Last month I gave lecture at Globis University on what it takes to build an AI startup today.
It’s no longer early days for AI, and most founders don’t have the connections and resources that drive toady’s multi-billion dollar seed rounds. However, as I detail, they still have several paths to success.
After the lecture I am joined on stage for a panel discussion by Reiji Yamanaka, the managing director of the Kibo Impact Investment Fund, and Kelvin Song, the program director of the Globis MBA program.
It’s a fascinating discussion, and I think you’ll enjoy it.
Podcast: Play in new window | Download | Embed
Subscribe: Apple Podcasts | RSS | More
Transcript
Welcome to Disrupting Japan, straight talk from Japan’s most innovative founders and VCs.
I’m Tim Romero, and thanks for joining me.
I have a special in-between episode for you today. A few weeks back at Globis University, I gave a lecture to aspiring founders on the best way to start a generative AI startup right now in this time of intense AI competition and funding levels.
I cover the different AI business models, promising application spaces, and how to know if you’ve got an AI startup idea with a good chance of success.
Now, the first 30 minutes of this episode is the lecture itself, and then I’m joined on stage by Reiji Yamanaka, the managing director of the Kibo Impact Investment Fund, and Kelvin Song, the program director of the Globis MBA program. And we dive even deeper into these ideas and also talk about how generative AI is likely to affect us all.
I hope you enjoy it. So let’s get right to the presentation.
Presentation
Today we’re going to talk about how to build a generative AI startup and some important things to keep in mind if you actually decide to do that. Now, before I tell you what we’re going to cover, I want to kind of tell you what we are explicitly not going to cover. So first, we’re not going to talk about the transformative nature of AI in general, the explosive growth of the market. There’s already way too much chatter about that, and I assume if you’re even thinking about starting an AI startup, you already know it. Second, I’m not going to offer general advice about starting and growing a startups, although this is a topic that’s very close to my heart. I want to focus on what can add the most value to you in this particular seminar. If you want to talk about general start advice, talk to me later. I’ll point you in the right direction or ask questions afterwards or during the panel discussion.
We’ll begin today by talking about four common exit and growth strategies. This is a bit unusual. I don’t normally recommend that seed or pre-seed companies focus too much on exit strategy, but these are not normal times. With generative AI, you need to plan your end game from the very beginning. We’ll spend the bulk of our time talking about actually building your AI startup. We’ll cover some key strategic considerations, and also talk about a few of the most promising targets for AI disruption. Does that sound good? Well, before we get to it, why should you listen to me? And that’s a totally reasonable question. So, I’ve been in Japan for, wow, over 30 years now. Currently a partner at Jira Ventures. It’s $300 million corporate venture capital firm that invests in green energy, next generation energy, generation technologies. But in my time here, I’ve started four of my own startups I’ve sold two, bankrupted two. So, 50 50, not too bad as far as startups go.
I’ve done a lot of angel investing. I’ve taught entrepreneurship and corporate innovation at New York University’s, Tokyo Campus. I’ve brought foreign startups into Japan as a country manager. I was tapped by TEPCO to come in and help them spin up TEPCO Ventures. I left TEPCO to run Google for Startups, Japan, swearing I would never go back to energy CVC. After four years at Google, I decided to go back to energy CVC because right now what’s happening in energy is just fantastically exciting. Oh, and I also run a podcast called Disrupting Japan, where we sit down and we talk with Japanese startup founders and VCs, not so much about their specific company or portfolio, but what it’s like to be an innovator in a culture that really prizes, conformity and what problems these startups are trying to solve. So, you know, please like, and subscribe and all that. Okay, let’s get into it. Let’s talk about your exit strategy and the possible business models that stem from it.
Now, as I mentioned before, this is not the usual way of doing things. At seed stage, founders don’t normally need to think too deeply about the exit strategy. If you build a rapidly scaling business that adds value to your customers, M&As IPOs, both paths remain open to you and it’ll eventually become clear to you which is your best option. It’s not something you normally try to optimize for, not at seed anyway. However, these are not normal times, and if you’re building a generative AI startup, your exit strategy will greatly influence how you run your business. So, the four exit and growth models in increasing order of complexity are the Peter Pan, riding the hype, rapid customer growth and sustainable revenue growth. Now, we’re going to move really quickly through the Peter Pan and riding the hype, since they don’t really require growth at all. And honestly, only sustainable revenue growth needs a path to profitability. So we’ll be spending by far the most time on that one because sustainable revenue growth, it’s the most flexible, it’s the most complex, and in most cases, it’ll give you the greatest chance for success.
Now, as a founder, you need to understand your options and consciously make a choice of what your goal is. And on the other hand, if you’re thinking of joining a startup, you need to make sure that you’re a hundred percent aligned with the strategy that startups actually executing. Not just the one they’re saying they’re executing. Now, before we get into the details, I need to point out that, that even if you fully embrace the potential of AI, and it’s transformative nature, the VC market has never seen anything like this level of hype and focus. We’re already way beyond what we saw during the dotcom area. So, just how hot is the AI market right now? Well, it’s stupidly hot right now. Between 60 and 70% of all VC investment is flowing into AI startups. In fact, if you remove the AI investments, VC investment is actually down sharply. So yeah, these are not normal times. So, let’s talk about exit and growth strategies.
First, the Peter Pan, also known as I’ll Never Grow Up. Now this is not a serious strategy, but sadly, it’s the one that most startups are executing either consciously or unconsciously. They raise some pre-seed, maybe some seed money. They play with some cool AI tools, they ship a me too product, and then they disappear in a year or two when they can’t raise a Series A. Now I include this for two reasons. One, because like a surprising number of founders seem to be fine doing this. Some people just seem to want to have a founder experience and run what I’ve heard called a starter startup, which I mean, if that’s what you want to do, fine. But it seems like kind of a waste of everyone’s time. We are here at the beginning of what might be one of the most important transformative shifts in technology. And man, it makes sense to swing for the fences. Don’t waste your time chasing small dreams on this. And two, I include the Peter Pan model for those that are thinking of joining an AI startup. If you’re thinking of joining a team that is not clearly executing one of these other strategies, they’re probably never going to grow up and you’re probably wasting your time with them. Okay, let’s move on to more profitable strategies.
