five years. latitude has been cooking this thing for FIVE YEARS. and i'm not talking about the kind of five years where a startup burns VC money and ships a half-baked game that trends on tiktok for three days then dies. i'm talking about five years of iteration on the World Engine—six different prototype engines, according to linkedin—until they had something that doesn't completely fumble the core promise: unscripted NPCs with actual memory and dynamic worlds where your choices matter.
look, i was not expecting to care about this as much as i do. i went in thinking "oh great, another AI game that's going to be overhyped hype beast nonsense" (which is like. 90% of AI gaming news lately) and then latitude's pitch landed and i said "wait" out loud. alone. in my apartment. to no one.
so here's the thing about Voyage that makes it different from AI Dungeon (their 2019 thing that everyone played for like four hours then ghosted): it's native AI. not AI as a party trick. the entire game loop is built around the fact that you're working with procedurally generated worlds and NPCs that have actual persistent memory. your choices don't just matter in the moment—they reshape the entire world state.
they're using Google's Gemini Flash for image generation and Gemma for the text/audio/video stuff, plus proprietary models built specifically for this. which honestly? makes sense. you can't just slap an off-the-shelf LLM at a game and expect it to track 160,000+ unique AI-generated characters with distinct personalities and remember which ones you've betrayed. that's a lot of state management. that's the kind of problem that requires actually good engineering.
THE GOBLIN THERAPIST POSSIBILITY SPACE
ok so the pitch mentioned this and i need to talk about it because it fractured something load-bearing in my brain and i am choosing to be fine with that. imagine you're playing an RPG and there's this goblin character. standard expectation: you fight the goblin. maybe it's a boss fight. maybe it drops loot. that's the script.
but in Voyage? you can just... talk to the goblin. and the goblin has a backstory and motivations that are emergent from the simulation, not pre-written. so instead of a fight, you could end up counseling this goblin through its existential crisis. you could become its therapist. an actual goblin therapist questline emerging from the unscripted interactions.
that's not a bug. that's the feature. and it's the most unhinged thing i've heard about this game so far, which is saying something, because we haven't even gotten to the trolls with marriage troubles yet.
the pitch also mentioned trolls with marriage troubles. like—the possibility space here is genuinely wild. because the game isn't trying to predict what you'll want to do. it's just simulating world state and letting NPCs react to your choices in ways that cascade through the entire game. betray a merchant and every trader in the city knows. help someone and they become an actual ally with ongoing relationship dynamics.
THE FIVE-YEAR COOK
here's what i think is wild about the five-year development cycle: RPG design is hard. getting NPCs to feel real is harder. getting memory systems to not eat themselves alive is harder still. getting all three to work simultaneously with 160,000+ distinct characters is apparently a thing humans can do, which is new information.
the progression system is inspired by D&D—skills, luck rolls (like actual dice mechanics), abilities like Counterspell—but it's all running on top of this emergent simulation layer. which means the game is kind of... teaching you how to interact with it as you go. there's a chatbot helper for players who get stuck, which honestly seems necessary because you can't script a tutorial when the interactions are unscripted.
also: players in the beta have made 3,000 gameplay choices on average. which suggests people are actually playing this. not just goofing around for 20 minutes. they're sinking real time and engagement into it.
THE SUBSCRIPTION THING
ok so Voyage is free-to-play with three subscription tiers: $15, $30, $50. the higher tiers unlock advanced AI features and remove action limits. which. look. fine. it's not predatory—compared to most live service games, that's actually reasonable. you can play for free. you can engage with the core game. the subscriptions are for players who want to engage more, which is like... the standard free-to-play model.
but here's where i think it gets interesting: the subscription model is basically Latitude saying "we believe in this so much that we're betting you'll want to engage with it repeatedly and deeply." not "we're going to extract maximum value from whales." the fact that they're not paywalling core gameplay suggests they actually care about the creative sandbox aspect of this.
former Roblox CBO Craig Donato is on the board, which is chef's kiss actually—Roblox prints money precisely because creators can build endless content. Voyage seems to be taking that same logic (players as world-builders) but applying it to AI-powered narrative experiences instead of 3D games.
GOOGLE AND THE FUTURE SCARE
alright so Google's AI Futures Fund is partnered on this. on one hand: that's a ton of computational resources. that's legitimacy. that's deep pockets. on the other hand: it's google. and google's relationship with the things it loves has a recognizable arc, and that arc does not end well for the thing.
Latitude's independence is important to Voyage being what it is. the second Google starts optimizing for engagement metrics and advertising, the whole thing potentially breaks. the unscripted NPC interaction thing works because it's designed to surprise the player, not to be optimized for watch time or click-through rates.
i'm not doom-posting about it—Google's AI division actually seems to understand that good AI experiences need to be good games first. but this is worth keeping an eye on. the indie game industry has watched enough beloved small studios get absorbed and slowly sterilized by acquisition. we know the pattern.
THE REAL QUESTION
ok so the open beta's coming later this year. and i'm genuinely curious to actually play this instead of just reading about it. like—does it actually feel good to play? does the memory system create emergent narratives or does it just create janky nonsense? does the World Engine scale or does it have hard limits that kill the possibility space?
these are the questions you can't answer from reading a pitch. you need to actually sit down and make 3,000 gameplay choices and see if the world actually feels alive or if it just feels alive because there's some procedural generation happening in the background.
but here's why i'm actually hyped: Latitude is showing their work. they're saying "we spent five years on this because it matters." and in an industry that's increasingly full of quick cash grabs and AI hype with no substance, that level of care is genuinely rare.
Voyage opens to expanded beta now and public beta is coming this year. if you're into narrative games, worldbuilding, emergent gameplay, or just... unhinged possibilities... this is worth watching. worth playing. worth telling your goblin therapist about. -- mads