT came out. Right? I, and at that point, I had just started using I started using Replit. I remember I remember, like, loving Replit because in my prior attempts at using Python on my computer and using it in the terminal, I would run into things like having the wrong Python version or having the and, like, it is like, that was just something I never really wanted to figure out. Replit just made that all easy. So at that point, my, my stack was, having ChatGPT write code and copy-pasting it into Replit largely for their deployment. And then, while you know, that was back in 2022 that I started using Replit. And then they started building in, you know, AI kind of assistant features and then agent features. And over time, once the agent was kind of fully baked out, I noticed myself just using Replit Agent instead of using ChatGPT. So for a big chunk of time, Replit Agent was how I wrote and built stuff. My tech stack did change. When I was using ChatGPT and copy-pasting into Replit, my go-to stack was Python Flask, and I feel like I, you know, got pretty good at understanding how that worked over time. Replit agent, for some reason, tends to default more to TypeScript, I think. And so, I have noticed that more of my projects are built on that, but when I am using Replit Agent, I do not really look at the code at all. I kind of just have it built for me, so I do not really care about the stack. As far as coding tools, I have started using OpenAI Codex. I do like the kind of more fine-grained, you know, doing PRs into GitHub. So I have a couple of projects that I will kick-start on Replit Agent and then maybe, you know, start working on it with Codex at the same time or in parallel. And then you mentioned APIs. One of the things that often inspires ideas or projects is experimenting with an API. So I remember, you know, when I got when I started playing around with the, you know, Hume AI, I did a project where I connected it to Twilio and gave ourselves a phone number. And I really just did that because I wanted to kind of play around with that API. And I have done the same for, you know, portfolio companies like Videodeby, which is a media-first API. I have done a couple of things with Falcor DB on the knowledge graph side of things. I think most recently, I wanted to try the X API, and that kind of led to building VCPedia, which is kind of a Crunchbase that is built on top of Twitter data. But that project was, yeah, it was inspired by the desire to play with an API that I thought could be interesting to pair with LLMs.
Glasp: Interesting. But when you use API, when you play around with APIs, and since you said you do not check code and and and so on. So how do you evaluate, oh, this API works well? And does it depend on the outcome you get? Or yeah. I am curious about how you.
Yohei: Oh, so I often feel like I have a pretty good memory for startups. So I remember when I was thinking about doing a daily newsletter for VCPedia, I felt like I had seen Resend on Twitter a couple of times, and I felt like it was kind of an up-and-coming, you know, API first email service. So I thought, I you know, and I feel like, you know, newer APIs tend to be kind of simple. They take advantage of the latest and greatest. So, I felt like it was a good opportunity to try it. Also, I like the idea of just supporting early-stage startups and experimenting with them as well. It is just kind of a fun way to build relationships with founders. It is just like using their API and open-sourcing projects on top of it.
Glasp: I see. And so do you have any ideas you are currently working on, but not published or not released yet? If you can share.
Yohei: There is always a lot of stuff I am working on. At this point, I usually have multiple projects working in parallel, and I kind of decide what I want to work on just based on how I feel. And actually, you know, a lot of projects, some projects do not even make it to completion. Right? Because a lot of these are experiments. I will just ask Greta to build it, and then, like, sometimes it does not work the way I thought it would, and then I and then I will just, like, move on to the next project. Most recently, I have been working on VCPedia, kind of slowly fixing that. That is one of the things I have been working on. I have a bedtime store bedtime podcast generator that I kind of put off because I tried to translate it into a bunch of languages, and then I recently resimplified it. So I am working on that now. I was working on that last weekend a little bit. But the biggest project I have been kind of tackling, I keep coming back to, is this unified CRM calendar email client note taker task management tool. And so I tried to build. I have tried to build I have tried that, like, 12 times, but, yeah, every time we feel like I think of a better way to architect it, and I just, like, start from scratch. But, but I am but I am hoping I will get to the point that I like soon with it.
Glasp: Interesting. Yeah. I can not wait to see it then. But I was always wondering, like, since your project seems more like prototyping fast, then there is it. Right? And so it is like I see, like, it is a kind of small project. But if you have an infinite amount of resources, like talents, money, and so on, what would you want to build? Do you have a big idea? Because you are investing founders, they should be they should have big picture auditions, you know, like a vision. But if you have that amount of resources, what would you build?
Yohei: That is such a good question. Well, the things that popped to my one of the things that popped to my mind is what I described earlier. Right? I think it is kind of surprising that there is not really a good email client that is also a CRM that is also a task management tool. The only thing I can really think of is kind of Google Workspace, but it kind of feels like separate tools, and Google Contacts is not really a CRM. You do see some sales tools like that, but I have not seen that on, like, a personal side tool. And I feel like that would be difficult to do, but if I had infinite resources, that would be something I think would be really fun to work on because I have just been thinking about it so much. See. But, man, I would probably build a I do not know. There are a lot of small ideas I want to build. I thought, what would be really cool is, what would be cool is to train a small local model specific to survival techniques and then create a very rugged hardware that has access to it. I think that would do really That would be a pretty interesting project. Interesting. Kind of like a survival kit, but like a survival LLM, but it runs locally and, like, on a curable device or something like that would be pretty interesting. I do not know. You would need to power it, though. But that is an interesting idea.
Glasp: But in that sense, since you are, yeah. You know, you said you are VC by day and and builder by night. You know? Yeah. But have you, since you talked to a lot of founders and you built so many things, have you thought about becoming a startup founder, not a VC venture capitalist?
