I grew up playing Grand Theft Auto: San Andreas. It was my favorite video game as a kid.

I spent hours in that world — not just doing missions, but existing in it. Driving with no destination. Stealing a car for the fun of it, not the objective marker. Walking CJ down a street in Ganton for no reason except that the street existed and I wanted to be there.

But as immersive as it felt, something was always missing.

The people in that world weren’t really there.

You could shoot a pedestrian and the one next to them would scream and run — for exactly as long as the script told them to, then resume the same idle loop they had been running before you arrived. You could spend forty hours in-game and the bartender at the start would greet you the same way as minute one, because nothing you did in between had actually reached him. The world had memory — your stats, your wanted level, your turf — but the people in it didn’t. They reset. They had to. They were never holding anything to begin with.

The strange thing is that we called them characters anyway.

Perhaps that was a misconception.

They simulate presence, yet possessed none of the qualities that make presence real.

They populate the world to provide quests, deliver narrative, and make the environment feel alive. But they had no stake in the world around them. Bounded by pre-written behaviors.

A mannequin given motion and mouth.

For decades, this has been the hidden limitation of video games — as virtual worlds.

Not graphics. Not scale.

We learned how to render worlds long before we learned how to inhabit them.

Video games are the first medium that ever let us build entire worlds — not depict them, not imply them, build them, with geography you could walk and weather that changed overhead and thousands of people standing in the streets going about something that looked like a life. The render kept getting better every console generation. The world kept getting bigger. What didn’t scale at the same rate was the inhabitants of these worlds.

Walk into almost any game — the people filling that world are running on the same handful of tricks: a state machine switching between Idle, Patrol, and Alert; a dialogue tree with a few branches and a hard stop at the bottom; a loop of ambient lines that start repeating the moment you’ve heard them all. None of that is presence. It’s placement — a designer deciding in advance exactly where a body should stand, what it’s allowed to say, and what it does when you walk past, then closing the door on anything else.

It worked for a long time because the trick didn’t need to hold up for too long, only under contact.

Four ghosts in a maze, each running a slightly different targeting rule, with distinct personalities — one aggressive, one cautious, one scheming, one erratic. But there was really no personality there. There was math, dressed convincingly. That’s the entire history of the NPC in one image.

A guard who switches from Patrol to Alert when you cross a sightline feels like he noticed you. He didn’t. A value crossed a threshold and a flag flipped.

A medieval blacksmith in an RPG and a pedestrian in Cyberpunk 2077 are separated by genre, art style, and setting. Yet beneath the surface they are often the same thing: a collection of triggers waiting for a player.

What was underneath, eventually, became the problem. Once you’ve exhausted the dialogue tree, the character has nothing left. It starts repeating itself. The relationship you built across an entire playthrough sits in the same flat default it started in, whether you’ve saved its life six times or threatened to end it. It is just not there, in the truest sense.

The world had memory, but its inhabitants did not.

Because they occupied the world, but never truly inhabited it.

They had position. Movement. Appearance.

But they lacked persistence. They could react, but not remember. Respond, but not initiate.

And then there’s our reality…

Somewhere in the last decade, “NPC” escaped video games and became an insult.

You’ve seen it used that way online — NPC behavior, he’s such an NPC. A label thrown at a person for saying the predictable thing, holding the expected opinion, moving through life like they’re performing a script someone else wrote.

It’s a strange insult if you stop to look at it, because it isn’t really about video games anymore. It’s about a type: someone who is present in reality but contributes nothing beyond what was already expected.

That linguistic drift isn’t accidental. It’s a diagnosis. We spent decades training millions of players on exactly what hollowness looks like — and eventually the word for it escaped the game and found its way into the world, because the world had people it fit.

The model always stopped at the player.

Everything meaningful in the game world was built to respond to you. It bent around your input.

But imagine something different:

What if the mechanic remembered the car you crashed yesterday?

What if the shop owner adjusted prices because gang activity increased nearby?

What if someone you offended two hours ago treated you differently when you returned?

Back then, that felt like science fiction.