Our second strategy is to ride the hype. Now, I mean, all startups need to do this to some extent. We see hype driven business plans in every hot startup sector, but I mean, AI has everything dialed up to 11. Riding the hype is a pure play. This is pure marketing. Your product is your stock. It’s a very good play for the right kind of team. So, what do I mean? We saw safe super intelligence raise a pre-seed round at a $5 billion valuation. They got a nice step up to a $30 billion valuation at seed. Thinking machines just raised 12 billion. Open AI acquired Johnny Ivey’s IO for 6.5 billion on the promise that they would eventually build something cool. Now, these are obviously really big numbers, but if you haven’t worked in VC or startups for a while, it might not be apparent just how bonkers these numbers really are. So, when investors invest in seed startups, they’re generally looking for a path to a hundred times return. They’re not expecting a hundred times return, but they need a path to it. Now, even assuming these are somehow special, they’re only looking for 50 times return or 30 times return. This is implying a trillion dollar exit, which would make it not only the biggest startup exit in all history, not only twice as big as the biggest startup exit in all history, but twice as big as the next five biggest startup exits in history combined. And that’s not, keep in mind, this is not a bet on the technology. This is a bet on one particular company. So, yeah, thing things are things are a little frothy.
Now, these valuations are based on the star power of the founders, and that’s important to note that none of these companies have an actual product or even a prototype yet. So, shipping a product is actually a very dangerous step for hype driven companies because that’s when people are able to compare the hype and the reality, and very often the reality just doesn’t live up to the hype. But anyway, if you are extremely well known in the industry and you can line up an A-list VC to give you that credibility with other investors and to help you exit when the time comes, this could work for you, ride that hype as far as it’ll take you. Generally however, this is an avenue that’s not open to most of us. Riding the hype is important to understand, and we’re going to come back to it later.
Okay, now we’re getting to the good stuff. Our third strategy is rapid customer growth. Profitability be damned. Simply put, this involves focusing all of your efforts into acquiring a lot of paying customers as quickly as possible without worrying too much about being profitable in the long term. Now, this strategy can be extremely successful, but you need to keep in mind that not all customers are created equal. You need a particular type of customer to make this work. So, let’s look at the on, again, off again, acquisition of Windsurf by Open AI for an example. So, why was Open AI willing to pay $3 billion for this company? And why didn’t Google want the company preferring to pay 2.5 billion or 2.4 billion to license the tech and then just hire away the executive team and the top technical team? So, if you don’t know, windsurf is a coding assistant. It uses AI to help write, edit, and test code. Now, windsurf is actually number two in the market. Cursor has higher growth, higher revenues, and a much larger market share. But Windsurf was an attractive option for Open AI. So, what did Open AI want with windsurf? The platform itself has some value, of course, but Open AI wanted the data from the users, not the users themselves.
So, Open AI has 400 million active users, Windsurf, hundred thousand or so surrounding error. The real value Windsurf offered Open AI is the integration with the developer’s workflows and the rich training data that that gave them access to. They could see not only the prompts in the generated code, but they could see how that code get gets edited, what gets thrown away, how it it’s refined. That information is invaluable. Now, Microsoft already has this with a copilot, but Open AI design. Google has access to the workflow of a large number of developers, but they seemingly don’t have the AI integration and toolings in place to take advantage of it. So it makes sense for them to hire away the team. Now, cognition is now in the mix offering to buy the remainder, what’s left of windsurf? I’m unclear exactly what they plan to do with it. In fact, it’s kind of funny. This particular example I’ve had to rewrite three times in the last four or five days because this story just keeps changing so quickly. So, rapid customer growth only works for the right kind of customers. It won’t matter how many people download your app, use your chat bot. But if it gives you visibility into real world workflow, or if it’s something that enables access to a new type of training data, that has tremendous value.
Fourth and finally, of sustainable revenue growth. Now, fundamentally, this strategy is just building a regular startup that happens to be based on AI. And for the rest of the presentation will be talking mostly about executing this model. Now, clearly there’s a lot of overlap between the rapid customer growth and sustainable revenue growth models. There’s really only two key differences between the two of them. So for sustainable revenue growth, first you eventually you need a path to long-term profitability. And second, you don’t really need to have access to new types of training data for foundation models. But let’s talk a little about growth. When we talk about growth for an AI startup, we’re talking about massive growth and not just the top 1% of gen AI startups that grab the headlines and that everyone knows. So, these graphs come from research done, yeah, again, by Anderson Horowitz. And after the first year, the median B2B generative AI startup is doing about $2 million in ARR. The median B2C is doing about four. And if you look at the top quartile, this is about 5 million and $9 million respectively in ARR. Now, this is a lot of growth, but it’s what you need to be shooting for. Your competition is growing at this level, you need to do better. It’s also important to notice that the gap between the average and the top quartile is increasing. So, it’s looking like there’s going to be a lot of winner take all markets in AI.
Alright, we’ve gone over the different growth and exit models. It’s time to build your AI startup. So, we’re going to talk about four important strategic considerations and those are the thin wrappers, whether you want to be a generalist or a specialist. How to plan for very rapid foundation model improvements and the importance of getting customers using it. And then we’ll talk about some promising targets for AI disruption. So, let’s talk about thin wrappers. Now, thin wrappers are relatively simple front end that makes using AI more accessible for a specific use case. A wrapper takes user input, transforms it a little bit, calls an API passes that output back to the user. There are thin wrappers supporting social media, post generators, HR chat bots, study tutors, email responders, therapy bots, I mean, just a whole host of applications. Now, the term thin wrapper has been used dismissively because a thin wrapper puts you in a very vulnerable position.
You’ve got no real IP, nothing stops the foundation companies from moving sideways into your niche and pushing you out. And furthermore, since thin wrappers tend to be quick and easy to build, you’re going to have a lot of competition no matter what you’re doing with them. But I’ve changed my thinking about thin wrappers recently. For all their downsides, they can be extremely useful. And there’s a couple of reasons for this. First, the balance of power has actually shifted a little bit. The performance differences between the various foundation models is not as large as it used to be, and it seems to be getting smaller and you’re switching costs from one model to the next are extremely small. There is no lock-in across different models. And second, even if you don’t own any valuable IP yet, you own a customer relationship. And that can be extremely valuable if you use it correctly, both for the workflow integration that we talked about before and for the horizontal integrations we’re going to talk about a little later.