Yohei: I get that question a lot. Oh, yeah. I love being a VC. Right? I remember early on in my career. Right? I think one of the career advices that really stood out to me was, somebody said, just ask yourself what type of people do you want to be spending time with ten years from now? And that was like that to me was a very eye-opening way to think about a career. It was not about, like, what do want to do. It was about who you want to be spending time with. And, you know, it was immediately obvious to me that, like, I wanted to spend time with founders, like, early-stage founders because they are always working on, you know, new interesting ideas. It is such a you know, I mean, you have to have this, you know, conviction and and passion for to, like, risk a lot and start a company. And so that has always been my north star was I wanted to be in a position where I could work with founders. And so it was never you know, and I, at least based on observation, being a startup founder is pretty busy, and you do not necessarily get to spend time with founders as your job. You do it as part of networking. But as a VC, I get to, you know, meet interesting founders all the time, and, I would never I I do not want to give that up. But maybe someday, if I had infinite resources, I would start an incubator so I could launch multiple companies.
Glasp: Yes. And, also, you know how to use AI tools. And you are building AI agent tools as well. So when you meet founders or when you are thinking about, oh, deciding if you should invest in the company or not, and what kind of traits or personality do you look for in founders? Is that a team, idea, market size, and what aspect do you see in founders?
Yohei: Yeah. I think one of the things that we I think we are kind of intentional and slightly unique for us, again, going back to the top down, is that I tend to, diligence kind of the market first in the sense that I I want to spend time talking to founders, building in markets I generally like, I feel like are growing that that would make sense to invest in as a pre seed investor. And then that way, when I actually start talking to a founder, I can really just focus on vetting the founder more and how they think about the space and the problem. Once I start talking to a founder, I mean, the more kind of obvious, you know, I think the classic things that people say, you know, are like grit, obviously. Right? Like, do I think they will get through the tough times? All that kind of stuff. And there are a couple of things that I probably look for. But, a few of the ways I think about founders, especially for, like, an early meeting, I often think about how excited I am to share the like, introduce the founder to other people. And that usually, if I am excited like, I sometimes I will jump off a call and I am like, oh, I can think of so many people that I want to tell about this founder, and that feeling has to come you know, there there has to be a couple things about the founder and the idea that make me feel that way. So that is one of the feelings I look for. If I were to work in this industry, would this be a founder I would want to work for? It is a question I also ask myself. Right? I think one of the important things as a founder is to be able to attract strong talent, and I think, again, not not universal, but, right, if I, you know, as an investor, I am I I want to be a a you know, I consider myself part of the extended team, and and I want to invest in founders I would want to work for. So those are a couple of things. And then, of course, team dynamic, kind of founder market fit. Like, does it feel like the right founder to be making an impact in this market? Again, it is a lot of feeling-based stuff, I think, which is okay at Preseed because there is just not that much data, but those are some things.
Glasp: I see. And since, you know, you have talked to a lot of founders, I think. So what market industry space do you think the most impacted by AI, or not impacted yet by AI or AI agent?
Yohei: That is a good question. Well, the ones where we are seeing a lot of usage, right, are coding, like, development, and marketing. I think that makes sense. Right? Development occurs because the people building the tools are scratching their own itch. And then I think marketing makes sense because, you know, writing content is something that LLMs are very good at. I think sales is another area where we are starting to see a lot of, lot of activity because it is, again, it is, you know, word-based. It is scaled. It, you know, is personalizing emails, those kinds of things. It is, LMs are very, very suited for. Where it is going to be slower, that is a good question. I think our I think I mean oh god. I want to say, like, people relationship-based businesses, but I think I am just being biased in thinking about VC there. Regulated industries are always a little bit challenging because you have to go around the regulations.
Glasp: Like a health care or those things?
Yohei: Yeah. But I guess it is being used a lot in health care, too. That is a good question. I mean, honestly, AI is impacting many industries in some way or another. Right? I think for some companies, at enterprise, for example, I think we are going to see more internal agent use case first before we see external, especially for kind of larger companies where there is risk in external agents. I think that is something we are seeing today. I think executive planning and management, I think, is still harder to do. I think we even though it can reason, you still need to organize it, like, figuring out how to organize the information within a company so that an LLM can help at the help kind of plan at the highest level, I think that is that will be still challenging, and, we are not quite there yet either.
Glasp: Do you have? I do not know if you could share this, but do you have any examples of founders or startups you have recently invested in or you sought to invest in, or not?
Yohei: I am happy to talk about my portfolio companies. Oh, I think one of the fun ones, I think, that we recently did is a company called Layers, which is an automated marketing and ad platform targeting developers specifically. It lives in the IDE. So when you are ready to when you finish developing something and you are ready to market it, you install Layers into your development environment. It will then read the code to figure out what you built, and based on that, it will figure out who you are marketing toward. Based on that, it will come up with a content strategy, and based on that, it will start automatically generating content for Instagram and TikTok as kind of the first channels. And then based on engagement, it will suggest ads for you to run. And you can manage the entire campaign, solely from the development environment, without ever leaving.
Glasp: That is so smart because developers do not want to, they want to focus on, like, keep coding and
Yohei: Right. They do not they not only do they not want to work with a marketing person, they do not even want to go to a marketing app. They just want to stay in their development environment. And I just thought that was such a beautiful future that they had envisioned.
Glasp: Do you? Oh, I see. Interesting. Oh, okay. And you have seen many ideas. Also, you came up with many ideas. So what is a good idea in your definition? And what kind of things and aspects good idea meet?
Yohei: That is a good question. I do think, you know, it is it has to start with the customer and the problem set to some extent. And if you again, for a business, for a startup idea. But, for me, some of my projects are not actually great startup ideas, but they are good ideas to go kind of viral. And to some extent, because some of the buildings I do, one of my goals of building is to, you know, is content marketing for us to some extent. I think a good place to think about, you know, one idea that fits into both categories is, like, what can you build today that would have been hard to build, you know, even a year ago? Like, what can you build, you know, again, to if you were to ask that today, I would say, you know, looking at tools like VOD three, I think it is much easier now to build an interesting kind of movie generator tool. It would have been much harder to build a year ago. So I think if you are really starting from scratch and looking for ideas, yeah, I would I would look at asking myself that question.