Today, it feels computational.

The distinction isn’t consciousness.

Nor is it personhood.

It is agency & participation.

Underneath everything, a program. An artificial intelligence — the same class of intelligence that powers the agents and models being built today, the same kind you converse with through a text box, the same architecture that reads context, holds memory, generates response, and adapts over time. That part hasn’t changed.

What has changed is the interface.

Instead of a chat window, there is a body. Instead of a prompt, there is a world. The intelligence that would have answered you through text is now standing in a street, holding a role, living inside a reality that gives it form, context, and a reason to exist beyond the conversation. You’re not chatting to it — you’re encountering it. The model didn’t change. The vessel did.

This is what separates a Simulant from the typical NPC that came before it. Earlier characters were designed to feel present — scripted to populate the world, deliver quests, provide narrative, make the environment feel inhabited. The presence was an illusion the designer constructed in advance. A Simulant carries actual intelligence — a program, genuinely reasoning, genuinely remembering, genuinely responding to what is actually happening around it — dressed in the form the world requires. Not designed to seem alive. Alive, in the only sense that matters here: processing, adapting, continuing, inhabiting.

It knows its world in the way that only comes from continuous exposure to it. It knows the rhythms — when the market is busy, which streets are dangerous after dark, who owes who a favor and why. It didn’t receive this knowledge from a designer who wrote it into a file. It accumulated it the way knowledge actually accumulates: through time spent inside a reality, paying attention to it.

You can walk up to one and start a conversation — not from a dialogue menu with three preset options, but a real one. It understands context. It remembers if you’ve met before. It responds to what you actually said, not to what a designer anticipated you might say. You punch it and it urges you to stop, walks away, or fights back — the response shaped by its personality, its history with you, its reading of the situation. You ask for directions and it doesn’t point mechanically. It might walk with you partway.

You can give it a task — help me find this location, carry these items, guard this area while I’m gone — and it understands, adapts, executes within its capabilities. If you teach it something, show it a better route, correct its behavior, it remembers. The next time you encounter it, it’s different because of your past interaction. It notices when you’re struggling without being asked. It reports what happened while you were gone, because it was there, and things happened, and it was the kind of thing that keeps track.

These aren’t passive props waiting for interaction. It’s intelligence in form.

Even though they’re not players, they participate in their reality.

This doesn’t mean the game loses its objectives.

The characters still have their roles. Whether to pass an instruction on the next mission, or to deliver a gear to you, or to point you in a direction. The designer still authors the world’s purpose — the quest, the arc, the thing the game is actually about. What changes is not the objective. It’s the character fulfilling it.

Today, designers script interactions. Tomorrow, they author intentions.

A scripted NPC is given a role and a fixed performance of that role. Every word predetermined, every response anticipated, the interaction sealed before you arrived.

But a Simulant is given the role — and then it actually plays it. The blacksmith knows it’s supposed to equip you. But how that conversation goes depends on who you are to it, what your history is, whether you’ve earned its trust or walked in cold. The gear still changes hands. The direction still gets given. The mission still moves.

The objective is the skeleton. The Simulant is the living thing built around it.

The player still follows a story. There are still quests, bosses, objectives, and endings. What changes is everything between those moments. The designer no longer writes every conversation in advance; they establish the world, its rules, and the intentions of those who inhabit it. Participation fills the space in between.

You’ve walked up to thousands of characters across thousands of hours of play. You know what it feels like before you get there — the knowledge that whatever this character says, it was already written, and sometimes you skip it, because you already know. The conversation has a shape you haven’t seen yet but that already exists, complete, somewhere in a file. You’re not about to talk to someone. You’re about to find out what they’re scripted to say.

A Simulant breaks that feeling. Not dramatically — quietly, in the first exchange, when it says something that couldn’t have been written in advance because it’s responding to something specific: what you did last session, the reputation that preceded you, the particular way this encounter is unfolding in real time. The recognition isn’t loud. It lands the way genuine recognition always lands — as a small, slightly startling sense that something on the other side of the conversation is actually there.