And finally, the large AI companies are not aggressively moving into verticals. And when they do so, they do so through acquisitions rather than building products. So open AI is far more likely to acquire your startup than it is to build a product that competes with it. So yeah, thin wrappers are a valid strategy, but there’s one critical thing you need to keep in mind. It’s that these are an entry strategy. You’re going to need to fatten up your wrapper really quickly. And in very general terms, you can make your wrapper taller or you can make it wider. You can make a wrapper taller by adding functions that better help users accomplish the task at hand. This can be things like additional processing of the API output additional editing tools or other kinds of transformations. Alternatively, or additionally, you can make your wrapper wider by integrating horizontally into the team workflow. So is your customer feedback tool, is it accessing support databases and online forums to see how you’ve interacted with this customer in the past? Can approval requirements be integrated into the generation tool? That kind of thing.
The next thing you want to think about is do you want to be a generalist or do you want to be a specialist? And there’s no right answer to this at all. You can make a great deal of money, either as a generalist solving a small problem for a billion customers or a specialist solving a really big problem for a few hundred customers. And in the world of AI, like everyone is talking about potential users and the numbers of hundreds of millions or billions, it’s easy to forget this. And yeah, if you’re going for rapid customer growth, it’s better to go a little wide. However, if your goal is to sustainable revenue growth, you’ve got a lot more flexibility. So, for example, I’m in the energy industry. If you have 50 transmission system operators who are using your solution, you are basically the industry standard. If you’ve got a hundred airlines using your software, you are considered a force in that industry. So, while there’s no general answer to the generalist versus specialist question, there’s usually going to be a correct answer for your particular product and your particular team. Generally, I advise if you can specialize, you should specialize. So for example, if you’ve got deep insight into customer needs or existing relationships, focusing on that market will give you an advantage. The same is true as if you can integrate into industry specific, but poorly documented APIs.
Now, keep in mind that specialist markets can also be huge. So for example, coding assistance is general, but test driven development in Java, Python and go is specialized. So, I don’t know, conversion from COBOL and FORTRAN to modern languages and finance specialized, but absolutely huge. The key to make specialization work is having a real edge and, and a real insight into the market needs and not just a story. And what we’ll talk about this a bit more when we go through the potential targets for AI disruption. In other cases, you want to be a generalist, of course, there’s not really a clear line between generalists and specialists. It’s all shades of gray. If you want to be pedantic, you could argue that anything other than the foundation models themselves is a specialization. But generalists have a lot more potential users and they have the potential to reach scale very quickly. But in practice, this is hard to pull off unless you’re early to the market. So, my two pieces of advice to budding generalists out there is first to make sure you’ve got a clear advantage in the market, either in the product itself or in your go-to-market strategy. And if you’re defining your advantage using words like better or easy to use, you probably don’t really have an advantage and you should give it a lot more thought. My second piece of advice is to always have an answer to what comes next. So specialists start when they start to get traction, they usually have a pretty clear path on deeper integration. Accelerated growth, the path for generalists tends to be a little less clear. So you’ll need to get a lot of thought in advance. It’s just like you need to plan on fattening up your thin wrapper. You need to have a clear plan on how you’re going to expand your generalist product once it starts to get traction.
Third, let’s talk about what happens when foundation models improve. So, one of the best ways I found to know if an AI startup has a sustainable business model is to consider what would happen if the foundation model performance suddenly became twice as good. Would such a change make your product? And by your product, I mean the part of it that you own, the part on this side of the API. Would it make your customers willing to pay more for your product or less? So ideally, you want generative AI to be the weak link in your value chain. So, that means that your customers value your particular workflow, your industry knowledge, your task level integrations, whatever. That’s what they value. When the foundation models improve, they will value your product more. Now, on the other hand, if your value adds in automatically refining user prompts or small improvements to the API output, it’s quite likely that improvements to the model are going to make your product less important. You’ll be adding less value to your customers. And by the way, later we’ll talk about how this highlights opportunities and hybrid models that combine more proven predictable AI approaches with generative AI.
So if you’re trying to build a sustainable business, you also need to plan for the possibility of a doubling or tripling of your API costs. Now this may seem like an odd thing to say the whole industry is constantly talking about how inference costs are coming down and they’re coming down fast. And yeah, that’s true. They are coming down, but current prices are heavily subsidized by VC money and by fierce competition. So it’s okay for these companies to lose money now, but it’s very, very likely that prices are going to increase as soon as growth begins to stall or when the market begins to consolidate around a clear winner. So, make sure you’ve got a plan to respond and make sure you’re adding enough value that you can absorb that kind of a price increase.
My final piece of advice is to get real customers using and actually paying for your offering as soon as you possibly can. Now, obviously this is really important advice for all startups, but it’s especially critical for generative AI startups for two reasons. First, VCs are drowning in AI generated pitch decks from AI startups claiming to be disrupting some industry or another. So VCs have to develop quick filters to get through this. Some look at the founding team, some look at where the introductions came from, but everybody pays attention to paying customers. So, if you’ve got paying customers at seed, you’re already ahead of 95% of your competition.
Second, customer feedback is more critical for generative AI startups because the demos and prototypes can be just so jaw droppingly impressive. There is a very real risk that you’ll trick not only your clients, but that you’ll trick yourself into thinking that you have solved a problem that you really have not solved. Now, maybe you have but you’re not going to be sure until your customers are actually using your product in production. AI is still new buyers haven’t yet had time to calibrate their demo to reality expectations yet. So until you have customers that are paying you to solve their problems, you’ll never be sure. Now, of course, once customers start paying you and then they continue paying you month after month and they start recommending you to other people, well then you might really be writing a rocket ship.
Okay, let’s talk about five promising targets for generative AI disruption. Now, this is not even a remotely exhaustive list, it’s just to give us all a few ideas for some discussions and maybe some brainstorming later. And the fact that these targets are so obvious to me might well mean that the best startup idea for you is something different. Something that you’ve got some special insight on that isn’t obvious to the rest of the world. So first, number one, legal tech. I have no doubt that over the next 10 years, AI is going to transform the legal industry just as fundamentally as the spreadsheet transformed the accounting industry in the eighties. And for those of you who are too young to remember that, which looks like most of the room, there used to be a thing called bookkeepers and their job was just to add up columns of numbers and the big accounting firms. You went in, you worked as a bookkeeper for a couple of years, and then you became an accountant. Spreadsheets just eliminated that whole class of profession. It also eliminated the career ladder. It also eliminated the billing model for these companies. So the big accounting companies had to reinvent themselves as consulting companies.