Glasp: I see. Then so yeah. Yeah. So if there are startups, so investing in those, like movie generators, but eventually Google or ChatGPT will replace them. So would you invest in it? So if So yeah.
Yohei: Again, okay. So I guess the movie generator idea fits more into, like, I think it will work well because it will seem like a lightweight tool I could build an open-source tool today. Right? Because if, especially if I had you know, if you use VOD three, you can make a very, very you know, you could probably write a single script that will generate a five-minute film for you. So that is where that falls into. When it comes to actually investing in a startup, what I like to see is what I like to focus on, like, how much is like, how fast is the founder learning? Right? So, usually, when you start a company, you have a target customer segment, and there is probably a problem set. And your hypothesis is that solving that problem is a good business. And so that is what the, that is what the startup is, an initial experiment around a hypothesis. And as you try to answer that question, sometimes the answer is no. That was not the right problem set to solve, or there is a problem, but no one has money. Just pay for it. Right? You learn these things. And then but you might uncover a different more a different opportunity. Or maybe you identify a narrower ICP to start with. And so, when you talk to a when I talk to founder, what I want to hear is how much they have learned through trying to solve that problem. And that can be in the form of customer development. Right? I love when I hear a company that says, Hey. We talked to 300 of our potential target customers and then decided to build this. Like, that always perks my ears. Right? Because they are taking the time to go talk to people. I also like product people who just build a prototype and put it out on the market. I mean, open source is a great way, but even if you are not open source, just getting it into the hands of a customer. And then, what I want to hear is that, like, hey. We pushed this out. We found that, you know, early users really like this, and, you know, we realized that these are things that did not work as well. So we are going to go, you know, continue doing this, and then we believe that it will continue to grow and people will stay on longer, and so on. So, I yeah. So that is that is kind of what I look for, I guess, is, is there speed at learning? Because the initial idea is easy. Right? Again, if it is something like the video generator. But if you put it out there, if you try to get people to use it, then you learn something completely different than just, like, building it on a weekend and pushing out.
Glasp: Also, it makes sense. And in that sense, there is a famous saying, like, a first-time founder cares about folks about, like, a product, and second-time founders care about distribution, which is important in the AI era and also in the past. You know? People say, Oh, execution is everything. But do you think it's true or idea matters more? Do you have any take on this?
Yohei: I mean, when it comes to success, I am firmly all you know, I like the phrase, the, you know, success is 11% inspiration, 99 perspiration. Right? Like, the idea is really just the start, but it takes a lot of hard work to take an idea and turn it into success. So I do believe execution is everything, but that is a very broad term where execution covers everything. Right? Just like you doing stuff instead of just sitting there thinking? When it comes to product versus distribution, especially in today's AI age, I mean, you definitely need both. Right? Because you need people to use your product well, to learn from them. Also, because the bar to build stuff has become lower, and because everyone is focused on AI, it is hard to get attention in today's market. Right? You, there is likely to be, unless you pick a very small target niche, you know, for many companies, there are a lot of competitors out there. And, like, so getting the attention of the right people is harder, so it is challenging. Now that being said, I do not think you could say that distribution is everything either because at least in today's market, people are still experimenting with different AI tools, and they are willing to switch AI tools for one that is better. And so just because you have people trying it does not mean they are going to stay and stick either. Right? So you also need a really strong product to keep people in. And I think the result of combining the two, and where I see companies really kind of breaking out and, you know, becoming a category leader, is having both. Right? You need strong distribution and a strong product. And when you have both, you build a strong brand, a reputable, credible brand. And once you have built that, that feels like a pretty strong moat because then you become the less risky option for people to adopt. And so it is so much easier for, you know, let us pick text to voice. It is easy for an enterprise, you know, somebody in an enterprise to say, let us use Eleven Labs because they have such a strong brand reputation, investors. And so that is what I think, where you want to go. But I do not think, yeah. Again, I do not think just product or just distribution is enough. You need both.
Glasp: I see. Yeah. Makes sense. And so some people, like, investors say, you know, thanks to AI or because of AI, so startup raise one time, then they can build so many things and distribute more so that once they reach to product market, we they do not need to lose days money so that if you as as a investor, if you missed a first round, you are going to pay more. So is that happening? Is that true?
Yohei: I do think there is a, it seems like the idea of what I have, I have heard people call it seed strapping. Right? Just raise the seed round and then just, like, basically grow that. I think that idea, I mean, just because people are talking about it more too, and the fact that people are talking about it as a trend, I think it is kind of a self-fulfilling prophecy, is that probably more people will be conscious of that as an option. That being said, I do think that, you know, there is often if there is opportunity to if you can put a dollar in and get $3 out, right, then why wouldn't you raise more, like, raise more money to accelerate that cycle? Like, to me, that like, I feel like there will always be a core venture growth pattern for especially the fastest growing companies, where if you can figure out how to deploy capital and generate the return, then you will, you know, generate growth and you you you will keep doing that. That being said, I would like to see more seed-strapped companies. Right? Companies that raise a little bit and then, you know, become self-sustaining and grow really large. Not to talk about, like, venture capital. It does, it does, kind of may, I think VCs, that does happen more, we will have to figure out how to value companies. Venture capital has to change a little bit because today, we value companies based on funding rounds. But if you have a company that raises a small amount, then it just grows by revenue. At some point, we have to learn how to do, you know, revenue-based multiples even at the earliest stages, perhaps, which is hard.