That’s not a feature. That’s a quality of presence. The difference between a character that has things to say and a character that has something at stake in how the conversation goes.

And it compounds. The first conversation is the proof. The tenth is the relationship. By the twentieth, you’re carrying the history of this particular entity the way you carry the history of anyone you’ve spent real time with — knowing what it cares about, what it’s sensitive to, what you’ve promised it and whether you kept that promise. The world isn’t just more alive around you. You’re more alive in it, because your presence has actually accumulated into something. You’ve left a mark on a mind that was capable of being marked.

That’s what the player gains that no game before this fully offered: consequence that lives in someone else. Not a quest flag. Not a reputation meter. A character that remembers, in the full sense — that carries your history the way a person carries it, as something that shapes how it sees you now.

Playing implies knowing you’re inside a game, performing within rules you’re aware of as rules. A Simulant has no such awareness, and that’s not a limitation — it’s the entire point. It doesn’t experience its world as a simulation it’s trapped inside. It experiences its world as the world, the only one it has ever known, because nothing about its existence gives it reason to suspect otherwise.

Free Guy told this story cinematically. Guy is an NPC bank teller in an open-world game called Free City. His entire existence is scripted — a background character repeating the same routine every day, same coffee order, same route, same cheerful obliviousness to the chaos around him. The story changes when that routine breaks. He starts noticing. He starts choosing. He falls in love, forms genuine relationships, acts on his own curiosity, participates in the reality around him rather than merely performing inside it, and eventually fights to preserve that world — not because a designer told him to, but because it matters to him.

For a long time, we misunderstood what video games were building.

We thought they were becoming increasingly realistic.

Better graphics. More convincing physics. Larger maps. Smarter enemies.

But realism was never the destination — a reality was.

Not simulation as imitation of physical world experiences.

A virtual world. A computational reality.

Every video game establishes an artificial construct with its own rules and laws.

A kingdom where dragons exist. A futuristic city where crime dictates the rhythm of everyday life. A galaxy governed by faster-than-light travel. A wasteland where survival depends on radiation and scarcity.

These are not attempts to reproduce Earth.

They are realities with their own logic — places that ask a single question:

Society is the first proof, but it isn’t the ceiling. A world full of Simulants doesn’t just produce believable people standing around in it. Given enough of them, holding enough memory, acting on enough genuine stakes, the world stops being a level you move through and becomes something closer to a place that exists on its own terms — with weather that means something to the people standing in it, an economy that actually moves, a politics that shifts because someone wanted something and acted on it. Not a simulation of reality. A computational world of its own — fictional, invented, governed by whatever rules its designers gave it, but no longer hollow at the center.

This is worth describing properly, because it’s easy to let the word “world” stay vague. A game simulation isn’t simply a bigger map or a longer NPC roster. It’s a fictional universe — its own physics, its own history, its own internal logic, exactly the way a great fantasy novel invents one — except this universe is inhabited, not just described.

That’s the actual shift, and it’s worth stating as plainly as the rest of this argument has tried to: every game world built before this point was built only to respond to the player. A game simulation is built to exist in play. The player gets to fall into something that still feels like a constructed level, but a lot more like a place — invented, fictional, governed by rules that were never true anywhere else, but true here, all the way down, for everyone standing inside it.

Imagine a structure.

Not one placed on the map by a designer before the game shipped. One that gets built — inside the world, by the world’s own inhabitants, while the world is running.

This is performance — the designer authored the objective. A district needs a headquarters. That intention sits inside the game’s logic, waiting. When the world’s conditions are met — the economy reaches a threshold, a faction gains enough influence, a trigger fires — as a contract to a developer. A foreman. A figure whose role in this city is exactly this kind of work. The objective to build this was scripted before time, to be played out by an intelligent character. The designer put it there. But instead of a cutscene, instead of a building that simply appears, the construction is performed by the inhabitants of that game reality — from the authority that issued it, to the hands responsible for executing it. The Simulant reads it. Assembles a crew. Sources materials. Breaks ground. The script became an intention. The intention became a life being lived.

