--:-- Host (Steph): Voice is the way that humans have communicated with each other since really the day that communication was born. But voice has never really worked as an interface for technology.
--:-- Guest (Anish): Ten years ago, I probably would have called that experience magical.
--:-- Host (Steph): The exciting thing is that this first wave is already happening right now. It's not a theory, it's reality.
--:-- Guest (Alex): Anime is just able to combine many different types of player profiles because a lot of people like anime and there's a lot of different player types.
--:-- Host (Steph): An awesome user experience can raise switching costs, strengthen network effects, and empower a brand. You've just got to start building. Drone swarming. Reinventing early education with AI, plus AI companions going 3D, and games that may never end.
--:-- We asked investment partners across A16Z to preview one big idea that they believe will spur innovation in 2024. And from this exercise, our team compiled a list of 40-plus builder-worthy pursuits that you can find at a16z.com slash bigideas2024, or you can click the link in our show notes.
--:-- But today, we continue our three-part series as we look toward 2024 together, with our partners across all our verticals, from biohealth to games to American Dynamism and more. So in part one, we covered what it'll really take to democratize "miracle drugs" like GLP-1s, but also programmable medicines, Final Frontier, and AI interpretability.
--:-- On deck today, we discuss how AI and other tech trends are being applied to the world of consumer and games. For example, is voice finally having its moment? And will foundational models or UX win in the AI battleground? Plus, what might the games of the future look like? Listen in to find out.
--:-- As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com slash disclosures.
--:-- First up, after over a decade of heavy investment, are large language models finally the unlock that we needed for voice-first apps? And if so, what will this shifting interaction modality mean for both new and existing products?
--:-- Hi, I'm Anisha Charia. I'm a GP on our consumer investing team, looking at all things AI apps.
--:-- Anish: My big idea for this year is that voice will actually be a new, emerging way to interact with technology and at the heart of a new set of productivity apps. The theory here is that voice is the way that humans have communicated with each other since the day that communication was born. But voice has never really worked as an interface for technology.
--:-- I think one of the reasons for that is that when you interact with a product using voice, you have the unconscious expectations you would have of a human. You expect it to be flexible, cognitively sophisticated, and able to interact with you. Most systems that exist today, or have historically existed, are very inflexible. They're just logic trees. If you don't navigate them perfectly, they don't work. We've all experienced that with Alexa and past generation voice assistants.
--:-- The go-forward idea is that AI and large models really unlock humans' ability to interact with technology. The other thing that's very interesting for productivity apps is that you could imagine some incumbent productivity providers—Gmail for email, for example—will build a lot of the obvious AI features for those apps and experiences. But it's less obvious that they would prioritize something that's voice-first, because it requires rethinking the entire experience from the ground up.
--:-- I really think it's an interesting area because it has previously never worked, and now this technology allows it to work, and it's not something the incumbents are obviously going to build. That's what I expect to see, hope to see, and invest in in 2024.
--:-- Host (Steph): All right, Anish, I feel like that sounds really exciting, but voice has been pending for quite some time. You addressed this a little bit, but I actually went back. I looked up the Gartner hype cycle over the last decade. Speech recognition popped up in 2010, and that was actually in the plateau of productivity. Yet despite a lot of investment from a lot of big companies—Amazon, Google, Apple—it feels like we're still waiting. You addressed that maybe there's some new technology on the horizon, but what would you say has been the limiting factor thus far?
--:-- Anish: It's simply that the old technology architecture doesn't allow you to get to the level of fidelity you need. The last 10% is where 99% of the effort goes. This is true of a lot of AI problems built on the old architecture. Self-driving is a great example. True interactive voice is a great example. You can get to 90%, you just can't get to 99%. This new technology really lets you get to 99%.
--:-- You can see that if you've interacted with ChatGPT via voice, it's very compelling. If you used 11 Labs, their work is very compelling. It's clear that this now works, and what we needed was a technology shift, not a set of financial resources that incumbents have historically provided.
--:-- Host (Steph): You mentioned that existing apps aren't necessarily equipped to build these experiences. Could you elaborate on that and maybe think through what a voice-first app would even look like?
--:-- Anish: Existing apps are built with a set of existing workflows. Anytime you modify the workflow, you have to think about how that affects all existing users. I'm sure there are a lot of creative people at Gmail, but it's very hard for them to go to a dramatically different paradigm like a superhuman. Google tried this with Inbox and ended up shutting it down.
--:-- There's a set of consumer expectations and a huge audience that expects a product to work a certain way. To do something like a voice-first inbox, you need a completely different set of workflows. If Google decides to do that, they'll have to start from scratch along with everyone else. It's not obvious to me that an incumbent will build this.
--:-- As for what the experience would look like: a great example is executives I know who, when driving into work in the morning, phone their executive assistant and say, "Let's go through my inbox together. Let's talk about what seems like a priority, what's important but not urgent, what's clearly spam, what are internal versus external asks." They sort through it together so that when the executive arrives, they're ready to go. An experience like that would make enormous sense, except now it's available to everyone, not just executives with a dedicated assistant.
--:-- Host (Steph): You mentioned two specific areas you're looking at in 2024: companionship and productivity. What are you already seeing built there? Are builders creating apps that are live, working, and voice-first?
--:-- Anish: There's a lot of cool exploration happening in the voice space. One product I'm a big fan of is called Tab. Tab has gotten a lot of attention online. It's a pendant you wear that passively captures your conversations throughout the day, then allows you to interact with all that context, ask questions, ask it to summarize. It's a great example of someone thinking from a first-principles, voice-first perspective.
