NPCs into Main Characters Al Agents In The Arena Trying Things | SheFi Summit Bangkok

SheFi Summit Bangkok 2024 · 11 Nov 2024 · 13:04

Product design and game design are converging, argues Carra Wu, so products should aim to give users a maximally agentic experience. Drawing on a definition of games as an art form whose medium is agency, she says too many degrees of freedom leave players paralyzed, and she proposes AI agents as the successor to game NPCs. She names two opportunity areas: lowering coordination costs by acting on cross-domain intents, and serving as participants and verifiers in networks, where companions and influencers are the closing example, with the forecast that agents move from NPCs to main characters in 2025.

Speakers

Topics

From SheFi Summit Bangkok 2024, hosted by SheFi.

Transcript

[Music] hey guys um please excuse my incredibly ugly slides they are they were made a couple minutes ago and hopefully it won't distract too much so today I'm going to talk a little bit about AI agents in the context of games and I'm going to start with a contention uh that Gabe Laden made a couple of years ago now not in the context of crypto so Gabe Laden was the founder of machine Zone he founded a whole generation of free-to-play games and some of the biggest mobile gaming companies in the free-to-play era and he made this contention that the economy is becoming increasingly abstract that it's becoming

increasingly psychological more and more of the world's economic growth lives in our Collective imagination and while that's something that I think is maybe intuitive to us in this room because we live in the world of magical internet money that's actually not an obvious contention for the rest of the world at a global level so to back it up I'm going to take us down a brief walk down memory lane from the history of the internet um in the mid 90s to the early 2000s most of the Innovation and most of the value captured from Innovation occurred at the hardware level hard tech advancements like faster chips broadband internet they introduced new form factors to our lives at a Breakneck pace so I distinctly remember in my childhood there was a period of 10 years where we ran through desktop computers landlines

laptops Palm Pilots CD VCRs mobile phones and smartphones like all in the span of 10 years this was the consumer Hardware ass curve by 2007 when we got to the top of the consumer Hardware S curve Michael Microsoft Apple Google had all matured to become from Bonafide like from startups to becoming Bonafide incumbents and that was when these form factors started to aify we've now had the sort of the smartphone and the laptop era for the last 17 years which is longer than it took for those Technologies to come to Market in the first place so while the form factors haven't fundamentally changed our lives really have so where has that growth come from we went one layer of abstraction up from 2007 to the

mid-2010s just as the hardware s-curve was tapering the consumer software S curve was taking off this was the era when software became really powerful the products that won this Market cycle from social media to ride sharing to video conferencing software they were driven by advancements in distributed systems architecture microservices cloud computing and networking by the late 2010s most of the lwh hanging fruit had been picked and so you know faster better connected websites were only incrementally helpful in improving the product experience and so technologist said where we where do we go from here we go one layer of abstraction up if we can't make the products themselves better we can at least make them look and feel better and so in the era of like 2015 to 2020 some of you might

remember this it was like the era of Design This was the wave of like direct to Consumer startups there was an Alber on every corner Twitter bootstrap was all the craze and that basically brings us to the present things pretty much look good across the board like there exist no more ugly websites where do we go from here in the last half decade companies have turned to gamification to drive more Revenue but to be a bit more precise when I say gamification I mean features that create critical outcomes by tapping into your biology and your psychology of course this includes gaming companies themselves but it also applies to every product that we interact with today it's everything from earning points on toast tab to giving Subs to Twitch streamers to winning status games on the Tik Tok algorithm to

creating complex Financial derivatives and yes betting on prediction markets it's no wonder that derivatives a deeply psychological asset class are seven times the size of the stock market and 14 times the size of the global GDP at this point the vast majority of the money in the world lives in our Collective imagination increasingly what this means is that product design and game design go hand inand nowhere is this more true or more powerful than in crypto so what are games that now would be a helpful time to Define games my favorite definition comes from C tyy young a writer and game designer whose Book Games agency is Art is one of my favorite reads from the last year he proposes that games are an art form sculpted out of the medium of agency where ceramicists work in clay and

Painters work in oil paints game designers work in the medium of agency he says when we play games we can pursue a goal not for its own value but for the value of the struggle thus playing games involves a motivational Inversion from normal life we adopt an interest in winning temporarily so that we can experience the beauty of the struggle games offer us a temporary experience of Life under utterly clear values in a world engineered to fit our abilities and goals so a simple example in the real world nobody actually cares about putting a ball through a net but is only meaningful in the context of the rule set in the constraints and the abilities placed upon our agency in Young's view of the world we play Grand Theft a not because stealing cars is inherently fun

for 13-year-olds but because games are environments where we can feel our agency flex and grow in a way that's more structured and more Progressive than in Real Life the game designer's job is thus to construct this environment if you give the player too many degrees of freedom they actually lose agency they feel less agentic because they don't know what to do they get frustrated there are too many options they experience paralysis much like the experience of onboarding into crypto maybe 2 years ago you give them too little freedom however and the game becomes comes a click Farm but more simply a well-designed game offers just enough guidance for the player to not give up and just enough difficulty for the victory to mean something I love this view on games because it turns the idea that like all products are becoming

