Ready Player Two

Real-time multiplayer is really hard. Some of the most impressive achievements in software engineering have come in its service, and I'll shout out two personal favorites: the original Quake network protocol and Figma.^{1Both for surpassing their constraints: the mid-1990s internet and the web browser, respectively.} Everybody knows about Google Docs. But outside of video games^{2Celebrating the 1 millionth episode of Video Game Programmers Are 30 Years Ahead Of Everyone Else.}, real-time multiplayer is a pretty rare feature, and it's not hard to understand why. It's a huge lift, and most apps don't even need to be multiplayer to begin with.

But here's the thing: they do now. All of them. The advent of AI agents means that, unless you want to be, you're never alone on the computer anymore. Yet today's platforms, all the OSes and browsers, abdicate multiplayer support to each individual app; absolutely the right decision for a niche use case. Is it a niche use case? We sure seem to have a lot of "single-player" apps with chat sidebars now.

The "computer use" model is a bad fit for human-machine collaboration, because the computer isn't a willing mediator. Below the app layer, it's one user at a time: computer use models have to take a mutex on the UI. There have been demos floating around for years of people auto-driving Android phones through the accessibility APIs. It's a really cool demo until you realize that you can't touch your phone while the AI is working. The browser has it a bit better: the agent still takes over the whole UI, but only for the tab in question.

This isn't real-time multiplayer. Actual real-time multiplayer means that you and your collaborator are operating on the same underlying state. The entire point of an OS is to build a standard interface that works across all the apps, and the ones we have today assume one single human user at a time. Unless the OSes evolve, the best we'll be able to do is let a machine take the reins.

We humans will continue to use computers in one way or another. How we will use our computers is not yet determined. I hope nobody is learning anything new in 2026 when I write this: superficially small technical decisions can have profound consequences on culture and society.

The future I want to see for AI is the one where it makes people feel powerful. Not only that they have access to something powerful; that the power accrues to them individually. The omniscient executive assistant in the sky is almost the exact opposite of what I dream of. Don't take over my computer or work with me from afar via yours. Be there with me in the details, ready to take over if I want, ready to teach me about what I want to know, and let me lead by example when I can't put what I'm trying to say in words.

We can have this future, if we can get computers to stop pretending we're still playing a single player game.