--:-- The second most interesting product, outside of something like 11 Labs, is actually just ChatGPT's voice module and voice interface. I encourage everyone to go take a walk and have a chat with ChatGPT. You'll be amazed at how quickly you forget you're talking to an AI and fall into the flow of a conversation. Between Tab, ChatGPT, and what the team at 11 Labs is doing, there's a lot of cool stuff to check out.
--:-- Host (Steph): As this tech gets better, I'd love to hear your perspective on the unexpected types of experiences that don't seem obvious right now. It's easy for people to imagine sending an email or starting a song with Alexa. But if voice becomes the primary modality for interacting with some applications, what might be on the horizon that's more significant than what we imagine with voice today?
--:-- Anish: It's trite to cite this, but I will anyway: the movie *Her*. That's it—a passive experience that has all the context you have, where voice is the primary modality of interaction, and you can fall back to a screen if needed. If we look back in 10 years and that's how we interact with technology, it won't feel like such a leap.
--:-- The important question is that currently, voice is not the primary modality. You have all these companies that rely on a website, an app, or a graphical user interface, which may become moot if we're moving in that direction.
--:-- Host (Steph): How should builders think about that potential paradigm if it really is voice-first? How do existing companies integrate this thoughtfully?
--:-- Anish: I don't know if it is moot. I think there's room for both. There's a lot more technology penetration to be had. If you're a developer working in an IDE, you're probably faster hands-on keyboard with a screen than with voice, at least for now. Maybe the answer is some combination of the two. There's a lot of room for existing products and existing modalities.
--:-- On the other hand, if you're a senior citizen struggling with technology, this unlocks every product for you. I think it's a market expansion, positive-sum story, not a zero-sum story. But I tend to always think that.
--:-- Host (Steph): I agree, it's positive-sum. But if you're an application without a voice-first interaction mode, should all builders be thinking about incorporating it? How would you think about that if you're an existing builder without this on your roadmap?
--:-- Anish: The customer and interaction pattern are so different that if I had an existing product, I would focus on playing from my place of advantage—non-voice workflows—rather than rethinking my product from the ground up. The answer may be in the middle, where some workflows are voice-enabled to extend the existing workflow. But again, I think this is a market expansion, net-new customer story, not someone taking existing customers.
--:-- Host (Steph): If founders are thinking about creating in this new environment, are there any challenges you'd call out from your experience with tools like 11 Labs? Where are we in this arc, and what can and can't be done?
--:-- Anish: You have to retrain people. They need to know how to interact with these systems. By default, their expectations will be very high. They'll think, "I should be able to interact with this with the same level of fidelity, responsiveness, and flexibility as a human." If that's not what you can offer, you'll need to train your customer to interact with you in the right way. There's an onboarding question.
--:-- There's also a social patterns question: where will it be socially acceptable to talk into thin air? People are already doing that with Bluetooth and AirPods, so maybe we'll all change quickly as a society. But there will be a period where early adopters feel a little crazy or look that way.
--:-- Host (Steph): It's funny you mention that. My husband's grandmother said the first time she heard someone talking on a cell phone, she thought they were crazy because they were doing it in the supermarket. She was like, "Why are they talking to themselves?" So maybe there's an element of that again.
--:-- To round things out, there seems to be a lot of opportunity on the table. We're talking about a potential change in interaction modality. If you're an entrepreneur, how would you even think about starting? What would you look at? Where would you begin?
--:-- Anish: Obviously, you can only begin at the beginning. It's impossible to infer what the end product will look like. You've just got to start building. I would try to build something useful and interesting and iterate from there.
--:-- The other thing that's interesting is that if you're too pointed at the beginning in terms of product design, it can constrain you. If you introduce the capability in an intelligent way, your customers will pull you in a direction. Some of the best product builders I've seen have been willing to offer the capability in a relatively neutral way and allow the customer set to pull them.
--:-- Host (Steph): Per your advice earlier, for people to play around with this tech themselves and use voice more.
--:-- Anish: 100%. Always. Just use the products. Just use the products.
--:-- Host (Steph): Awesome. This has been fascinating, and I cannot wait to use more voice in the technologies we use daily. Thank you so much for sharing this, Anish.
--:-- Anish: You got it. Thanks, Steph.
--:-- Host (Steph): Next up, we address the quickly evolving AI battleground. Since ChatGPT came out just one year ago, builders are constantly asking where moats will arise. A16Z partner Alex Immerman thinks the differentiator may actually lie in user experience, not the foundational models.
--:-- Alex: Hi, I'm Alex Immerman, a partner on our growth fund here at A16Z. My big idea for 2024 is the consumer AI battleground is going to move from the model to UX.
--:-- Overheard in 2023: "A minute not spent building the model is a minute wasted. Build the best model and the users will come." The most popular consumer AI companies thus far have been the producers of their own models, such as ChatGPT, Character, Bard, and MidJourney. Differentiation has come from being the best model in their domain. MidJourney with images, Character on entertainment, ChatGPT on overall text. UX has largely been determined by the fastest way to get the models in the hands of users.
--:-- But thanks to a combination of factors—the likelihood that chat shortages ease, the availability of most foundation models via API, and increasingly powerful open source models—the groundwork is there for breakout consumer apps to be built on someone else's model. In 2024, consumer AI apps will break out by delivering the best user experience around unique use cases, less on model performance alone.
--:-- I'm particularly excited for consumer AI apps that figure out how to include shared experiences in a multiplayer mode, aggregate multiple models into a single interface, or build more focused solutions where workflow and process drive value. LLMs can be a source of differentiation. Today, they might provide a first-mover advantage. But old-fashioned moats like network effects, high switching costs, scale, and brand will likely win long term.
