5 days ago
How Agent-Native Gameplay Can Blur The Boundaries Between Reality And Fiction
Ben Kim is the cofounder of Nitrode.
Every few years, the video game industry invents what many people expect to become the future of play—VR headsets, non-fungible tokens (NFTs) and currently, non-player characters (NPCs) powered by large language models (LLMs).
I've observed that these waves tend to dominate conversations for a year or so, then settle into a niche because players lack interest, usually due to high costs or because these technologies don't fundamentally change how they interact with games.
LLMs are redefining both how games are built and what games can be. It's important for developers to look beyond AI-powered NPCs and more intelligent matchmaking between players in-game. In my experience working in the space, agent-native gameplay, which refers to titles that run on autonomous AI agents, can deliver new possibilities and create new experiences for players. Agent-native gameplay is possible now due to technological advancements, but take a look back, and you'll see early examples that echoed elements of what is possible today.
Previous Attempts At Creating Ultra-Immersive Games
There have been several previous attempts at creating ultra-immersive games, such as Façade and Evidence.
One notable one? EA's Majestic, which was released in 2001. According to The Obscuritory, a blog centered on obscure games and software, players gave information including their fax and phone numbers, email addresses and instant messenger names. The Majestic team then interacted with players to 'guide them through the game's available 'episodes.''
However, The Obscuritory noted that the game ultimately fell short for various reasons, including poor timing, a lack of technical breakthrough and ultimately insufficient user interest as a result.
How The Technology Has Caught Up
Up until recently, the technology stack wasn't robust enough for developers to build true agent-native games. However, in 2025, researchers studying machine learning explored what web-browsing AI agents could achieve. The researchers presented WebGames, 'a comprehensive benchmark suite designed to evaluate general-purpose web-browsing AI agents through a collection of 50+ interactive challenges.' They found that 'a substantial capability gap, with the best AI system' only attaining a 43.1% success rate. By contrast, human performance had a 95.7% success rate. However, in my view, that 43.1% is still significant—it's close to 50%.
The technology stack has caught up in several ways.
There are now fully autonomous agents that show that LLMs can reliably chain together dozens of steps. Cognition's Devin is a well-known example of a fully autonomous agent currently on the market—it can read documentation, open terminals, fix bugs and ship pull-requests without any human input. With fully autonomous agents, video game developers can create email agents that send players unique messages and clues as part of the gameplay experience.
The web automation tools available today, such as Browser Use and Browserbase, enable web agents to seamlessly execute tasks on websites, such as making purchases and booking services. What this translates to is that agents can be leveraged to deliver real-world products and experiences that enrich gameplay.
There are various solutions on the market, such as ElevenLabs and Cartesia, that can not only handle text-to-speech but can also emulate specific voices. Video game developers can leverage this technology by incorporating voice agents that can call players, sounding identical to their favorite in-game characters.
What Agent-Native Games Can Do
Combined, the capabilities mentioned above enable developers to create agent-native video games where players aren't just immersed in a game, but in a new world.
There are various possibilities.
For example, with voice agents, developers can create an adventure game that, with the consent of everyone involved, clones your friend's voice. Then, your real phone would ring, and you pick up and are offered to go on a journey. The way you respond would dictate the outcome of the game.
With web agents, one example of a game could be one where an agent organizes a select number of packages to be delivered to your door, or sets up a short trip (within a pre-selected distance constraint). Your actions would determine what happens next in the game.
As for email agents, imagine a whimsical role-playing game (RPG), in which each agent-drafted email is a combat move, replies from humans branch the narrative and sentiment analysis feeds a meter that could lead to clues being revealed to players.
Gameplay agents present another opportunity. For example, imagine agents becoming your game buddies. Your job is to outperform the agent in the game you're playing. When you're not there, the agent pushes you ahead.
Ambient life simulations are possible, too. A farming game, for instance, could tie its weather to real forecasts in your city, while a browser agent sells surplus crops on an actual e-commerce site to fund in-game upgrades.
However, with these possibilities come risks. Agent-native games could go too far, putting players in dangerous situations. Developers must be mindful of maintaining player safety, getting consent when it's called for and providing warnings when needed. They should also build in-game mechanisms that enable players to quickly and safely stop playing.
Agent-Native Games Can Ultimately Blur The Boundaries Between Reality And Gaming
Agent-native games invite players into worlds where AI agents can weave their actions into plots.
Because agent capability is continuously improving, developers nowadays are limited less by production constraints and more by the scope of their imaginations. Today's autonomous agents supply video game developers with the missing gears that enable them to blur the boundaries between reality and gaming.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