Legal faces, the same problem today, these big partnerships, you go in as a junior associate, you spend three years doing contract review and pulling documents and stuff that AI can do really well, it’s going to destroy the whole career path and it’s going to cause a lot of legal firms to figure out how they can reinvent themselves. See laws and contracts they play into LLM strengths. They’re tech space, they’re highly contextual, they’re filled with specialized jargon. And the labor costs to work with them are exceptionally high. Consumer facing legal tech is going to transform the way all of us interact with the law. In fact, I think in the future, AI could well be the primary interface normal citizens use to interact with the law. However, selling to law firms is not easy. They’ve got unique infrastructure and workflow and they do not tend to be early adopters. It’s also important to keep in mind that their existing vendors are not standing still. Companies like Thomson Reuters who’ve been in legal tech for decades are actively adding AI features to their offerings. So, make sure you’re bringing something new.
Number two, sales optimizations. The generative AI is already being used to support human sales staff to optimize online sales. And even in fully autonomous agents that can sell on the internet, this can be a great target because sales is metrics driven and tends to be very open to new technology. Keep in mind however, that this openness is a two-edged sword. On one hand it makes it easier to onboard your first few customers and prove that you can deliver. On the other hand, there are thousands of competitors out there who are making exactly the same claims you are. And so if you do not deliver and deliver very fast, you’re going to quickly be tossed aside. Two quick pieces of advice for those thinking of developing gen AI sales tools. First, if you don’t already have a background in performance marketing, you might be surprised at the complexity of the industry and the amount of AI that’s already in widespread use. This sector is fiercely competitive, so make sure you understand the state of the art before diving into deeply. Second, please don’t be an asshole. There’s already so many dark patterns and abusive practices and online sales don’t, don’t make things worse. I mean, gen AI, there’s so much potential to do genuine good and make a tremendous amount of money doing it that you don’t need to do something like that. Be on the good side. Build something you’d be proud to tell your mom about.
Number three, construction and physical project development. There’s a lot of great and highly specialized generative AI application we rolled out in the space. If you want to build anything, and in many cases, if you want to tear down anything, you need permits and you need to ensure you comply with various regulations. And now that sounds painfully obvious, but the challenge is it’s very hard to know how many different agencies are regulating you, who they are, and what regulations actually apply to your project. It all depends on what you’re building, where you’re building it. And believe it or not, sometimes when you are building it, there’s no simple way to find out what those regulations are. A large project will likely have to comply with regulations from multiple agencies and multiple levels of government, both local and national AI can be an absolutely fantastic tool for simplifying and speeding up this process.
Number four, hybrid AI solutions. So AI’s already in widespread use across many industries. It’s well trusted and has been for over a decade longer in many industries. So traditional AI already dominates predictive maintenance, demand forecasting, quality control, energy management, process optimization anomaly detection, all kinds of classification and sorting from medical scans to inventory management. In many of those cases, however, innovation slowed and the problem often is in the interface, either into or out of the system. It’s a communication problem. In some cases, generative AI can be the answer. This target’s not open for everyone. But if you understand the strengths and limitations of how traditional AI is being used for a given application, the potential for meaningful innovation here is absolutely huge. And finally, horizontal integration, or as I like to call it, death to SaaS. So because of the SaaS boom over the last 20 years, many companies have information that’s spread across dozens of different and poorly integrated SaaS products. And it’s not for lack of trying proper integrations next to impossible because there’s so many of these systems and they change frequently. So, gen AI has a potential to provide universal interfaces that can reach across multiple SaaS pro products and enterprise platforms. Now, it’s clearly not the right tool for every job. You don’t want to build an accounting or a compliance system on top of generative AI. It’s not meant to be a system of record. However, it’s potential to provide quick overviews and insights and areas from project management to customer satisfaction to product quality is absolutely incredible. So AI will likely replace many of these SaaS products entirely in the coming years. However, since the database vendors and the big data companies are already rolling out generative AI tools here, you’re probably going to want to specialize. If you understand the workflow and the needs of a specific industry, there is a lot of potential in being a SaaS killer.
Now, some of the programmers in the audience might have noticed something missing from my list here. What about Pure Tech place? If you’ve developed a way to improve a foundation model, speed up training or inference, what if you’ve come up with a better way to do fine tuning, wouldn’t that be a good path to be acquired by one of the big AI companies? Well, maybe, but probably not. Now, I am by no means an expert on the technical aspects of generative AI. So, if you are an expert and you believe that you have a game changing tech and other recognized experts who are not part of the project, have told you that you have game changing tech, then yeah, go for it. But from what I’ve seen, when you dig into them, most of these technology acquisitions, they turn out to either be products that have proven themselves through rapid customer growth or startups founded by famous people who are riding the hype.
So, it’s not enough just to build great tech. You need to have the market validate your tech and your startup and validate the fact that it is in fact, great. So, I want to thank you for coming out tonight. I’m looking forward to the panel discussion and taking questions later. I’m easy to find, connect on LinkedIn, check out Disrupting Japan if you get a chance. And thanks very much.
Pannel Discussion
Reiji: Okay, thank you. Thank you everyone for joining. And I’d like to thank online participants. There are about 140 online participants here. So thank you for coming here and thank you very much Tim, it is our privilege to have you here in Globis. Thank you.
Tim: Thanks for inviting me. I always enjoy doing these.
Reiji: And thank you for giving me the stacker of Disrupting Japan. I’m a big fan of these programs. Thank you.
Tim: Thank you. Thanks for listening.
Reiji: Kelvin, thank you. Would you like to introduce yourself quickly? I know that you have a deep technology background.
Kelvin: Yeah, it is really nice to see everyone here. Some of my students over here as well, and probably online as well. My name is Kelvin. I’m from Singapore, but I’ve been living in Japan for over the past 17 years. Now I’m the program director for the English MBA programs in Globis. I’m also leading the course development in AI and related subjects.
Reiji: Thank you. Tim. Your presentation was so insightful. Thank you very much. I’d like to start from this question. How can we enjoy the path for sustainable revenue growth? I think many of you are interested in studying up. How can future entrepreneurs define the good area where they can enjoy a sustainable revenue growth?
Tim: Yeah, I think well another thing, I think that over the next five years it’s going to be hard to identify an industry that is not influenced by AI. It’s going to just become like the internet where it’s just a tool that everybody uses by default. So, I think the key to identifying a market, and I talked a lot about them in the examples I gave, but in general, you’re looking for something that you don’t need precise answers because it can’t be trusted to give precise answers. And that involves synthesizing a huge amount of information. So that’s why legal tech is good. That’s why a lot of applications in marketing and sentiment analysis is really good. But it’s fundamentally, when you’re starting any company, you’re looking at solving a problem. It might turn out that once you start designing your AI startup, when you’re talking to customers, you might find out that the AI’s actually providing a very small part of the value. It turns out to be one tool you use.