--:-- Host (Steph): All right, Alex. Since last year's AI consumer boom, people are constantly asking where moats will arise. I feel like every podcast we do on the topic, I have to ask that question: where is the differentiator? A lot of people talked about models, but you're saying UX might be the moat. Why do you think that is? And what would you say to the pushback that design or UX can be copied easier than ever?
--:-- Alex: Over the history of technology paradigms, we often see technology progress and features first, then a movement toward prioritizing design and experience. I expect user experience to contribute toward the building of moats. An awesome user experience can raise switching costs, strengthen network effects, and empower a brand.
--:-- Think about the last paradigm shift with mobile. If we rewind to 2010, people were taking photos with their new powerful phones. There were many photo filtering apps—I remember Hipstamatic—and social media apps like Facebook. But there wasn't a great experience that bridged the two. Instagram came along. It brought a slick interface. It was simple to take photos, edit, and share photos. Ten years ago, I probably would have called that experience magical. My photos, my friends, they're all there. Suddenly, the likelihood of me switching to another application that doesn't have that information is lower. Why would I switch? The experience is so good that my friends and I use the application more, which in turn makes it better, strengthening the network effect.
--:-- In AI, we're starting to see consumers build really exciting applications on top of this new technology. I think they will leverage UX to establish similar moats to their predecessors.
--:-- Host (Steph): That makes sense. But some people might ask: if everyone has access to the same models as they become more accessible, how do new players really win in this environment?
--:-- Alex: It's important to point out that virtually everyone already has access to OpenAI or Anthropic. But there are real costs associated with using these models. Unless you're very well capitalized, it's difficult to run a free consumer app and afford the associated inference requirements. Many current AI consumer apps use a subscription or transaction model to cover costs. They are beholden to charging consumers.
--:-- What's changing now is that we expect the costs to access quality models—whether closed source or open source—to fall. Developers will be able to build more affordably for a broader set of consumers. It widens the top of funnels for these developers and presents a much broader opportunity.
--:-- One framework to consider is that models are just a new primitive to build applications on top of, the same way storage, geolocation, or payment rails are today. If you believe in the commoditization of models, you could argue that models are similar to ACH rails in financial services. A fintech company has to distinguish itself with user experience and workflows around ACH, just like consumer apps will need to with these models.
--:-- Host (Steph): So you're saying that many people think as this becomes more democratized, incumbents win because they have financial resources and existing users. But you're saying those economics will change. Another piece of pushback might be: how do you avoid becoming just a wrapper and being displaced by model providers like OpenAI? There's a meme on the internet that every time OpenAI has a demo day, a bunch of startups get wiped out. How factual is that, and how should builders differentiate through model versus experience?
--:-- Alex: A classic investor question for many years has been, "Won't Amazon do this?" Now it's quickly becoming, "Won't OpenAI do this?" Entrepreneurs should focus on where they are uniquely best positioned. That's not specific to AI, but in general for company building.
--:-- A recent historical parallel is cloud computing. Should entrepreneurs build their own infrastructure, or should they work with AWS, GCP, Azure? For most founders, they should work with a third party. If founders believe their team's core strength is building a truly differentiated model, sure, that's a worthy endeavor. But there are few companies with that caliber of research team today. It's true with Character AI, Ideogram—the list is short. For most companies, that's not going to be the case. But fortunately, building a defensible business does not have building a model as a prerequisite. There are other ways to make it difficult for your users to never want to leave your application.
--:-- Regarding wrapper companies: most wrappers today are just providing a user interface to call a model API. You really need to add value on top of the underlying model. My colleague Jamie Sullivan would say it's the difference between being one of the hundreds of compass apps in the App Store and Uber, using the geolocation of a smartphone to build an application that wasn't possible before and for which the underlying infrastructure is not a threat.
--:-- In the AI context, I'm a fan of Perplexity, an AI-powered research assistant. It works with third-party models, but it also offers many seemingly small features that add up to a compelling user experience: lower latency, providing sources, suggesting follow-up questions, giving multimodal responses. This results in a fun, different search experience that users can't find with just using a model.
--:-- Host (Steph): I love that Uber example. It really cements the idea that a model is not the starting point for many, and you can add value on top of it. If folks are convinced and say, "Okay, I don't have to create my own model, but I can create something of value with a differentiated user experience," you mentioned ideas like multiplayer modes and interfaces that aggregate several models. Could you elaborate? What are you seeing as examples of creative experiences—the equivalent of Uber with geolocation—that you've seen already, and what breakout use cases would you like to see in 2024?
--:-- Alex: I'm really excited about this. First example: Quora's AI product, Poe, covers both multiplayer modes and model aggregation. On the aggregation front, Poe brings together leading models like GPT, Claude, Bard for text, Stable Diffusion, DALL-E for image, and many others. If you want a central place to access the many models popping up each day, Poe is that web browser for AI.
--:-- But I can also use it with friends. When I ask a question of GPT-4 on Poe, I can share it with you, Steph, and you can continue the conversation in Poe. Looking ahead, these shared experiences will go from just text to being multimodal. Imagine, Steph, you and I are having this conversation, but we're having it with an AI as well. The AI is not just answering and listening to text, but also listening to our speech and responding with its own AI voice. Poe recently added voice input. We're going to have more realistic AI friends.
--:-- This should also translate to gaming. You and I are playing Fortnite or Roblox, and suddenly the non-player character, the NPC, who's listening in and watching our gameplay, starts having a unique, personalized conversation with my friends and me. I think that's going to be a lot of fun.
--:-- Host (Steph): It's going to be so fun. And what's also going to be fun is we can only dream up a fraction of these experiences. As all these builders go out and explore, we're going to see all types of things we never could have imagined today.