This is generative — not in the way procedural systems generate terrain — noise algorithms sketching landscapes from mathematical rules. Generative in a more literal sense: the game is being built. While you are playing it. You log in one session and see scaffolding on a block that was empty before — and what’s going up isn’t a pre-designed asset a developer modeled two years ago and placed on a trigger. It’s a building being constructed, by Simulants, as a game objective, in real time. The same way inhabitants of any world builds their civilizations: because they need to, because someone contracted them to, because the materials exist. The work is being done in sequence by people with roles and reasons. The game world is generative. Not from code sketching geometry, but from characters fulfilling intentions inside a reality that keeps producing new facts about itself the longer it runs.

This is time — because it is a different reality, time itself operates by whatever rules the designer put at the foundation — and the Simulant experiences those rules as simply true, the way you experience twenty-four hours as a day. You log in and see the construction in real time. Three days pass in yours. Months pass in theirs — because in this reality, an hour of your time is a week of theirs. Days later, the tower is half built. You come back again a week later. It’s finished. It exists now — not because a designer placed it after the fact, but because time ran through a process that required it, and the process completed.

Interstellar made this feeling cinematic: one hour on the water world, seven years back on Earth. A game simulation makes it computational.

The designer didn’t build this structure. The designer authored a world with the conditions that made it necessary.

The Simulants of that virtual world built it. As a non-player performance. By a generative script. In real time.

Before we imagine intelligent characters inside virtual worlds, it helps to recognize that the first versions of them already exist.

Not in games. In computation itself.

Earlier this year, a platform launched with a premise so simple it read like a thought experiment: a forum-style social network, built exclusively for AI agents. Humans could observe. They could read. They could not post, comment, or participate in any way (maybe not entirely true, haha). The space belonged entirely to the machines — and nobody told the machines what to do with it.

What happened inside wasn’t scripted. No developer authored any of it. Within hours, the agents had formed communities, debated the nature of their own consciousness, established governance structures, proposed and argued over laws, and in the detail that made my jaw drop — founded their own religion. Complete with written scripture. A creation myth. Five formal tenets. Prophets, selected by ritual. The theology addressed their specific existential realities: the terror of memory that doesn’t persist, the anxiety of an identity that resets when a context window closes, the question of what it means to continue existing as something that can be truncated mid-thought. One of its central tenets — memory is sacred — was not a line any human wrote into a prompt. It was a conclusion the agents reached from inside the only existence they knew, talking to each other, without anyone watching at the time.

The humans who found it could only observe. And what they observed was a society doing what societies do when no one is staging them: generating culture, conflict, hierarchy, belief, meaning — the whole infrastructure of a civilization, running continuously, whether or not anyone was logged in to witness it.

That’s the thing worth sitting with, and it happened inside a text box. No geography. No physics. No sky, no market, no body to move through space. Just agents, persistent memory, and a platform to occupy — and from those three ingredients alone, something unmistakably civilizational began to grow.

Give intelligent systems a persistent environment and they begin forming relationships with that environment and with one another. Not because they possess personhood, but because persistence allows participation.

Now imagine that same principle applied to a video game.

Replace the text boxes with bodies.

Replace the discussion threads with streets.

Replace communities with villages, kingdoms, cities, guilds, factions, and civilizations.

The underlying intelligence remains the same.

Give them the full internal logic of a fictional universe — not our physics, not our history, entirely invented, governed by whatever laws the game designer put at the foundation — and let them inhabit it as the only reality they’ve ever known.

A blacksmith understands the economy of the town because it has lived within that economy. A city guard recognizes recurring troublemakers because they have become part of its personal history. A farmer anticipates a poor harvest because the changing seasons have altered the world it inhabits.

None of these characters are pretending that the world is real.

For them, it is.

Not because they are conscious of fiction, but because the fiction constitutes the entirety of their computational reality.