Reiji: I see. So, you need to focus on the problem solving for customers.
Tim: Yeah. What is the problem you’re solving? How are you making people’s lives better?
Reiji: I see. Kevin, I’ll ask the same question to you. What is the key to sustainable revenue growth? You are teaching the course of the data science and AI and I’m curious about your insight about this.
Kelvin: Yeah, first of all, I’d like to touch on what Tim measured about preciseness. Because generative AI is essentially not precise. What you get from generative AI will always change. If you have something that you need precision, then you probably have to go back to the traditional way of doing software engineering, which is traditional what we call expert systems, if else programming. So, that is why when Tim mentioned about the hybrid AI solutions, combining the traditional way of doing it, AI versus predictive AI, injective AI, I see a lot of potential in combining all these three areas. And when it comes to thinking about AI, not all the AI solutions are created equal. And what I mean is that depending on the solution, it requires different level of accuracy and precision. For example, if you’re creating something like a recommendation engine, I’m talking about predictive AI in this case. If you’re thinking about recommendation engine, it does not need to have a high level of accuracy. As long as you can recommend something that people tend to use or tend to like in 10% of the time, it’s already providing some value. But if you’re thinking about things like autonomous driving, things that are mission critical, the level of accuracy or precision has to be very, very, very high. So, I think some of these things are the things that you should be considering as you think of different solutions for different problems.
Reiji: Okay, thank you. I’d like to move on to the next question. Many of you are interested in starting up that may be related to AI or how can AI help to startups in general, not only AI focused startups, but startups in general. How can AI help their management?
Kelvin: No, my background is actually from software engineering. So, looking at how AI is doing, Tim mentioned a lot about coding assistant, vibe coding is the key word right now. I feel very threatened because of my background as software engineer. But at the same time, I also see a lot of opportunities. And nowadays, anyone would be able to create apps very, very easily by just using natural language. So in the past, you probably need to take tens of hours, hundreds of hours, months, or years to create the apps. Now you can do it in a matter of minutes or even hours. It’s really amazing. It’s really amazing. And you do not need to have a strong technical background in order to create apps. So, what that means is that in order to create a startup, you probably need to do, MVP, minimum viable products. So, that is prototyping. You can easily do it using vibe coding. Very easy to create prototypes. But on the other hand, after you have created the prototype, if you want to make it into a production quality product, I think my personal view is that you still need to have engineering skills because if you’re just using vibe coding, you do not know what the AI is creating. It can be very dangerous in terms of security. One example that Tim mentioned was cursor. So, he mentioned Cursor as the number one share in vibe coding. Just one week ago, for those of you who are following the news. One week ago, there’s was a problem with cursor in which some users downloaded an extension unwillingly and unknowingly. And that extension turns out to be a Trojan horse that can actually invade into the developer’s environment at quite a huge chunk of, I think $500 million of crypto was being stolen from the developers. So, this is how dangerous vibe coding is because if you do not understand what is going on with the code that is generated by the AI, it can be very, very dangerous. Alright. So yeah, it’s a long answer to a question, but the point is it’s very easy to do prototyping.
Tim: No, I agree with that. I think that that’s a very good general rule. I spent a number of years doing software development myself. AI tools without significant tooling around them anyway, they tend to be very good at rough approximations. They’re incredible at building prototypes and demos. They’re incredible at doing like first pass market research, trying to figure out who’s spoken on things, what the state of the art is. But two things that I kind of always keep in mind about AI and kind of nurture my inner skeptic on this. So first is, I mean, I use perplexity. I use chat GPT extensively daily. And when I ask at things I don’t know much about, I’m always amazed at how much it knows. But when I ask, it’s about things that I know a great deal about. It’s far less impressive and it makes mistakes. So, it’s a first pass. And the other thing, the reason I’m skeptical about some of the claims on the coding productivity is, I think it was Microsoft announced there was like 25% of all new code is AI generated. And I’m sure that number will stand up in court. But when you look at data available from people who aren’t trying to sell you AI, you don’t see like open source projects having their backlogs completed. You don’t have these bug. And so out there where you can see it, you don’t see evidence that this is a transformative tool. Well, it is a transformative tool, but it might be transformative in the same way that IDEs were transformative. The IDE gave developers five X productivity and gen AI could do that, but you need to be skeptical of the output. So anything that you need kind of a rough, quick take on AI is amazing for.
Reiji: Thank you. Thank you. So, we are very hopeful about AI, but we also need to be very careful about how to use it. What is a key skill for future entrepreneurs in taking advantage of AI while avoiding the pitfall?
Tim: Don’t believe your own hype. That is the most dangerous thing that can happen in any industry. And like, when I’m talking to founders, sometimes it’s hard to know whether they’re just pitching or whether they’re the true believers. I mean, you never want to lie in a presentation. You can bullshit, but don’t lie. There’s a fine difference between the two. I mean, but you need to just keep in mind that it’s not magic. Occasionally ask chat GPT or philanthropic or your tool of choice, ask it about something you know a lot about. And that’s always really sobering. You have to keep reminding yourself what these limitations are, just because even when — it’s like an optical illusion for the mind. So, any type of optical illusion, you can know logically what it is. But you can’t help but see it. Our brains are wired to it. We can’t outthink ourselves. And I think large language models kind of push the same buttons where even if like, we know logically these are the limitations, it’s really hard. You can’t help but be impressed. So, you have to consciously check and double check that instinct.
Reiji: Thank you. We needed some kind of critical thinking for the innovation. Kelvin, same question to you.
Kelvin: Yeah. Reiji, you mention critical thinking, which I think is very, very important because if you go into understanding how large language models work, it is basically an order complete technology. So, it predicts the next word and then the next word, and then the next word based on the words that you give to the large language models. So what that means is that if you give it something that is general, it will give you something that’s general as well. It basically works using conditional probability. So depending on the conditions that you give, this is the output that you’re going to get. So in order to get a good output, you have to provide a good question. So, how do you provide a good question? It depends on critical thinking and also depends on the domain knowledge. A person who has a lot of domain knowledge, so Tim mentioned that if you are an expert in a particular subject, you can actually get better output of what you get from generative AI compared to just normal person using generative AI. So yeah, so understanding the questions and also have the ability to identify what a good answer is from the generative AI. So, it goes back to the same skills. You need to have to have critical thinking. You need to have the domain knowledge. If not, you will not be able to even identify, whether this is true, this is false, or this is bad from the AI.