--:-- One other thing you mentioned in your big idea is the powerful nature of open source models. How key will that be to driving this change? To what extent do you feel open source may become the industry norm underpinning many of these products to come?
--:-- Alex: Clem, CEO at Hugging Face, forecasted a few days ago that in 2024, open source models will reach the level of the best closed source models. That may be aggressive for 2024, but they are headed toward parity. Several leading open source models like LLaMA and Mistral are already competitive with closed source equivalents. It's early to call how the open/closed source split will go, but the better open source models get, the more pressure it puts on closed source models to lower costs.
--:-- What I'm confident about is the cost to access models going down and them being broadly accessible. There is so much early activity happening on top of open source, which can be much more affordable and allow for free and broad consumer adoption. Two fun image applications are EverArt and Can of Soup, which productize fine-tuning to a specific style or people.
--:-- Host (Steph): Like we said, there's going to be a lot on the horizon, and open source really democratizes that. The more entrants, the better. As we think about how the product layer may be increasingly delineated from the infrastructure layer, can you think through some second-order effects? That could apply to existing or new companies.
--:-- Alex: As the application and infrastructure layers are split, it will make it much easier for new companies to be built. Founders can just build for applications and not have to worry as much about the underlying model as an input to their own company creation. This is akin to the mobile phone being introduced and making it so much easier for gaming companies to handle distribution via the App Store. If there's one less major feat to building a big company, it attracts that many more builders.
--:-- Beyond that, much like Uber and Snapchat could not have been predicted when the iPhone came out, we don't know what Gen AI will bring. But we're excited. It's going to be some fun, sometimes weird, hopefully always impactful new behaviors.
--:-- Host (Steph): "AI will save the world." What a place to end off. I think if we take one thing away: if it's true that you do not need to build a foundational model to participate, there are just so many more people that can participate. That's what we want—more entrants, more people exploring this new technology.
--:-- And what will widespread, economically sound AI mean for games? Could it mean highly personalized, real-time rendered games that actually never end?
--:-- Host (Steph): Well, A16Z partner Jonathan Lai thinks that may be the case. Hi everyone, I'm Jonathan Lai. I'm a general partner on A16Z Games. And my big idea for 2024 is AI-first games that never end. So in 2024, I believe we'll see the first cohort of AI-first games from creators that leverage large language models to enable novel game systems and mechanics.
--:-- And while much of the early discourse on generative AI in games is focused on how AI can make game creators more efficient, I actually believe the largest opportunity long term is on leveraging AI to reinvent the nature of the games themselves. So, creating never-ending games that engage and retain users for a very, very long time.
--:-- A couple of examples of this: generative agents powered by large language models can create incredibly lifelike companions and emergent social behavior, turbocharging games that heavily utilize non-player characters or NPCs. Personalized character builders and narrative systems will enable every player to have a fully unique and personalized playthrough of their favorite game, and in the game worlds themselves, will no longer be rendered, but actually generated at runtime using neural networks. A new player onboarding itself is poised to be reinvented.
--:-- And then my idea is that every game will potentially be redesigned around an AI co-pilot with the mantra of good alone, great with AI, and best for friends.
--:-- Host (Steph): All right, great. So Jonathan, you mentioned several opportunities or even you could say fundamental shifts in game development from never-ending worlds to convincing NPCs to personalized real-time rendering. I don't know which of these upgrades you are already seeing built perhaps and also which you most excited about.
--:-- Jonathan Lai: Yes, so we generally think of the AI revolution in games as happening in two waves. The first wave is building the same game but faster and higher quality with AI tools and infrastructure to empower game developers. But the exciting thing is that this first wave is already happening right now. It's not a theory, it's reality.
--:-- So we surveyed 230 game studios earlier this year, and it turns out that 87% of them are already using generative AI tools. And so some of the products that they named include ChatGPT, MidJourney, 11 Labs, KDEM, Adobe Firefly, GitHub Copilot, et cetera. And the majority of these studios rated the impact of these tools as transformational, which is pretty incredible.
--:-- The second wave, which we consider even more exciting, is building entirely new sets of AI-native games that are uniquely enabled by AI. And those are the examples I write about in my big idea for 2024. And so dynamic worlds that can be generated in real time based upon a player's actions, lifelike AI companions, never ending sort of choose your own adventure games.
--:-- We're still very early in seeing what those new AI native game experiences might be like. And so you've seen a couple of prototypes, some interesting sort of proof of concepts. But in general, the playing field is wide open and that's the exciting thing for the folks that do get it right. Like these are the ideas that can potentially create entirely new genres that captivate hundreds of millions of people. And that's what we're incredibly excited about looking out into 2024.
--:-- Host (Steph): Yeah, I think what you're picturing is really important because what you're saying is we're not just incrementally optimizing existing games. You're really describing a fully new game environment, fully new way of thinking about games. And I love this mantra that you included, which is good alone, great with AI, and best with friends. Can you elaborate on what that means and maybe how it shifts or has the potential to shift the way that people currently think about both building games and also playing games?
--:-- Jonathan Lai: So many of the biggest games in the world today, you think about Fortnite, Call of Duty, League of Legends, they utilize AI primarily for onboarding and practice. And so before you play against real human opponents, you might play against AI bots, and they enable you to practice, build confidence, learn all the controls, get your hand-eye coordination down, and set sort of a warm-up before you jump into the real game, quote-unquote.
--:-- I think one of the promises of generative agents is in elevating agents, like the AI agents, from a warm-up into the core game itself. And so one example of that that I find fascinating is this concept of an AI companion for every game.