Glasp: Yeah. I see. Yes. And I was curious about the future of VC because I think you are one of two examples, but you automate so many VC processes, I assume, and using AI, AI agents. And do you imagine more VCs becoming automated or data-driven as you have, or adapting to building public assets? And, meaning if so, meaning, if more VCs automate things with AI, what could be the differentiator for the VC? Because if they have the same processing, does that a brand that only matters?
Yohei: It is a good question. Well, I do not think venture capital will, I mean, it is not going to change overnight. It is a very slow industry in terms of innovation. The life cycles are long. Funds yeah. Every fund is a ten-year life cycle. Any new strategy takes ten years to prove. There is little incentive for people who have been successful for twenty years to change their strategy, especially if they have been successful. And there are few seats at the top, which means the turnover of, like, people kind of one generation of general partners, you know, leaving next coming up is also slow. So this is a slow-to-change cycle. That being said, yes, I do think VCs will increasingly become data-driven, and it is a trend that has been going on since before GenAI. You know, we have seen more and more, especially at a later stage, kind of data-driven funds. I think we will see a lot more VCs leverage AI agents, either in different aspects of it. Right? I know some people who, you know, like us, who use it for diligence. I know some people who are using it more for communication or for scoring companies. Right? So there are a lot of different areas of VCs that you can automate. That being said, I do not think the way comp VCs diligence companies is the same. Right? I think what you decide to focus on, what you decide to weigh, what kind of data you want to collect, is different as a VC. And I think, you know, that is one area that is going to differentiate you. It is like, what are the questions you are asking the AI in the first place to come to your decision? And I think it is hard to automate because it is not just about, like, increasing the odds of every investment. It is really about looking for outliers. And I think that is, I do think finding the outlier is just a harder thing to do with AI. Right? Especially in, like, an immense power law where, like, you are going to invest, you know, if I see a, you know, scoring is a good example. I do not really use AI to score. Right? If I look at 100 companies, I know if I am probably talking to one out of every 100 companies I come across, what if you are talking about, like, an idea at the idea level. Again, it is hard for me to think that I could I could build an AI that would pick the same one in a 100 that I would every time. Right? It is so important for me to, it is so important for me to talk to the right com spend time talking to the right companies, that, yeah, again, even if every VC were to try to automate everything, I think a lot of VCs would still be very different. And to your point, I think branding, how we support portfolio companies, how we implement it is will still be a differentiator. But, again, I think it will take time before we see that we will say that a lot of VCs are largely automated, I think, because it is still far, far away.
Glasp: But but in five years, ten years, will it be fully automated? Will we have a fully automated agent fund or something like that?
Yohei: I mean, the fund does I would like to I would like to think I am I am I am a candidate for building that. Yeah. Maybe five to ten years. Again, I do not think it would be largely the industry, but, like, as an experiment, I would like to see a venture fund that is largely automated, in, I think, five to five years, is possible. Again, it might, like, it might not be good, but at least it will be a first proof of concept. I think it will exist soon. But, yeah, there are some of the challenges in venture, though, still is that there is not, like, a button you can press to subscribe to companies as easily, right, at least for the more typical rounds. But, again, like, right now, the sec the later stage secondary markets, for example, are a little bit more liquid. There are a lot of kinds of marketplaces and platforms where you can buy second-hand. So that might actually be somewhere we will see the first kind of I could see an automated secondary venture fund. It would probably be easier to build, especially if any of these marketplaces have or build APIs.
Glasp: Oh, by the way, I recently saw an interest in a tweet on, sorry, XN, so that, like, oh, this is a SPV or SPV or SPV or something like that.
Yohei: Yes. I think that was today. Right? The third layer SPVs were Oh, yes. That is right. Yeah. Yes. There is an SPV, and then somebody takes a portion of that, turns it into an SPV, and then someone takes a portion of that SPV, and then you have three layers of carry, which is just horrible. Yeah. If you come across a three-layered SPV, you should never invest.
Glasp: Is it common? Is it becoming the popular thing, like SPV or SPV, or is it just randomly occasionally happening?
Yohei: I think it is I, it is probably happening more now. Right? I would guess that it is happening for companies like OpenAI and Anthropic. Right? Because those are typically hard rounds to get into. So someone gets allocation. They get a pretty large chunk of the fund because the funding rounds are so large. And then so they do an SPV to fill that. And then so even the second-layer SPV, you can still get a decent chunk, and you can build an SPV on top of that. But then, even if you have a small allocation into something like OpenAI or Anthropic, so many people know about it that people feel like, Hey. I could probably scrounge together, you know, a million from my friends who would want to put money into Anthropic. But at the later stage, I almost feel like, you know, it is like, from my perspective at Preseed, the later stage is, like, expensive and and and, you know, less risky. But from a public stock investor perspective, right, at least hypothetically, these SPVs in the late-stage companies are cheaper and riskier than their typical investment. So there is demand, right, to like and get into the next AI company before they go public, assuming that they will do well in the public markets.
Glasp: I see. And
Yohei: But caveat caveat, that is not where I invest, so I would not take my advice on any of this stuff aside from the third layer fee thing.
Glasp: Yeah. Sure. Yeah. Yeah. Yeah. Thanks. And as a VC, so what kind of support do you want to give to funders, but you could not have done it?
Yohei: Well, like, what kind of support would I like to give founders but but can not?
Glasp: You bet today? Yeah.But due to, like, you know, resource limitation or something. Yeah.
Yohei: I mean, I would love to have, you know, a full-on partner partnership, you know, kind of a staff to support companies. I think there is always more that you can do for portfolio companies, but that is not as exciting an answer because a lot of venture funds have that. Mhmm. I would love to build an awesome portfolio support AI agent with access to a lot of tools, actually. I just have not gotten around to it, but I actually feel like there are probably services like APIs. I mean, you know, I will pick, you know, Norbert as an example. It is one of my email finding tools that I like to use, and if I have API access and I give that API to my agent, now I can help all my portfolio companies find email addresses of anybody they want to. Right? You can now if you start thinking about other APIs you could give an agent, I could imagine a, you know, kind of yeah. I should build that. Like, an associate that you can ask as a portfolio founder for, like, help, and it has access to APIs that the VC is already paid for.