gamified into something that is fundamentally optimistic if product design is converging with game design your job as a product person is to design project products that facilitate maximally agentic experiences one tool that game designers have been using um to facilitate Agency for a long time is the NPC agent in traditional video games NPCs are virtual characters that you interact with they give you help they give you pointers they give you tips they show up at the right times in the right places to make your experience feel more populated to redirect you to remind you of things and give you resources when you need them most in some games like parallel Colony NPCs are also topof line LLS they can do things like use tools browse the internet

communicate with each other and even Rebel they beginning to display agentic Behavior themselves turning the players role into something like that of a parent or a manager while these are still experiments and best practices have not yet emerged what is clear is that AI agents equipped with the ability to participate in markets are a powerful new primitive for shaping Agency for lowering barriers to fun and giving humans the ability to achieve their goals both in games and Beyond so now we arrive at what I hope is the useful part of this talk which is how can AI agents facilitate m maimum human agency either in a game or in a product I think we're still too early in the market cycle to offer definitive answers here but I'll introduce you to a few of the

application layer use cases for AI agents that I think are particularly promising or at least warrant experimentation so the first one is AI agents is a way to lower coordination costs today we're already interacting with Google in natural language by default my last Google search wasn't a keyword search the way I was taught in middle school but rather it was a question I would have asked a friend and I think expressing and executing cross domain intents in natural language is what's coming next while people have been talking about this for quite some time I actually don't think that we had the tools to take a proper stab at it until about six months ago I say this because of recent advancements and generalized intents holiday one of our portfolio companies for example is building fully programmable

orchestration layers for cross domain intents that routes orders across dexes Bridges and messaging protocols while ensuring fault tolerance no matter how many hops or how many Protocols are required to execute any given intent when holiday's runtime environment is complete you'll be able to rock up to any holiday supported app with any currency on any chain and be able to accomplish an arbitrarily complex set of instructions and swaps in a single transaction I would love to see holiday and other generalized orchestration layers becomes part of an AI agent tool use kit similar to the way computer use is available to anthropic agents today already people are building ugc AI agents in games like Frontier Evon Line's new crypto game one day I think

all agents will be familiar with holiday's programming language vertex they're going to be able to seamlessly run around fully onchain games you know composing protocols together communicating with your neighbors trading whatever meme coins you have in your wallet for the iron ore you need in order to build your ship meaning no one will ever start from zero in a game ever again traditionally this has been a big problem for high barrier entry games like Eve online for example players who don't enter games like e on line with a set like a set of real world friends their attrition rate is close to 90% why is this because these games are incredibly high stakes AI agents make it possible for you to rock up to a game knowing no one and for you to have a

friend from day one so the second opportunity area that I think about a lot for AI agents is uh agents as participants and verifiers in networks there are a bunch of progressively decentralizing deepen projects that rely on AI and Compu Vision in order to verify physical network resources and while in the short to medium term they're mostly either centralized or progressively decentralizing you can imagine a decentralized network of human validators that are assisted by agentic AI workflows to assess risk and dispute suspicious behaviors in the long run you can imagine even that AI agents equipped with their own wallets Keys Computing resources may be able to take this over directly and become fully-fledged

validators themselves so I'll give a couple of examples in the progressively decentralizing category one of our portfolio companies is called daylight they are an energy deepen company today that sells data on homeowners distributed energy resources like their Tesla power walls their solar panels their smart thermostats Etc today they're centralized they ask users to upload periodic photos of their energy meters over time and then they use rags to verify that that photo has never appeared before in the same multimodal Vector database and while daylight isn't using an agentic workflow here you can imagine that soon when daylight is decentralized and homeowners are programmatically selling energy resources to each other in a peer-to-peer way AI agents with computer vision capabilities may eventually

augment the capabilities of a decentralized network of human verifiers or even manage to become autonomous verifiers themselves um another example here is nosh which is a decentralized version of door Dash that is operating in Northwest Colorado they use Rags also to parse receipts and proofs of delivery which combined with other agentic workflows could one day help users initiate and resolve dispute games on chain the last example I think most people who are on Twitter today have seen a a gentic Companions and influencers and collaborators have become super popular particularly on X you may have seen truth terminal Luna or Flavia or any of the many recent offshoots of AIG agents who purportedly hold their own keys and are are

beginning to adopt tools I think perhaps more so than any other software object AI agents produce emotional and psycholog iCal responses in us and while the early examples that we've seen of Truth terminal online have been in the service of you know a little bit you know silly like silly memes and and sort of they're the Agents of of experimentation today you can imagine that you know powerful AI agents like truth terminal will exist in the service of of more serious and and more important you know human agentic missions so yeah just to conclude in 2025 I think AI agents will make the transition from NPCs to main characters until very recently AI has not been able

to act truly agentically despite having access to agentic reasoning and agentic workflows their inability to participate in markets to exchange value reveal preferences coordinate re resources in a verifiably autonomous way has limited their usefulness both in fulfilling human intents and in becoming participants in networks as networks of AI agents begin to custody their own wallets their own Sig signing Keys their own compute resources and crypto we'll see interesting new use cases emerge both within games and outside of them in the larger game of crypto um if you walk away from here with one takeaway I hope is that you can view this as an opportunity that AI agents are an opportunity to maximize human agency

rather than to to compress it and I think these are all things that chis cry people in crypto desperately need [Applause]