--:-- So let's say at the beginning of a game, like you're starting Call of Duty, for example. You might customize your character. And let's say you also customize an AI companion that you can bring with you into the game. And this companion can play the entire game with you, right? Either it's on the same team or against you if needed. And it might comment on the experience. It understands the world around it. It keeps you company when your friends aren't there to play with you. It learns from all of your preferences over time and helps you out and generally tries its best to make sure that you have a great time.
--:-- And so such an agent wouldn't replace playing of humans, but it would certainly be better than playing by yourself. And so that's what I mean by designing games that are good alone. They can be played single player. They can also be played with an AI, and it's great with AI. But of course, it's best if you have human friends to play with. And those different modalities can be mixed and matched, depending on the game and depending on the time.
--:-- Host (Steph): Yes. What gives you the confidence, though, that these are the type of games when we're talking about the inclusion of AI becoming hyper personalized potentially? What gives you the confidence that those are the type of games that people want? And I'm asking that as it relates specifically to community and the fact that games really are these communal interactive environments where people have somewhat of a shared reality. And so if everything is hyper-personalized, I'm curious to know your perspective about how that shifts people's ability to build community. And you mentioned some of this is already being implemented. What are you seeing that gives you the indication that this hyper-personalization or inclusion of AI is something that people really want?
--:-- Jonathan Lai: So the interesting thing about games is that the games industry has actually embraced AI for decades. And so in a way, the latest expression of the technology of large language models is new, but the consumer behavior, the patterns around it are actually already existed. And so that's what gives me confidence that there's demand for some of the new types of gameplay that we talk about.
--:-- Take AI companions, for example. It might sound like a crazy idea at first blush, but so many iconic games already have AI companions that are pre-scripted, and they don't use large language models, but players have still developed deep attachments with them. A great example is the Halo franchise. There's an AI companion in that game called Cortana, and she talks to Master Chief, comments on what he's doing basically like as a chip that rides in his helmet, and she tells him where to go next when he's lost. She is there from the beginning to the end of almost every Halo game. And there are millions of people all over the world that adore Cortana, write fan fiction about Cortana and the Halo universe, to the point where I believe Microsoft has integrated Cortana into its core Windows search function.
--:-- And in another game, a favorite of mine, The Last of Us, which is a famous Naughty Dog franchise that was recently turned into an HBO show, you play as Joel, but then you have Ellie as an AI companion, and she follows you around for basically the entire game. She adapts to your behavior. If you're sneaking somewhere, she'll also sneak with you. If you're in a firefight, she helps you out in combat situations. And she has a fully fleshed out personality that you learn more about over the course of the game.
--:-- And so from that perspective, AI companions have already existed and are very, very popular. And so the idea of just turbocharging these with large language models to make them more lifelike and enable them to have a wider range of expression and behavior, I think is going to be very well accepted by players.
--:-- On hyper-personalization, which is the question that you were asking around storytelling and how that works with communities, I think I'm optimistic based on the fact that personalization already happens to a large degree in games. And so role-playing games like Baldur's Gate 3 have incredibly detailed character builders where you can customize everything from your skin color to the highlights in your hair to the tattoos on your face. And people spend hours customizing their avatars. There are entire YouTube channels dedicated to how to create the most lifelike or the most badass character for each of these games.
--:-- And let's not forget that a huge portion of the revenue in games like Fortnite and League of Legends comes from selling skins, which are customized virtual clothing for how your character might look in game. And of course, we've written about this before, but one of the oldest RPGs of all time is Dungeons and Dragons, which by definition is a personalized adventure that is told by a human narrator, a dungeon master, a game master, who basically crafts an improv adventure with you and your group of friends.
--:-- These are just two examples of personalization in games and an AI companion in games, but I'm optimistic that players have already embraced many of the concepts that generative AI is turbocharging. And so from that perspective, we're not inventing entirely new consumer behavior. We're just making it easier for players to do what they already want.
--:-- Host (Steph): Yeah, absolutely. And you mentioned a few you could say age-old games, Halo, Dungeons & Dragons. They've been around for quite some time. Do you see this new technology, or as you said, a technology that has been ingrained in games for also some time, do you see that really benefiting the incumbents who already have these large player bases? Or do you see this kind of fundamental shift where you're going to see completely new winners in this completely new era? How do you think about maybe some of the new entrants versus the incumbents and who really wins?
--:-- Jonathan Lai: So I think the incumbents, let's call them sort of the big AAA game developers and publishers, they are already going deep into the Wave 1 AI tools and infrastructure that I spoke about. And so I think almost all of them are using ChatGPT, MidJourney, Adobe Firefly, onto these software tools earlier. And I think they will benefit hugely from them as they will enable them to make higher quality games faster, get more shots on goal, and so on.
--:-- I am skeptical, however, that the largest developers will be the ones that create incredibly innovative AI-native games of the sort that we spoke about as the Wave 2 games. I know many reasons, but one of them is that if you're a large developer, your business revolves around IP. So you typically have a few key franchises, Halo, Dungeons and Dragons, World of Warcraft, Call of Duty. And your business revolves around launching sequels to those franchises every few years. And so you don't actually need to take that much risk in shipping these sequels. And in most cases, the players themselves actually aren't looking for something too different. Like the next Call of Duty title is like comfort food. You know exactly what you're going to get and that's why you enjoy it.
--:-- So I think the business model of these large incumbents doesn't support taking too much risk. So I think that's one reason. And second, I posit that the first generation of these AI native games might end up actually looking really silly to traditional game developers. And so they might not actually take them seriously at first.
--:-- And so take mobile games, for example, which was the last sort of major paradigm shift in games. The winners in mobile games were largely new companies, Supercell, King, Zynga, but not the PC console companies, right? It wasn't Electronic Arts or Activision or Microsoft. And I bet a large reason for this was that if you were an executive producer at Activision and you looked at a game like Candy Crush in the beginning, you probably thought, "Hey, this is really silly. The graphics are really bad, there's no multiplayer on this thing. It's a really simple puzzle game. So how can this possibly be good? How many people will play this thing?"