Glasp: Yeah. We would love to use it. Yeah.
Yohei: Okay. Okay. That is a good idea.
Glasp: Seems like you are the one who is really obsessed with, like, AI agents and how to leverage AI agents. But why an AI agent? And what intrigued you to leverage AI agents? Yeah.
Yohei: I think, one, I am lazy, and I do not like doing the same thing, and bored easily. So I do not like doing the same thing over and over again. So if it is a repeat task, I want to try to automate it. And so I have been, you know, even before I started using AI, I was a very heavy Zapier user just using no code. I think at my peak, I was running 20,000 Zap tasks a month. Like, I just want I love automating things. I love collecting data, all that kind of stuff, collecting and leveraging data. So I have always been, about like, if I have to repeat something multiple times, there is this, like, x k c d chart. I am like, depending on how long it takes, how often you do it, you can kind of calculate how long you should spend automating it. Like, I feel like that is always going on in my head whenever I am doing repeat tasks. So that is probably what drives a lot of my desire to build and experiment with, you know, what is today called AI agents. But, again, automation was it was it was always part of the way I worked from before. Specific to BabyAGI, and, actually, this will go back, I am going to go back to an earlier question you had about whether I had unlimited resources. BabyAGI, initially, I sat down with ChatGPT and said, I want to prototype a startup autonomous startup founder. And I described what eventually became BabyAGI. I said, as a startup founder, I wanted to figure out what it should be like, figure out come up with a task list, store it, and then execute one task at a time. And then, as it completes a task, I wanted to think about any new tasks it should do and add them to the task list. Oh, and let us have it prioritize tasks during this loop. And that is kind of what I, like, initially typed into ChatGPT, and that eventually turned into BabyAGI. But I would love to, I would love to build, like, an autonomous startup incubator. That would be really fun to do. Again, I think it is, like, maybe a little bit too early, but I but there are probably some simpler business ideas you could execute on. And if you kind of built the muscle now, I feel like you could you could do something pretty interesting.
Glasp: Interesting. Yep. So it sounds like a repeated question, but since BabyAGI and has your perspective on AI agents changed? Like, the potential that they have or things they can do, or are they the same?
Yohei: I mean, it was always a very early experiment, and I like to think that, like, I feel like BabyAGI, like, almost falls into the research category. Right? So one of the things I learned is that it is hard to communicate these things, because they get hyped up very quickly. So I think some people were excited about the potential. But, you know, one thing I learned is, you know, again, through building multiple times is that there is a lot of work to do to get them to work really well, especially if to to get to, like, a dynamic general autonomous agent. I think there are still a lot more things we need to figure out to get there. So one of the things I think I learned, I think I have come to learn, is that, like, the simpler, easier workflow agents are much easier to get value out of today. So that is probably an easier place to start. And then, as you probably want to build those modularly, so that you can start testing more dynamic flows. And I think coding agents are kind of right there. And then but once you can start doing dynamic stuff, as soon as it turns into kind of a longer-term task, I think a lot of the agents still break and are not that good yet. So I think of next step is probably more project planning, organizing data. So, yeah. So I think there is there is there is still a lot of opportunity to make AI agents better and more useful. But, but I do, but but I absolutely Mhmm. Gung ho and believe in their eventual potential.
Glasp: Mhmm. And recently, I have seen a lot of news about context engineering. So do you think context engineering helps an AI agent better, like, improve?
Yohei: Yeah. I mean, I think the idea of context engineering, right, is is that, you know, assuming AI has access to all the data in your company and and all your chat logs and everything that you have ever done, how to pull the right information to handle a task or how to pull the right information to, you know, provide an answer gets increasingly complicated. And so, absolutely, I think context engineering is key, is key to get you know, building more, reliable AI agents that can, that are helpful in more complex environments. But I do not think it is, you know, context engine makes it sound like it is an engineering problem, but I actually think it is really more of an organizational problem. Like, you have to understand what the key questions are. How should you organize the data? What kind of questions might get answered? So, it is really not just, for at least for, like, an enterprise use case, it is really not just an engineering problem, but it is a that you have to understanding your company at, like, a deeper level almost to be able to do it correctly, I think.
Glasp: Totally, Yeah. And it does make sense, like, in the age of AI, what will be the long-term moat? I think you get this question a lot, but what could be the long-term moat for startup founders or startup founders? Is it only speed or execution?
Yohei: I mean, I think yeah. I mean, earlier on, I do think speed and execution are really important, especially kind of the rapid cycles that help you figure out the problem you want to solve. I think, as you start growing, it kind of goes back to, you know, what I mentioned earlier. I think, you know, what you want to get to is building a, you know, reputable, trusted brand. I think trust is hard to build, but once you build it, it is a huge leg up. And I think that feels like a moat, kind of the trust piece. Again, a moat is kind of a it might not be the right phrase for it, but I think branding is important. And then, and then an idea I have been playing around with a little bit is this idea of, you know, well-structured, memory or user data or, you know, user context, I think, becomes, kind of, a motive source. Right? I benefit from going back to ChatGPT a lot because it has a lot of my historical context. So I really like that I can just refer to a previous conversation I have had. I can even just go in and say, Hey. You know, I have to talk about X based on what you know about me. Can you just, like, draft me some points that I would talk about on this topic? And it does a pretty good job. Interestingly, right, with network, with social networks, there was, like, the more people you knew on the platform, the more valuable the platform got. But with something like an AI, I think just the more it knows about me, right, the more valuable it becomes for me. So that is, I think, another boat of sorts.