Reiji: Is that something that I can learn if I take your class?
Kelvin: This is the reason why we teach in Globis. I think, if you want to get better answers about marketing, you need to understand marketing, same thing. So, in my perspective, you can say that I’m biased because I’m the MBA director, but I still think that MBA is very, very important.
Reiji: Yeah. Thank you. Now, Kelvin, I’m sure that you have se several questions to Tim. Would you like to?
Kelvin: Yeah, Tim. I’m curious because you’ve been living in Japan for the past 30 years, and so I’m curious, what do you see? How do you see the landscape of AI startups in Japan and also in investments in Japan?
Tim: So, I think Japan is pretty far behind in AI right now. I mean, not only in terms of dollar investing, but I’m not so much worried about that particular metric, but I see practical applications in Japan being rolled out much, much more slowly. And it’s a shame, I’m not sure why, I wish I had a happier answer to give you.
Reiji: I’m puzzled why that is. Do you think that the investment amount is not enough? Or entrepreneurs?
Tim: Well, I think it really is. I mean, anytime there’s a new technology, there is a cluster of innovation. There is a group of people who’s just trying to make things work and coming up with new ideas and pushing the boundaries. And so we saw this, I think the reason that American technology companies dominate this particular wave of startup innovation is because smartphones and cloud computing originated in like San Francisco and Washington. There was clusters of people just trying to do different things and experimenting with them. And they were way ahead of everyone else. China’s done the same thing with a lot of small batch manufacturing. It’s a lot of people like pushing the envelope, trying different crazy things. Everyone’s got two friends they can call up and say, well, how would you solve this problem? So, I think fundamentally, it’s not something that like, well, it just needs more money, or we need more AI courses in universities. I think that right now, just that’s where the cluster is, that’s where the innovation is. And anyone, if you’re in Italy and want to start an AI company, it’s highly likely you’re going to move to San Francisco because that’s where all the cool stuff is happening right now. So, I think that’s just the fundamental reason.
Reiji: Do you see any strategy for Japanese startups to be differentiated from America and differentiated from China as well?
Tim: For AI?
Reiji: For AI in the United States.
Tim: Yeah. Solve problems, solve problems. I mean, there are areas in Japan that have unique problems or clusters of innovation in Japan. Japan leads the world in like industrial robotics. There’s a lot of factory automation that is Japan’s way ahead of everyone else. So, there are those clusters here. They’re just not AI clusters per see, but that’s fine. So, you can apply AI anywhere you find this like cluster of innovation where you can exchange ideas with creative people who are like pushing the boundaries. So, play to the strengths.
Reiji: I’m interested in that approach as an impact investor. I’m interested in solving social problems. And many social problems in Japan are related to communication problem, as you mentioned, and AI can potentially solve it. For example, AI can help consumers reach medical experts. AI can help consumers or citizens to reach the government information. All of them are suffering from labor shortage, and they’re not very friendly to work with many citizens, but AI can help the interface. So focus on the deep problems and solving it by AI. I’m interested in that area.
Tim: I think that is a good area for AI, especially like I mentioned permitting and legal tech. So, I think that eventually we’ll see AI models like licensed to answer certain legal questions within a limited scope. And then citizens could just say, is this rent contract standard? What’s not normal about this? Is it okay to sign it? And it would say, well, these two clauses aren’t normal. Or I think that is an incredible use of AI and one that that will happen eventually. I think it’s deprioritized now, not because it’s not socially important, but because it’s not as profitable as many of these other applications. But yeah, I think that type of communication and making society run better, I think there’s huge potential for it.
Kelvin: Yeah. Probably this is also a question for Reiji because you talked about social impact investment because Japan is the number one aging population in the world. And all the other countries are going to have aging population like Japan. So, if we talk about aging population, there will be a lot of opportunities in healthcare, I suppose. And at the same time, it’s not only aging population, but we got declining population. Every hour we are losing 100 people in Japan every hour. So, there’s more than one person every minute. This how serious this problem is. This is probably a very nice thought, but I thought that, if we can solve the problem in Japan, maybe we can expand this kind of problem for the world as well, because other countries are going to face the same problem. So, I just want to get your thought on this.
Tim: No, I agree. I think that it’s a huge opportunity for Japan, and I’ve been kind of watching that space for a while. No one’s really kind of cracked the code yet. And I think part of the issue with like using technology in startups to address some of the problems with aging and elderly is there’s a certain amount of altruism required in solving these problems that doesn’t always align with the motivations of venture capital. So, sometimes it’s hard to navigate that. But as a whole, I think you’re spot on. Japan is a decade ahead of Europe and 20 years ahead of China, and the whole world will be facing similar problems. So, it is a tremendous opportunity to figure it out and to get things right and to drive real innovations. And I think large language models could really have a big role to play in that because so much of it is communication. I don’t think that AI will ever like, replace human companionship or anything like that. But I think, and I’ve seen this on Reddit, when people are talking about it, people form para-social relationships with AI, but I think it is equivalent to like pets. So, if you’ve got a dog, you know it doesn’t really have human emotions really, but we all kind of pretend it does, and we know we’re pretending but it works. And so I think there’s like, that level of companionship is something AI could do.
Reiji: Thank you. I’m a big fan of Taroimo and I really want to invest, if there is a Taroimo for senior citizens in Japan. Physical AI that can be very friendly and that that can solve various problems for our senior citizens. That Taroimo could connect the senior citizens with various our internet world, EC and so on. So, that’s my dream. I’m always interested in solving serious social problems, but I also want to talk a little bit about entertainment. Entertainment, animation, manga, those are one of the strengths of Japanese industries. So, how do you see the opportunity between AI and entertainment industry? Are there any opportunities around there?
Tim: Well, what I’ve seen so far has not been pleasant. No, it’s going to transform it. I wrote an article last year on advice for musicians who are competing against AI generated music. And I won’t repeat it here, but I think we’re definitely in a transition period where there is a tremendous amount of AI content being generated. It’s very poor quality, but even when quality gets better it won’t get better in a certain way. AI is incredibly good at producing the structure and the affect of a particular work, but it can’t really create something moving or new or novel. The problem is that for artists, before you can create good art, you’ve got to spend a few years creating bad art. It takes a long time to learn these skills. And right now AI is already better than most of the bad artists. So, 10 years ago you play your crappy little song or show your crappy little drawing, and it was still way better than most people could do and like you would get encouragement for doing that. And you could be paid for doing that. So, with that career ladder going away, just like the bookkeepers having the ladder locked out from them and the lawyers, these junior associates who are about to lose that career ladder, I’m actually kind of worried that the artists of the future don’t have that ladder. They don’t have a way to spend years being validated for producing bad art before they’re good enough to produce good art. And it’s something I’m actually I’m kind of concerned about.