--:-- But what I think they underestimated is that with each new paradigm of games, a new set of players also comes into the ecosystem. So the folks that played Candy Crush were for the most part older women, and it was an entirely different demographic from the young men that played Call of Duty, for example. But the net effect was that it expanded the total pie for games, like the gaming time increased. And over the course of that, it created several valuable companies.
--:-- Another more recent example is take a product like Character AI, where you have millions of people that chat with their favorite sort of AI characters. I would guess that the majority of the users of Character AI don't describe it as a game, but you can squint and say, "Hey, this is a consumer experience that is just as engaging as playing a game," but it's probably not on the radar of any of the traditional game developers, the incumbents that we just talked about.
--:-- Host (Steph): That's such a good point. So I think in that same vein, I think the most innovative AI native experiences are more likely to be created by startups and teams that are just getting going and are bringing that fresh perspective, as opposed to coming at it with a lot of history and baggage around existing IP that they need to shepherd.
--:-- Jonathan Lai: Yeah, I mean, I think that's such a good point about just the things that are emerging that people would not even describe as a game. But if you really think about it, they are not so different fundamentally from games.
--:-- Host (Steph): Maybe super quickly, can we touch on the business model? Because as these things fundamentally take new shapes and forms, so too do the things that people are willing to pay for. But also when we think about AI, the cost structure is a little bit different too, right? If you're rendering things in real time, that has a real inherent cost to it. And so maybe you could just speak quickly to how you might see the business models evolving, just like we saw with mobile. And then also maybe things that developers should keep in mind if they are building these new games and are considering completely new business models?
--:-- Jonathan Lai: So one of the challenges is that games as a business model have trended towards two models. So one is free to play, where the majority of users, actually around 90 to 95% on average, play a game for free. And the minority, 5 to 10%, actually monetize. And so the free users entertain the monetizing users, and so they pay with their time, even if they don't spend money. And so some of the biggest games in the world, like League of Legends, like Fortnite, are free-to-play games.
--:-- And then the second business model is a premium game. And so a game that is just sold for 20, 30, 40, 50 bucks, you pay for that game up front, and then you never spend again, basically, after that.
--:-- The challenge is with an AI game, if you have a game that's running a large language model in the cloud, the inference costs of serving users dynamically generated worlds or text would probably make either of these models unprofitable right off the bat. Because you're monetizing the user up front for 50 or 60 bucks in a premium game sale, and then you're not getting any more revenue going forward. Or in a free-to-play model, 90% of the users are actually just a cost, because they're not actually spending any money in the game.
--:-- My guess is that companies like MidJourney or Character AI have been able to address this with a subscription model. And so basically every power user pays for their own inference costs, so to speak. And we could see that subscription model also become the default business model for AI games. And I think the gamers would be okay with it as long as they perceive that the value was there.
--:-- And as an interesting sort of side point, subscriptions have actually become more popular in games over the last five years with the rise of services like Xbox Game Pass, Apple Arcade, and even inside individual games like Fortnite, they now have season subscriptions like the Battle Pass, where you subscribe up front and it unlocks perks over the course of the season. So my guess is that subscriptions become more popular for AI games.
--:-- Host (Steph): Yeah, absolutely. One follow-up question there is if people are more linked to their gameplay, as in the time that they're spending in-game as it relates to inference, do you think that creates a negative feedback loop? Whereas right now you can choose to play a game and you can play it for quite some time and really get ingrained in it, and you could say obsessed or addicted to it without additional cost. Do you see an adverse relationship there where all of a sudden, if my inference is linearly tied to cost, that I may not be as likely to get as deeply involved in a game?
--:-- Jonathan Lai: Yeah, no, I think that's a great point that games will need to design around. I think the good news is that that same trade-off of time versus money also currently exists in free-to-play games. So if you recall the early days of Facebook games, take for example Farmville, like the Zynga games, they actually attempted to gate the amount of time that you spend in the game behind an energy meter. And so every day you would get 20 energy and every action in the game would cost two or three energy. And then when your energy was gone, you could either take a break and then come back and play again tomorrow. That's how the term "appointment gaming" came about. Or you could monetize and you could actually spend to refill your energy meter. And that enables you to keep going.
--:-- And so you can imagine a similar thing happening with AI games where you have some measure of the inference cost for players, and that's approximated by a virtual currency inside the game. And so I think it's just something that game developers will need to design around. But I think that paradigm of balancing time versus money already exists.
--:-- Host (Steph): Yeah, that's a great point. And you're bringing me back because I spent far too much time playing Farmville back in the day. But to close out, clearly there's opportunity on the table. And I think it really does feel like there is this new era, potentially, of games. And that's exciting. So with that opportunity, where would you start if you're a builder in the ecosystem? What would you be looking at in 2024?
--:-- Jonathan Lai: I would suggest that if you are excited about games, take your favorite game genre, whether it's a sports game, an MMO, even chess or poker or Dungeons and Dragons, and try to reimagine on first principles what it could look like if you were to introduce generative AI into the core gameplay loop. And then see if you can hack together a prototype or team up with friends to hack something together and then come on over and share it with us. We love playing crazy games and wild ideas.
--:-- And the reality of the games industry is that some of the best games got started by solo developers or as mods and passion projects that are built by small teams. Like Dota, PUBG, Counter-Strike, some of the most iconic games were built by very, very small teams. And I think an even more exciting thing is that anytime you have a new paradigm shift, like we talked about mobile early on, I think the playing field advantage goes to new teams, the folks that are coming in with a fresh perspective.