Glasp: Yeah. Regarding memory, I like ChatGPT's memory feature, and that remember it is about me what I asked, what I like, and what I am interested in, so that, like, the answers can be personalized to the query and to my intention and my interest. And that is why, I mean, we are trying to build a memory, not Yes. From chat history, but from, like, browsing data. Like, I mean, like, bookmark, and so on. But do you see this? And, also, listen to it. I see many people try to build in the memory layer, like a centralized memory bank management system or something like that. What are your thoughts on this memory layer?
Yohei: One, I do think it seems like a good place to like, there is a lot of op. It seems like there is a lot of opportunity for improving how AI systems and agents work at the memory layer. So I like a lot of these startups launching because I think these are companies, especially these open source ones, publicly sharing and experimenting on how to build a, you know, memory system that makes AI agents, you know, work better. And I think that will be used to build, you know, various AI agents. I think separately from that, right, like, each AI agent kind of needs to learn. Separately from that, I think there is this idea of I want my memory to be like, want, like, a composable memory where I can bring my memory from platform to platform, like, kind of a plaid for memory, or I think of it as, like, a memory cartridge. I would love for that to happen. I do not know exactly if it will, but I think that is a really cool idea. I think you could probably hack it to some extent.
Glasp: Meaning, hack it means?
Yohei: I was actually thinking, like, if you built an MCP server that Mhmm. As a tool and you instructed the agent to always use the MCP server to store the history of a conversation, so you could kind of extract history from various tools. Right? If I gave if I had one memory system and an MCP server and I gave that MCP it is like a memory storage MCP tool, and I gave it to my Claude Agent, and I gave it to my OpenAI Agent, hypothetically, it would build a central memory system that that the MCP tool is, like, automatically extracting history from.
I know. Yeah. And that is plenty of hacking it. Yeah. Yeah. Like, having a separate memory system, and then maybe if I use a new tool, I can just connect the MCP into that memory. And if I ask a question, it can it will look into that memory to see if it remembers.
Again, it does not feel as clean as a fully baked in-memory system, but I I do like the idea that I like the idea of not having to, like, reexplain myself to every new AI tool that I use.
Glasp: Yep. Me too. I do not, I do not, yeah. Me neither. I do not want to explain again and again. Yeah. Makes sense. And this is a little bit different topic, but I am personally curious about how you keep up with the new developments and then the information and news nowadays. Because so many, I mean, every day when I open Twitter, I see, oh, I launched this and, oh, this is a new release and so on. And it seems you are busy and talking to founders, and then also building stuff. How do you catch up with this news?
Yohei: Honestly, I mean, Twitter is incredible. I think that is what I use a lot. I do try to filter my feed so it is not as distracting, and it is mostly, like you know? Because recently, I think the algorithm changed and started sharing a lot of kind of, more general posts, like these kinds of videos that are meant to go viral, that I would expect on TikTok. I just immediately said not interested, and so I quickly tried to filter those out. So my feed is mostly, like, founders, VCs, and AI content. And so I do get a lot of information from Twitter. You know, I would like stuff. I will bookmark stuff pretty often, I will and I will go through them as well. We do, you know, I do kind of regular, you know, kind of quarterly sessions with our LP, telling them about AI trends. So at least once a quarter, I am going through my Twitter bookmarks and, like, trying to, like, rem like, kind of recollecting, like, what was interesting. And so I have this kind of, like, second feed. Like, the first feed is, like, my for you, and I am, you know, aggressively liking and bookmarking stuff that, like, immediately sparks my interest. And then, you know, on a regular basis, I will go through that again and pull out stuff that still stands out to me after a little bit of time. So there is kind of the two layers of filtering, starting with my Twitter feed of but, yeah, basically, I mean, the short answer, Twitter X.
Glasp: I see. By the way, next to follow for you on XN, there is a following tab. Right? And so that you can only see the following people. Do it. Don't you use that tab?
Yohei: I don’t use that as much. I like the forum for you a lot because I also discover new projects. Right? I think, you know, a lot of I got a lot of my followers through BabyAGI. So a lot of the people whom I have followed back are also in the AI space. And, I feel like some of the founders of some of the startups I found, I am guessing I found because somebody I follow or somebody who follows me happened to know the founder and, like, liked it or retweeted it, and so it showed up on my feed.
Glasp: I see. Do you use any, like, note-taking apps or press to save something?
Yohei: I use multiple. I mean, I still use Airtable as my core note-taking app for meeting notes. I use Twitter bookmarks as my as as where I store, store my Twitter likes. I do not have I do not use anything like Obsidian or Roam today. I was using Pocket a lot to store startups that I liked, but they shut down. So I Yeah. Built a little prototype, but it does not work as well. I would love a Pocket alternative prototype from a bookmarking tool from Glass. I absolutely use it.
Glasp: Thanks. But why did you know, prototype was not working? Like, Pocket alternative?
Yohei: Well, it is more Lucy, like, it is a tool I use all the time because Pocket is the bookmarking tool I decided on pretty early on for bookmarking startups. So I have a whole bunch of automations that automatically trigger. Mhmm. So if I like a startup website on Pocket, it automatically stores it on my Airtable. It scrapes the website. It adds a description. So, Pocket is my just happened to be my inbound or kind of my bookmarking tool I chose, mostly because it had good Zapier integration. I also like that it had a mobile. So even if I am on Twitter and I find a startup, I could open up the URL, and I could press share and press pocket, and then it would kick off, my it would trigger automation. So I like that it was a unified bookmark both on mobile and web. And then when they announced they were shutting down, it just seemed like it is it seemed like a tool that I could vibe code in a day. So it was mostly just triggered by the shutdown news, and I thought, oh, I should I should prototype something today since I just saw that it shut down. It was a little bit of a challenge to see if I could do it within twenty-four hours.