Kelvin: I’m not an expert in manga, in anime, in Japanese art, but I’m a huge fan of them. And I know that a lot of anime, they are handwritten by animators. They’re hand drawn by animators mainly. So maybe AI can help to help us produce more of this Japanese content, which can explain to the world. Yeah, this is my take on that.
Reiji: Thank you. We’ll soon move on to the question, not answer. But before that, I have a final question to you. How will AI change the dynamics among big corporations, investors and startups? You know, this triangle, big companies, startups, and investors?
Kelvin: As I mentioned, I think previously in order to create something going back to engineering, which I came from, in order to create something in order manufacturing something, you need to have the manpower, you need to have the ability to do this. But I see a lot of things can be now outsourced to AI. I think manufacturing and engineering would be much easier. So, with that expect, I do see that probably the smaller companies would have the ability and the speed and the resource in terms of AI to be able to challenge bigger companies. This is my simple take on this.
Reiji: I see startup companies will have more power to support and enable the big corporations.
Kelvin: Yeah. And on the other hand, I’m not against big operations, but because of the way we standardize things, because of the way we make things efficient, there are so much standard protocols, it’s quite difficult to change because when we want to change from humans to using AI, if we actually have to change the entire workflow. And it’s not an easy switch from the current workflow to a new type of workflow. I see.
Reiji: I’ll ask the same question to you, Tim. You have observed this triangle in Japan for more than 20 years. How do you see the change?
Tim: I think it’s a really interesting question. I think that the impact on startups will be in terms of like, the power dynamics will be smallest on startups. Startups they’re always kind of nimble. They adapt new technology, they’re always running as lean as possible. The big companies, I think it’s really going to hurt. Manufacturing industrial companies they’ve been automating and using, let’s call it traditional AI for decades now. So I think the impact might actually be smaller there, but in white collar work, all of this work of like writing reports for managers and managers reading reports and synthesizing this to give to a project manager, to give to a portfolio, that’s all going to kind of go away. And that’s going to be hard because corporations are run by people and the whole reason people join a company is that they’re in charge of other people. And the number of people that you’re managing informs your salary. And so I think it’s going to be hugely transformative towards large corporations. And I can’t predict how exactly. I think it will give VCs and capitalists in general a great deal, more power in the short term because these efficiency gains will be concentrated on those who make those investments, those who have the capital to deploy into it long term. These kind of things tend to even out. So yeah, like kind of broad brushes. That’s kind of how I see it shaking out over the next 10 to 15 years.
Reiji: I see. How do you see the bargaining power balance between investors and entrepreneurs? I’ve heard the hypothesis that the startup companies don’t need to hire so many engineers, therefore they don’t need to raise so much amount of money for starting up. So, they may enjoy a higher bargaining power against their investors.
Tim: Maybe, it’s complicated. So, sometimes if you raise a billion dollars, you need to justify raising a billion dollars. So, some of these companies, I think are just hiring people whether they needed them or not because you’ve got to spend the money. Because if you don’t spend the money, you can’t raise the next round so your valuation can’t go up and the VCs don’t make money. So there’s a little bit of that game going on. But I think more fundamentally, it’s just this ebb and flow. It goes in cycles. So when things are bubbly and exciting, startups have a lot of power. They get good valuations, they get multiple term sheets, and when things aren’t going so well, the power dynamic flips and it’s just cyclical. The types of startups and the exact formation of technical versus sales changes every cycle and what they’re working on. But that just seems to be, it’s like the tides.
Reiji: We need to understand the hype cycle and when the hype is high, we should enjoy and take advantage of it while building the solid revenue stream. And getting ready for the downturn.
Tim: But it is worth noting that usually the biggest, most successful companies are ones that are founded in the downturns. And I think it’s just because it’s a filter. The only companies that can get funded at those times tend to be very strong, very innovative. They’re entering the market without a lot of competition. So yeah, I’ve started companies at both times. It’s a cyclical industry.
Reiji: Okay, thank you. Now let’s move on to the question and answer. Shall we start from taking Christian from the, the floor? Yes.
Student: Hi, my name’s Cole. I work in human resources, I’ve done for the last 20 years, and I’m endlessly fascinated by the future of work. I was particularly interested when you mentioned about the evolution of accounting through spreadsheets and bookkeeping and disappearing and potentially what could happen with the legal industry. Particularly with the sort of early career jobs and tasks and things. AI natives are coming at us quite fast. We used to talk about digital natives. Now, AI natives, particularly kids are using it at school now that they don’t know any different. One of the things we see with a lot of AI stuff now is people relying on their olden day skills to analyze and validate what AI is putting out. But I think the next generation coming through may not have those olden day skills. They were born with a chat GPT account, for example. Can you give a little bit more sort of commentary about what you could see potentially happening in the future when the AI natives potentially dominate a lot of the proficient?
Tim: I think you’re absolutely spot on there. I think it is a society-wide problem. The same thing with the bookkeepers being replaced. But the same thing is like, someone has to make bad art for a while before they make good art. Learning any skill, any job requires a few years of not being particularly good at it. And currently the education system across the world, is based on this idea that, the world does not need another undergraduate essay on Hamlet. It’s the only people who will ever read it are those that are paid to read that essay. But the value is that to create that essay, the student has to read certain tasks and analyze certain tasks and exercising that muscle. And Chat GPT is just like short circuited that completely. And there’s no putting that genie back in the bottle, it’s just not going to happen. So, our approach to education needs to change in such a way that doesn’t suddenly increase the load on our already overworked educators by like five X, and can take account for this new paradigm. And the sad thing is, the advances in pedagogy and in educational techniques happen on the span of decades where the advances in technology are happening in the span of months. So, I agree. I think this is going to be a problem society’s going to wrestle with in about oh, 10 years.
Reiji: Thank you. Let’s move on to the next question.
Student: It’s funny that you asked your question because at the moment I’m actually studying how AI can be used in education and how we can help support students and alleviate some of the pressures that they have, especially in Japan. This country is quite notorious for having a lot of pressures on their student to study and to do well, and to go to cram school afterwards as well. So they’re kind of spending their whole life I guess childhood, studying, studying, studying. And one of the things that a lot of theorists and well-known people in pedagogy and learning theories suggests is that your emotional state and your motivation plays a fundamental role in how you absorb learning and how you build your skills. And I wanted to ask, what would you think the potentials for AI in education is? And in what ways can it be used to kind of maybe add to or revolutionize the current methods we have to help students?