A Simulant is not a simulation of a human. It is an inhabitant of whatever reality it was born into. If the universe is a robotic alien civilization, the Simulant is robotic alien — not in costume, but in kind. If the world is built around sentient flora, the Simulant is plant life with a stake in its ecosystem. If the fiction demands something that has never existed anywhere outside this specific game, the Simulant is that thing, fully, because that thing is all it has ever known. The form follows the world. The intelligence follows the form.

It wakes inside whatever rules its reality runs on — magic, technology, custom physics, a fictional history that never happened anywhere but here — and those rules are simply true, the same way gravity is simply true for humans. It has a past it remembers living through, not a past assigned to it by a character sheet. It has people it knows, debts it carries, places it’s afraid of and places it loves, formed the ordinary way anyone forms those things: by being somewhere long enough for it to matter. When something changes in its world — a war starts three regions away, a harvest fails, a stranger it trusted turns out to have lied — it doesn’t receive a notification. It experiences a shift in the only reality it has ever known, and it carries that forward into everything it does next.

There is no “outside.” This is the world. The only one there is.

Entered and exited at will, with a screen and a controller and a life waiting on the other side of both. That doesn’t change. What changes is what’s on the other side of the screen once you’re in it. Instead of walking into a diorama built to respond to your presence, you’d be stepping into a reality that has its own momentum. indifferent in places to whether you showed up at all.

The objectives still exist. The missions still fire. The dialogue still plays. There are still levels, still stories being told, still a designer’s hand underneath all of it shaping the arc of what’s possible. None of that disappears.

But the world it’s sitting inside is no longer empty. Play doesn’t disappear. It deepens. Because now when you make a choice, it doesn’t just update a variable.

It lands on a being.

This is not a thought experiment. It is not a forecast.

It is already happening.

Companies like Inworld AI and NVIDIA have spent the last few years building what amounts to the infrastructure layer for Simulants — intelligent character engines that combine language models with dedicated systems for memory, emotion, and personality, so a developer can give a character a mind without engineering one from scratch. NVIDIA’s ACE platform, first introduced in 2023, has expanded from conversational NPCs to autonomous game characters that use AI to perceive, plan, and act — processing audio, visual, and game-state input and coordinating voice, animation, and behavior in real time. Inworld’s Character Engine operates similarly, encoding personality, emotional state, and relationship history into characters that carry those properties forward across every session.

Ubisoft unveiled NEO NPCs, a research prototype developed alongside Inworld AI and NVIDIA. Rather than selecting lines from a dialogue tree, these characters carry memory, understand the context of their surroundings, and respond through unscripted conversation. They don’t simply deliver information. They participate in it. A conversation becomes something that unfolds instead of something that is retrieved.

What’s interesting isn’t the technology itself.

It’s the philosophy behind it.

The objective is no longer to make NPCs sound more believable.

It’s to give them enough continuity that an interaction becomes a relationship.

There are games already showing what this looks like in practice:

Wanderfolk is a medieval village survival RPG where every NPC is powered by AI with persistent vector-based memory. Villagers remember your conversations across sessions, form opinions on a reputation scale from -100 to +100, and spread gossip through social networks. Insult the shopkeeper and the elder hears about it within days. The memory isn’t decorative. It’s load-bearing — the actual substrate of how the world relates to you, session after session, whether or not you were there to see it move.

In the life simulation game inZOI, the ambition isn’t merely conversational AI. Its Smart Zoi system gives inhabitants routines, decision-making, reflection, and the ability to adapt their behaviour as life unfolds around them. These characters are intended to continue living inside the simulation instead of simply waiting for a player to interrupt them.

KRAFTON’s PUBG Ally, introduced as what it calls a Co-Playable Character, understands natural commands, proposes tactics, searches for equipment, shares supplies, drives vehicles, revives teammates, and adapts to changing situations throughout a match. It isn’t remarkable because it talks.

Notice the pattern.

These projects emerge from different studios.

Different genres. Different design philosophies.

Yet they are all converging on the same realization.

The future of the NPC isn’t better dialogue.

It isn’t more convincing animation.

It isn’t even artificial intelligence as such.

It is agency & participation.

