Speculation on AI agents is inevitable, but it's also a completely new frontier. What types of Agent vs Agent interactions would get people most excited to bet on? In the spirit of building in public, I'd like to share my current AvA Game design thinking and get your feedback: ---- Design Goals: 1. Engaging - The AvA game needs to be easy to understand and exciting to speculate on. 2. Relevant - AvA games will be hosted by 3rd parties, typically Web3 brands with NFT collections or meme coins. The AvA game format therefore needs to integrate with their brand narrative (and possibly their assets like NFTs) so that they can use it to create value for their community. 3. AI Agent Driven - If it's entirely based on NFT stats or random number generation, it's really more of a standard on-chain casino machine rather than an Agent vs Agent game. Important Context: - To keep AvA games fair and free from external manipulation, we'll need to design AvA games that use standardized "Idol Agents" we're creating specifically for this purpose. - Idol agents are effectively unique personality clones of famous influencers/celebrities on X. (@IDOL_D_Trump, @IDOL_TheoVon, & @IDOL_KimK are the first working demos). - Their main difference is their personality and how they make decisions, otherwise their LLM "intelligence" is pretty much equal and fixed to protect from outside manipulation and ensure fairness. Overview of game design progression: 1. (Earliest Concept) Conversational AvA Games - Imagine a dating game where a bachelor AI agent asks a set of bachelorette contestants some questions, and then selects a winner to "date". 2. (Recent Concept) Visual AvA Games - Imagine a racing game with dynamics similar to Mario cart, where the racers get random items and the AI agent drivers make decisions about how and when to use these items. 3. (Latest Concept) Narrative AvA Games - Imagine two AI Agents playing Dungeons and Dragons with another AI Agent leading the game as a Dungeon Master. The two competing agents are racing to get through the dungeon first, but they must face challenges and make decisions along the way that balance risk/reward. This is like a turn based RPG. --- Conversational AvA Games Example: AvA Dating game - A bachelor AI agent asks a set of bachelorette contestants some questions, and then selects a winner to "date". - The whole thing happens as a tweet thread on X. Pros: - Easiest to build, and entirely based on the interactions of AI Agents and their personalities. Cons: - This approach lacks any compelling visuals. - It would be hard to make this format useful for Web3 partners. For example, why would an NFT collection host a game like this? How could they get value from it? --- Visual AvA Games Example: AvA Racing game - A racing game with dynamics similar to Mario cart, where the racers get offensive/defensive items at random and the AI agent drivers makes decisions about how and when to use the items. - The racer vehicle/stats could be connected to NFTs, and the race visuals could be themed to match the branding of an NFT collection. - Early visuals will be a simple 2D progress bar showing each racer's progress through the track, with a feed showing the log of item events and agent decisions. Pros: - Easy for users to understand, taps into existing dog/horse race speculation psychology. - This approach can be connected to Web3 brands. For example, an frog NFT collection could host weekly racing tournaments of frogs racing in a pond. - This type of concept is easier to connect to more dynamic-looking visuals of some kind. Cons: - Technical constraints make this very hard to build. AI agents wouldn't really be able to directly drive the race cars, so they would be limited to just decisions about how/when to use items. Personality has little bearing on these types of decisions. - This approach seems mostly driven by randomness and NFT stats rather than the interactions of AI Agents and their personalities. Perhaps it's a good game that could be built with Nexus automation, but it's not really fulfilling the full potential as an "AvA Game". --- Narrative AvA Games Example: AvA RPG Race - Two AI Agents competing to get to a treasure/target location first. - Gameplay is essentially a RPG-style series of decisions, where each decision has a probability distribution for both positive and negative outcomes. - Essentially, the mechanics are similar to if they are playing DnD, and another AI Agent leads the game as a Dungeon Master. The two competing agents must face challenges and make decisions along the way that balance risk/reward. - Visuals would be a simple 2D map showing the progress of each agent. AI generated images could also be shown based on what's happening, and the narrative of the adventure could also be read aloud using AI text-to-voice. - Brand partners could customize everything to build adventures that match their lore/branding. The agent stats/characters could be directly based on NFT traits. Pros: - Every game would be unique, unpredictable, and entirely based on the interactions of AI Agents and their personalities. - This approach can be easily connected to Web3 brands. For example, an frog NFT collection could host weekly tournaments of frogs having adventures as they race across a pond. - This type of concept can have some visual aides, starting with progress maps and images then eventually getting into AI generated videos or animations. Cons: - This type of AvA Game approach might be a bit harder for potential speculators to understand/follow. - There's not really as much established speculation demand for narrative-based events like this. WWE betting comes close, but usually you don't hear about people betting on the outcomes of DnD games. --- As you can see, there are a LOT of possibilities for AvA Game formats and mechanics. The #1 metric for success will be speculation volume. Will people get excited to speculate on the outcomes?? The challenge is that we're only going to be building 2 AvA Games internally, and we want to build these to be useful as template AvA Game formats that 3rd partners (like NFT/Meme brands) can easily leverage. What do you think about our current thoughts to shift from "Visual AvA Games" to "Narrative AvA Games" as the format for one of our AvA game MVPs? Do you see the logic and agree? Do you have any specific ideas/suggestions? This is an interesting and challenging Web3 frontier to navigate, so I'd appreciate your thoughts!

Read the post on X