Reiji: Kelvin?
Kelvin: Yeah. For education? I think the problem with education right now is that, we are trying to educate everyone in the same way. And that should not be the case because every one of us, we have different pace and we have different ability to absorb the materials that we have. And when it comes to learning business, we all come from different industries. So, we might be more associated or related with one particular topic than the other one. But the way we teach right now is that we teach the same thing to everyone. And fundamentally that is because we cannot assign enough man-to-man lecturers for every single person. But using AI, we probably can. So, I see one possible future in education is personalization, so personal tutors. So depending on your industry, maybe the AI will be able to teach you in a different way, depending on your pace. And maybe even when we talk about case studies in business, maybe the AI can also generate a case on the fly that is related to your industry that is related to your background and to be able to let you absorb a particular concept in a better way. So I think personalization is one way to go in education industry.
Tim: No, that’s interesting. And I think maybe what you’re saying about the ability to provide more one-on-one, not just education, but the evaluation. So, instead of writing another undergraduate essay on Hamlet, you would have an AI that would be talking to the student about what they learned about Hamlet and what they thought about this. And that scales where a professor couldn’t talk to all 50 students, an AI could. I’m sure there’s ways you could like, make sure that that student is monitored and not stick them in a Faraday cage or something. And so they’re not using AI to answer the questions, but I think we’re going to have to use it. Absolutely.
Reiji: Okay. Online questions.
Kelvin: Okay. So, here’s like a kind of general question, but ties into AI as well from Michael. In your experience, is there a lot of motivation for entrepreneurship by young people in Japan? He says, I don’t currently see it at my tech university here in Japan. It seems that big companies dominate and bureaucracy strangles motivation. How can persistent young graduates who want to start their entrepreneurial journey and AI, how do they do that in Japan?
Tim: Well, I think it depends whether your answer, if you’re asking for the point of view of one ambitious student or from kind of a structural societal point of view. So, if we’ve got an ambitious student on the line, I’d say that most universities have a good deal of support for entrepreneurs. There’s even larger programs like one run by the University of Tokyo that’s nationwide and can provide access to mentors, to financing. So there is a tremendous amount of help and mentoring available in Japan right now. I think it’s so much better than it ever has been. Now on a societal basis, yeah, there’s nowhere near as much of a passion for entrepreneurship in Japan as we see in say, San Francisco. But it is, it is there. I work with high school students sometimes, and there’s a tremendous amount of energy and drive and passion and it’s there. And the thing we need to keep in mind, I think, for the sake of argument, let’s say San Francisco is the most innovative place on the planet. How many people are actually doing anything innovative? Like 1%, half a percent, not really. Even people working at innovative companies aren’t really doing anything innovative. And so in Japan, if we can just nurture that half a percent of innovators and connect them to the right resources and more important, connect them to each other, that is utterly transformative. I mean, I’ve been here over 30 years, so part of this is seeing the incredible progress that’s happened in that time since I’ve started my first company. But part of it’s also practical, like what I’m seeing now. So, I’m very optimistic for Japan in that.
Student: Thank you for the opportunity to ask this question. So, me and Kelvin and some of us, we went to China recently, and we saw that I mean there were great advancements in that country. And my question is from this perspective that if you choose to build something with AI and with the growing divide that we see. Right now, there was a report that says that around 60-70% of the world’s top researchers are currently in China. And I mean the top AI researchers. So, the kind of technology divide that is happening, and if a single country gets that kind of superiority and when you build AI startups, do you think that there is going to be a risk in terms of data like where the data goes? Will people be comfortable using a service that you build in Japan, but you use the technology from China or some other country?
Tim: Yes, but it’s probably not as big as you think it is. A lot of countries, Japan and the US included, many buyers are very sensitive to where their data is stored. The EU has laws about where data can be stored and transferred. But when you’re designing your startup, when you’re thinking about what you want to do, where the foundation models are coming from, where the researchers are, unless you’re into like security or defense, it really shouldn’t really factor deeply into your thinking and your planning.
Kelvin: Okay, this is the last one. I’m so sad because there’s so many good questions coming in from online, but we only have time for one more. So Robin asked my question, since many AI startups now use the same open source foundations, how do you differentiate between a company with a genuine IP moat and one that is just a thin wrapper? What specific evidence of defensibility in “patent” do we need to look for beyond the application code itself?
Tim: Patents won’t give you anything. Well, I mean, they’ll give you some, but they won’t tell you if you’ve got defensibility. I’d say, like we talked about a bit in the presentation, look at what happens when the models get better. Does that make your hold on the customer stronger or weaker? So, if your offering gets stronger, when the models get better, you’ve got something. If the model getting better, threatens your position, then you don’t have a moat and you’re in danger. So, I think that’s probably, if you want to ask like one question to focus on, I’d focus on that one.
Kelvin: Okay. Thank you so much. So that will bring our Q&A session to the end. Round of applause for all the panelists.
Outtro
And we are back.
I hope you enjoyed that. And I hope that if you’re starting an AI startup right now, you’ll remember that the key to success is not how exciting generative AI is, but what you and your team can do with it that others cannot.
And immersing yourselves in a community of like-minded innovators is absolutely one of the best ways to figure that out. So we’ll be back in two weeks with a full episode. A
nd if you enjoy Disrupting Japan, please rate and review on your podcast platform of choice. Or better yet, you know, just tell people that you enjoy the podcast. It means a lot. But most of all, thanks for listening and thank you for letting people interested in Japanese startups and VCs know about the show.
I’m Tim Romero and thanks for listening to Disrupting Japan.

Enjoyed this informative episode with your nicely structured and well-articulated talk. I think you should do solo episodes, in which you just share your current views of the Japanese startup landscape, give tips for entrepreneurs, comment on new laws that affect them etc. Your podcast doesn’t have to be only guests and recorded panels. Sometimes you can monopolise the mic and let your audience benefit from your extensive experience and perspectives on how Japan is being -or should be- disrupted. ;)
Hi Fab,
Thanks for listening. I love doing the solo shows, and they tend to be some of my most popular.
https://www.disruptingjapan.com/category/podcast/solo-shows/
It takes me long time to write those solo shows. I seem to average about one per year. I would love to do more.
Tim