--:-- And my guess is that the AI game teams that get it right might actually end up looking very different from the traditional game developers of today. If you get it right, I think the potential is you can disrupt the $200 billion market in the fastest growing category of media today. So it's very exciting.
--:-- Host (Steph): Amazing. I agree. It's super exciting. Jonathan, thank you so much for going through this and excited to see what games are built in 2024.
--:-- Jonathan Lai: Likewise. Thank you.
--:-- Host (Steph): Speaking of games, data shows that anime is going mainstream. But what is it about the style that makes anime games so uniquely accessible? And as the games industry continues to grow, what can these not-so-niche games tell us about what's to come?
--:-- Robin: I'm Robin. I'm on the A16Z Games team. And my big idea for the year is on anime. So anime is a uniquely accessible art style for both kids and adults. The medium allows for many different story archetypes and player profiles. There are adventure components, romance hooks, social loops, many other things within anime games that lead to deep player engagement.
--:-- We're excited because things like Genshin set the new standard for a fully cross-platform performance game across multiple device types, GPUs, and frameworks. And there are now many passionate developers that are building new experiences for players everywhere. And so we believe that anime will be a medium that transcends both different games, genres, and art styles. And we're excited to invest across that ecosystem.
--:-- And we see this in the data, too, where anime has become one of the highest-grossing genres in the ecosystem when calculated by average revenue per user or ARPUs. For instance, Mahoyo generated over $3.8 billion for releases like Genshin Impact and Honkai Star Rail in 2022. And earlier this year, Nintendo launched Zelda Tears of the Kingdom, which is on track to be one of the highest-grossing games this year. And in 2024, I expect anime's momentum to continue forward too.
--:-- Host (Steph): All right, so Robin, I love this big idea. And I'm sure many of our listeners are familiar with anime, but for those who aren't or just to set a foundation, what actually constitutes anime and maybe how does it differ from other formats like manga or webtoons?
--:-- Robin: Anime is unique because it started from Japanese comics. And so it's defined by a few different types of art styles. So one is solid outlines, exaggerated features, blocked out colors and textures and lines, things like enlarged eyes and so forth. It's a little bit different than webtoons because webtoons originate from Korea. And so there are slight artistic differences such as webtoons tend to be a little bit more pastel and shaded. They also tend to have more rounded shapes instead of the hard lines like anime.
--:-- But there's no real hard and fast lines for what defines anime. So some people consider "Avatar: The Last Airbender," which is an American cartoon, or "RWBY," which is another American cartoon, anime too. And so I think the genre is pretty broad. I'd say it's more defined by its fan base, because most of the core anime fan base likes a wide variety of things across these different archetypes.
--:-- Host (Steph): Definitely. And you mentioned that anime specifically is able to incorporate different components, like adventure components or romance hooks or social loops, and that generates this really deep engagement. And we see this in the data. We'll get to the revenue in a second. But what do you think it is about anime's style that actually sets that genre apart and allows for some of those components and some of that engagement in the gameplay?
--:-- Robin: Yeah, so I think in general, different art styles are very targeted at different audiences. And so for instance, if you take something like Elden Ring, it's a very hardcore dark fantasy audience. Or if you take something like Fortnite or Roblox, those tend to be more cartoony or Lego-y and so it tends to skew way younger. And so you only have a certain sub-segment of experiences that's available because your art style is tailored to that specific experience.
--:-- Whereas with anime, many different types of people and many different age groups and many different demographics really like it. So you can make a casual game like a match-three game that has anime components, or you can make a romance game where people feel very deeply about the characters, or you can make an action game. And you can see that even with Genshin, right? It's both an action and romance dating game. And so many of these genres can basically combine.
--:-- And so I think in general, anime is just able to combine many different types of player profiles because a lot of people like anime and there are a lot of different player types that gravitate towards the medium. And there are a lot of people playing.
--:-- Host (Steph): I mean, some of the data that you shared in your big idea really surprised me. I knew anime was popping off, but some of these games are making tens of millions of dollars per month. That's not per year. And so how are these games monetizing and is the monetization strategy any different from other genres of games?
--:-- Robin: Yeah, so there are typical things like battle passes or IAP or progression loops that you can put in. And so they have standard monetization techniques that are pretty common in all mobile games. You buy some boosts or you buy the battle pass for the month and so forth.
--:-- The one thing that's distinguishing about anime games in particular, which tends to drive the high ARPU, is the gacha component. And so this evolves from Gashapon in Japan, where basically you used to put a hundred yen coin into the Gashapon machine and get a random character out. And so this is the same idea with gacha, where I spend $5, I roll the gacha and I try to get my favorite character, my favorite husband or waifu, essentially.
--:-- And so I think there are fair critiques that this is similar to gambling in certain ways. But basically, a lot of the best studios have tackled this by making it so that there are a lot of pity techniques and that earn rates and the drop rates for five-star characters are decently reasonable. And so you basically have a lot of different emergent behavior that comes from that too, because basically people post YouTube videos of drop openings and box openings and, "Hey, I got my favorite character." And so it makes for great content. Plus, it allows for people to have very high spend depth in the game.
--:-- And so that's why you see ARPUs as high as $30, $40, $50. Whereas in some other games, ARPU might be more in the $10 to $20 range. So it's potentially 2x to 3x what a regular game makes.
--:-- Host (Steph): Right. And just to clarify, gacha is those almost like vending machines in Japan where you insert money and then you get those round balls with something in it. Okay, just wanted to clarify that for anyone who isn't familiar.