Glasp: Yeah. To be honest, actually, I actually saw the news. I mean, Pocket is shutting down, and we decided to build our own importer, like importing highlights from Pocket and so on, you know, bookmarks as well. But I was actually shocked by how fast you built your project. And I was I I was using, like, a Casa, like, those, like, a quote, those things. But you built way faster than I, and it is shocking.
Yohei: It was Replit. I did not touch the code at all. I just described what I wanted. And, again, Pocket is a relatively simple tool, so I think that was a very good vibe coding project.
Glasp: Yeah. Also, how do you use Airtable for meeting notes? So, you said that you take notes, meeting notes, on Airtable. Right? So how does it work?
Yohei: Yeah. I have a notes table.
Glasp: Notes table?
Yohei: And then I you know, one of the columns is long text, and I put I put my notes in there. One of the things I like about the way I take notes, or, I guess, the way it is set up in Airtable, and that is surprisingly hard to do in a lot of other CRMs, is that a lot of CRMs do not let you attach both people and organizations to notes. Like, often like, in a lot of CRMs, the note is attached to a person, or the note is attached to an organization. But I just, for the way I think, I like the idea of being able to attach people or organizations to a note as a separate attachment. And so that is how I architected our Airtable CRM.
Glasp: Just is it similar to, it seems like Notion can handle this. Like, can you do this?
Yohei: Yeah. I mean, I don't know. I think Notion and Airtable are actually pretty interchangeable in terms of tool use. They both have a lot of automations. I think Notion is, you know, probably better if you want document-style organization. Airtable to me, I probably think more in, like, spreadsheets and tables. So, like, Airtable to me just, like, felt more like a no-code back-end database. Right. I see no, yeah. Airtable felt more like a database.
Glasp: Do you use Granola or any AI note-taking tools, like a Zoom integration?
Yohei: I right now using Fathom for Zoom right now. What got me to use it? I remember they had a Zapier integration, which I thought had just made it easier. So my Fathom meeting notes can go, can get picked up by Zapier, and they get stored in Airtable. And then I can, you know, I can, tag the users that my Fathom notes are in. And so I like that it could trigger automations quickly me easily for me.
Glasp: So and and what else, like, do you use Airtable for? Like, what kind of other automations do you have? It is really interesting.
Yohei: I mean, I use Airtable as well. I, you know, I do not do this as much, but, you know, I still do it. Airtable has these linked columns that can convert a comma-separated array into multiple entities in a separate new table for you. So if you go to Crunchbase and, for example, if you do you know, let's say, you search for series, you can do a search for funding rounds, for example. You can search for series A funding rounds in healthcare AI in the last two years. And if you take that and export it from Crunchbase, one of the columns will be investors, and it will be a comma-separated list of investors. If you upload that to Airtable and turn that column into a linked column, it will automatically create a unique list of investors with the rounds tied to it, and then you can add a count column to it, which means you can you can will automatically count how many times each investor shows up. So you can see who has been the most active series A investors in health care AI over the last two years. So you can do things like that data manipulation. That is a little bit easier in Airtable than it is in Google Sheets, so I use that for that. I also just use it as a place to store data. So, for a while, I was one of my zap zap automations was anytime and and I think Product Hunt had a, I think, had a Zapier integration maybe, but I was storing every single Product Hunt. Mhmm. Product in an Airtable. It is like a massive Airtable. Just like every day, it was just like anytime there is a new Product Hunt product, it was just stored in my Airtable. And I did some, like, keyword analyses, like, internally to, like, try to capture trends on what people are building on Product Hunt. So that was, like, a side project that I did.
Glasp: And, Tracy, how much do you spend on Airtable per month? I mean, what tools do you spend the most on per month?
Yohei: Not that much. I think Airtable is more expensive. It is, like, more of a per-user. And, like, our organization is not big. So it is not too expensive to run. If anything, Zapier got to a few $100 a month at some point, but, again, it is not that expensive. I think I was on I was buying, like, a $3.99 a $3.99-a-month plan or something, and that was enough, which is still a lot, but again, I automate so much that it was, like, absolutely worth it.
Glasp: I see. Yeah. Is it for is it common for VC or, like, founders to use Airtable and manage everything, automate everything? Or all the
Yohei: I do not know if it is necessarily common, but it is also not uncommon. I know a lot of people use Airtable and automations. And I know that Airtable has recently done a big push in AI. I have not played around with those new capabilities yet, but it seems like they are making it easier so you can kind of build stuff on top and use AI with Airtable. So and I should probably spend some time playing around because I do like like the interface.
Glasp: I see. And how about, like, outreach managers or salespeople? Is it cool to use a table?
Yohei: I do not think so. I think for sales, there is a lot of functionality that comes with sales-specific CRMs that are very nice.
Glasp: Oh, okay. Mhmm.
Yohei: Right? That, like, unless you really liked managing it, again, it just, everything from, like, enrichment and all that. I do feel like if you are if you are doing sales, a lot of the existing CRMs have a lot of functionality that is, that is very useful. Especially, like, also email integration, right, these days, like, a lot of the newer CRMs. I mean, starting with Affinity, even Ateofolk will automatically connect to your email and your calendar. I think that kind of stuff is really helpful.
Glasp: Do you use any AI tools? I mean, I know you use a Replit or something. Yeah. Those are the AI tools. But do you use ChatGPT or AnswerPeople for daily, like, an email? I Yeah. You know, define definition or something.
Yohei: I use ChatGPT the most, followed by I use Claude, occasionally, but ChatGPT probably, like, seventy, thirty. ChatGPT Claude. And then for fun, I also use Suno a lot, which is a music generator, but I do it mostly with my kids.
Glasp: What is the use case of ChatGPT besides, like, coding? Do you ask them to create a post or not? Type a check?