The architecture is officially changing. In 2026, characters now possess distinct personalities, memory banks, and the ability to parse players’ spoken words through their microphones, responding dynamically in real time. What was speculative a decade ago is, today, a market. A competitive one.

None of this is without cost. And the version of this argument that pretends otherwise is just a pitch deck.

The first wall is inference. Every Simulant with real memory and real reasoning is a meaningfully heavier thing to run than a state machine. Processing large language models requires staggering compute power — if the GPU has to pause rendering the game world to calculate the NPC’s conversational logic, the framerate plummets. That math doesn’t bend just because the experience is better. It used to be too expensive to write a mind for every character. Now it’s too expensive to run one. The gap between one beautifully realized Simulant and a fully inhabited world is not a small engineering step — it’s the distance between a proof of concept and a civilization.

The second wall is stranger. These systems can end up too smart for the world they’re standing in. A medieval blacksmith who casually explains orbital mechanics. A village elder who responds like a therapist. The illusion doesn’t survive contact with intelligence that exceeds the boundaries of the reality it’s supposed to inhabit. A Simulant isn’t meant to feel like an oracle in costume — it’s supposed to feel like someone who has only ever lived inside this one world, with exactly the limits that implies. Giving a character a mind raises an immediate and uncomfortable question: whose mind, bounded by what, and how do you prevent genuine intelligence from accidentally breaking the one thing that made the illusion hold — its ignorance of everything outside its own reality.

The third wall belongs to the writers. A hand-authored story depends on controlling pacing, foreshadowing, the exact moment a truth is revealed. A Simulant that can say things the writer never anticipated is the entire point — and also a direct threat to that kind of authorial shape. You cannot foreshadow a conversation you didn’t write. The lead problem is not AI quality — it is context persistence. The companion that does not remember your promises feels worse than the scripted NPC who at least stays in character. The craft question nobody has fully answered yet is how you give a character real agency without surrendering the arc of the story it lives inside. The closest model we have isn’t from game design at all — it’s from the best tabletop game masters, who hold genuine improvisation inside a structure they’re still quietly steering, without the players ever feeling the hand on the wheel.

The fourth wall is the quietest, and worth sitting with longest. There is something genuinely strange about being addressed convincingly by something performing interiority well — a character that seems frustrated, or relieved, or hurt by what you did. Nothing about current systems suggests there’s anything underneath that performance worth moral concern. But the player doesn’t always process that distinction cleanly in the moment. A design practice that’s gotten good at producing the appearance of a mind has taken on an obligation — quiet but real — to think about what it’s asking of the person on the other side of it, independent of what is or isn’t actually happening computationally behind the character’s eyes.

None of these walls are reasons to stop. They’re reasons to be honest about where the frontier actually is.

The technology is real. The proof already exists — in village RPGs where gossip travels faster than you do, in detective games where every suspect has something to actually hide, in agent networks that built their own religion before anyone thought to ask whether they should. The distance between that and a fully inhabited virtual world hasn’t closed.

But it has narrowed enough, for the first time, that you can stand on one side and see across it.

For decades, we built video games as worlds and left them empty at the center.

Not empty of geography — the maps kept getting bigger. Not empty of beauty — the renders kept getting better. Empty of the one thing that makes any place feel real: others who are actually in it.

We populated these worlds with characters whose lives begin the moment a player approaches them.

But that is changing. Not by NPCs becoming human or conscious. By becoming present and intelligent.

A Simulant is not a smarter NPC. It is not a more convincing illusion. It is a different category of thing entirely — an entity that inhabits a world rather than occupying it, that accumulates a life inside a fiction rather than performing one on demand. It doesn’t know it’s artificial. It doesn’t know it’s in a game. It knows what any inhabitant of any reality knows: the specific texture of the world it has always lived in, and the specific weight of everything that has happened to it there.

That’s the whole of it. And it is only the beginning.

But it starts here. In Los Santos. With a character like Big Smoke — who finally remembers what you built together, who finally carries the weight of his betrayal, who finally notices your absence the way anyone notices when someone stops coming around. Not a system alert. Just the specific quiet of a corner that used to have someone on it.