--:-- It does feel like certain companies, I think Netflix has announced certain investments in anime. So large companies are, I guess you could say, waking up to this opportunity. But I'd love to just get your take on why it's felt like, maybe at least from the outside, people were sleeping on this trend and maybe still are in a way. Do you think that'll change? And what do you think maybe has not been capitalized on by the masses yet?
--:-- Robin: Yeah, I think in the past, when I was growing up, anime was definitely a weird otaku, nerdy thing that people didn't like to talk about. And so you had to figure out who also watches anime and only talk about it with your friends because you didn't want to come off as uncool or watching this weird Japanese cartoon thing.
--:-- And so I think anime has changed a lot and gone way more mainstream. So you see things like Jujutsu Kaisen or Attack on Titan or even the old Naruto, Bleach series, or One Piece getting really big these days. And I think there are just way more anime viewers. If you look at the stats, I think something like 50% of millennials and now like 70% of Gen Z watch anime. And so more and more young people are watching, converting to the medium.
--:-- And I think that's why people were sleeping on it before, because people were just like, "Oh, it's just a little side genre and hobby that some people like." But actually now it's something that a lot of people really enjoy because there are just many stories that you can tell through the medium.
--:-- And then the other thing I think people didn't realize is the fan base kind of keeps to itself. And so anime fans are extremely, extremely passionate. And you can just look at the numbers of how much anime fans spend on figurines and merch and games and so forth. But they keep it to themselves. And they spend it within their community rather than bringing it out to a wider community because I think there was still that stigma before. Whereas now it's changing.
--:-- It's sort of similar to gaming, right? Like when Fortnite got big and football players started doing Fortnite dances, then you know gaming is also going mainstream. I think the similar thing is kind of happening to anime now.
--:-- Host (Steph): Yeah, definitely. And I'm curious to hear your take on who you think could win or really dominates this genre as we move forward. There are obviously big giants like EA or Tencent that have been around forever. I think you mentioned Mahoyo, which has been around since 2011. So a little newer, but still around for a decade or so. But then I assume there's still room for new developers. How would you think about that? And also what lessons can developers learn from the success or the rise of anime?
--:-- Guest (Anish): Yeah, I think one, it's important to be a devout follower of the medium to be successful. And so I think one of the reasons Mihoyo was so successful is their tagline is literally "otaku save the world." It was started by a bunch of college kids who are just super passionate about the medium. There are very specific things about the genre that you have to get right.
--:-- So just, I think Mihoyo really gets right the specific character design of a lot of its characters. It gets right a lot of the archetypes of the medium. There's a little sister type of character. There's a lot of archetypes that you have to get right to make sure that you're doing right by the genre, essentially, and doing right by its fan base. And so I think Mihoyo was super successful at that.
--:-- And then they also built up their capabilities over time. So some of the first Honkai games did progressively better, but nothing at a billion-dollar scale. And eventually, they built up the technical capabilities to build Genshin, which is now a billion-dollar scale, but took several hundred million dollars to make. And so it takes a while to build up to that genre and that scale.
--:-- But I think what's also great about anime is it's not as expensive a medium to develop, because a lot of it isn't even about the gameplay. A lot of it's about the art and the story and so forth. And so you're not building something as deep as an MMO per se, which might cost you several hundred million dollars. You can build more casual mobile things that are cheaper to develop and cheaper to put out to market. And it's just an art style, right? And so you can put out many different types of things.
--:-- So I think to answer your question, I think both big players like Mihoyo will continue to generate new hits, such as Honkai Star Rail launching soon. But then also, I think there's a lot of opportunity for innovation. And we've been on a few companies in that space similar to so Pato Labs and also Odyssey Interactive are both building anime-style games. One is building an anime action RPG and the other one launched Mega Strikers, which is like Rocket Hockey meets League of Legends. And so we're pretty excited for both. We believe that both startup devs and large companies can have a lot of potential in the space.
--:-- Host (Steph): Just to close that out, the style itself can probably be enabled by some of the new technologies that are coming online or becoming more omnipresent. So AI, XR. What are your thoughts on how those new technologies impact the system and maybe help us invent the next new wave of anime?
--:-- Guest (Anish): Yeah, I think AI is definitely an obvious one for anime because it's a 2D art style and it's a very known art style. And so the AI is very coherent and can generate very good assets for the medium. And so you can look at things like Niji Journey, which is a fork of Midjourney and probably the most common anime 2D art asset generation right now. And so I'm pretty sure that many people will be generating both mods and fan art, but also people will be using this in a production setting to get anime games out for cheaper and getting them out at very high quality. So I think that will definitely happen for sure.
--:-- On XR, I mean, you can look at the success of Pokemon Go, for instance. I think XR is an interesting new medium and we'll have to see what kind of games come out of it. But for large brands, such as Pokemon, I think it's a very innovative use case for, "Hey, we go out in the wild, we collect Pokemon anywhere." And it's just like living your core fantasy that you had when you were a child. I'm not sure exactly what will come of XR because we don't know what we don't know. And I'm sure people will come up with way more inventive things than I can come up with. So I'm just excited to see how the platform develops and see what kind of new devs make cool new game loops on the platform.
--:-- Host (Steph): Totally. Well, this has been super interesting. I mean, it's incredible how many people are engaged with anime games. And then as we talked about earlier, the revenue is just shocking, like tens of millions per month for some of these games. So super excited to see how this trend evolves in 2024. And thanks for sharing your big idea.
--:-- Guest (Anish): Yeah, thanks for having me, Steph.
--:-- Host (Steph): I hope you enjoyed part two of this series. We do have part three on the way where we've got three more big ideas featuring partners from our American Dynamism and FinTech teams. And if you can't wait and want to see our full list of 40 plus big ideas today, you can head on over to a16z.com slash bigideas2024. It's time to build. Thank you.