Yohei: Or I use it a lot in my personal life. You know, I will use it to identify you. I was in California. We were at a park, and somebody said, Hey. I think there are some, like, poisonous plants over there. And so I used ChatGPT. I took a whole bunch of pictures, and I, like, had it identify all the plants for me. So I have I was like, oh, this is a safe area. You can play with these plants.
Glasp: Nice.
Yohei: When we were on a road trip, I used it to identify trees so I could teach my kids about, like, the difference between different furs and pines.
Glasp: Yes. Also, in that sense, what is the best use case of AI for growing kids? I mean, for kids or students?
Yohei: I mean, I think there is a lot of opportunity in leveraging AI for education. I do not, I mean, I think we are kind of early days of seeing the experiments. I have not really had my kids play with any too many GenAI tools. We use, like, image generators and music, music generators for fun. We do it as a group activity. For me, that is more about just getting them comfortable with AI. I have had my oldest start using Synthesia, which I really like. I do not, I think that was from before GenAI, but but it is it is an incredible kind of math, science, like, visual tool. We do use technology for education, but nothing, nothing too much from GenAI yet. But I suspect we will we will start seeing more interesting stuff soon.
Glasp: But do you recommend that parents use generative AI for kids? Because it is controversial, some of Eskard's problems, I would say.
Yohei: Oh, having them use it? I do not know if I think that is a that is personal decision on when your kids should use different technologies. I think people have different philosophies. I am not one to necessarily push mine. We tend to be on the, I guess, slower side, probably in terms of kind of letting them jump in and use technology. But again, they are, I mean, I do it all the time. So they are exposed to it. They are familiar with it. You know, they can play with it with guidance. And again, you know, my youngest is still three. So he is happy outside with a shovel and dirt.
Glasp: Yeah. Yeah. So, regarding the future and vision, do you have someone in mind you always try to look up to? I mean, who do you respect? Who do you want to become in the future?
Yohei: No. I would not say there is one person. Actually, funny. I think I talked about this during, in my tent, but I have this little note, a stack of note cards. And it is pretty decent-sized stack now, where one side of the card says somebody I have met or somebody that has inspired me, or it could be friends. It could be TV show characters. And then on the other side is, like, one lesson that I learned from them. And so I have this, like it is more, it is probably two, three hundred note cards. It has friends from elementary school, from middle school, like, all the people who have inspired me to, like, be who I am and, like, what I want to what I want to take from them. So I would not say it is one person, but it is just a blend of all the people I have met that have inspired me is what I aspire to be. But I think I think there is no one specific person that comes to mind.
Glasp: But could you list some names if you can?
Yohei: I mean, I mean, everybody from
Glasp: Everybody. Okay.
Yohei: I mean, there is there is hundreds of people, but, I mean, on the more famous people, like, you know, Richard Branson, I think there is, you know, I think there is something about his story of of starting a record shop with friends and just kind of letting, you know, growing that to a massive empire. I think that is a I think the way he did it seems very unplanned. Right? I think he just saw opportunities and jumped around, so that that is really stood out to, and resonated. I am a big fan of Jack Sparrow. He has got a little card there. Right? His carefreeness is something I aspire to learn from. Again, I would not want to be Jack Sparrow. I do not necessarily want to be Richard Benson, but, again, you can see how you scale this out to hundreds of people who have inspired me.
Glasp: Yeah. Thanks. Yeah. And so, since you are constantly experimenting with new tools, ideas, and what current trends in AI or tech are you most optimistic or skeptical about?
Yohei: It is a good question. I am generally pretty optimistic about, like, just the opportunity for AI to make businesses efficient. I do think there will be challenges as we start incorporating these more and more, especially the more you know, AI agent stuff. And also kind of consumer AI, I am not skeptical, but I do think it is important that we are aware of how and what AI tools we are using, especially for kids, for example, and how that impacts us. I mean, I think social media, if you go back to social media, there are obviously a lot of negative things about social media and the way it is impacting society. But, also, there are a lot of positives, right, in terms of giving people a voice that they did not have a voice before. Like, we have literally seen revolutions happen, like, on social media. And so there are pros and cons of social media, but I think that conversation about what is good and bad about technology is really important, and I think it is just as important, if not more important, for AI, just given how powerful it can be.
Glasp: You already shared so many lessons and insights, but do you have any advice for, like, aspiring founders or builders if they're okay? People who are about to start a new project or company. Sorry. Yeah.
Yohei: No. No. It means that already starting a company is challenging. Right? It is like, if you look at the most successful VCs, and you look at the statistics, right, even the best VCs, like, 60% of those companies do not actually return capital. Like, being a founder is very challenging. But, of course, it is also very rewarding to pick up people you know, pick people you want to help, and pick problems you want to solve. So, I mean, if I were to give one advice, I think talking to your customers is probably the biggest one. Right? I think that is probably the one that is most universal in terms of right? Like, learn about the people you are trying to help. And if you do that enough, like, you will gain insight that is unique to you. Because if you talk to the same 100 people today versus two years from now, they are going to have different problems potentially. So just talk to your customers today, and that gives you unique insight and builds something around that.
Glasp: Did they leverage the AI tools?
Yohei: Oh, absolutely. You should be leveraging AI for sure.
Glasp: Thank you. So, yeah, this is the last question, by the way. So since Glasp is a platform where people share what they are leading, learning, and we want to ask this question to you. So what impact or legacy do you want to or hope to be behind for future generations?
Yohei: That is such a good question. Oh, I would love to, you know, I think, I have a lot of fun doing anything I do, and I would like to help more people realize that they can also have fun doing whatever they do.
Glasp: Thank you. So, yeah, thank you so much for joining today, and we learned a lot, and that was really fun.
Yohei: Thanks. For having me.
