Founder Xiao Hong is extending Manus from an agent that completes work into a creative tool for building, revising and sharing browser games.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [Manus](https://x.com/ManusAI/status/2104684331029742014)
Why it matters #
Manus is extending an AI agent from completing work into making editable, shareable software. The demos show a broader creation workflow, while repeat play and reliability remain the tests that matter.
Manus says its September 28th release of Manus 2.0 can turn natural-language prompts into playable browser games, adding a Game Dev environment to an AI agent largely sold on getting work done. Founder Xiao Hong (@Red_Xiao_) is betting the agent can move from executing tasks to helping people make things they can play.
That extends Xiao's original pitch. In a 2025 interview with the South China Morning Post, he described an AI agent as something that “does not only think and answer questions, but interacts with its environment, collects feedback.” Manus' new game workflow makes that idea visible: users can generate a game, play it, then change its code, assets and scenes rather than accept the first output as finished.
From agent to game-making tool
Manus' official Manus 2.0 announcement describes Game Dev as a combination of coding, image-generation and video-generation systems. Users can start from a template or an idea, inspect and edit the result, adjust characters and movement, and publish a playable website. For multiplayer projects, Manus says users can provision an always-on Cloud Computer to host the game.
That workflow is more specific than asking a chatbot to write a block of game code. It puts generation, hands-on editing, publishing and server setup in one product. Manus' new Faraway demo is a first-person exploration game set above a sea of clouds; its page describes the content as AI-generated and potentially user-edited. The thread also includes examples such as Frostline Zero, Critter Kart and Neon Alley Brawl, alongside task replays and playable demos.
Manus presents the showcase as 20 games. The demos illustrate the range of projects its system can assemble, from racers and arcade games to exploration scenes. They are company-produced examples, however, not an independent measure of how reliably Manus handles unfamiliar prompts, fixes bugs or supports a game after people start playing.
Returning players, stable multiplayer sessions and useful editing tools would show whether the output can support a real creative workflow. Manus' launch material explains how a user can revise the result, but the showcased games alone do not establish how much work went into each one or how they perform with sustained play.
The founder's execution thesis
Xiao has built Manus around the idea that AI should act, not stop at producing answers. The broader 2.0 release carries that thesis into a suite of creation tools, including a timeline-based video editor, as well as agent automations and a new desktop workspace. Game Dev gives the pitch an especially tangible form: a prompt can become a thing friends can open and play, and then the maker can edit what the agent produced.
The product also reflects Xiao's path as a repeat founder. Before Manus, he built Monica, an AI browser assistant. In a profile published by 36Kr, Xiao was described as a product-focused founder whose first startup was in business software before he turned to AI applications. His preference has been to build products on top of foundation models rather than train one himself.
Manus 2.0 arrives as Manus resumes independent operations. Manus said on September 1st that its founding team would continue leading the business. RuntimeWire previously reported that the transition involved users backing up data ahead of a deletion window; the product relaunch gives Manus a substantial new creation feature to put in front of customers. (Manus) RuntimeWire's earlier report on the separation covered the user-data transition.
For Xiao, games broaden the audience for an agent whose earlier demos centered on research, websites and business workflows. The bet is that people who have never opened a game engine may still have a game idea worth trying, if the first playable version is close enough to make iteration feel possible. The editable code and asset controls keep the user involved after generation instead of presenting the game as a finished, opaque artifact. Manus is entering a field with dedicated prompt-to-game tools. Rosebud AI, for example, also offers browser-based game generation, editing and publishing. Manus' proposed difference is the connection to its general agent and its Cloud Computer hosting, rather than a claim that prompt-to-game creation is new. The opportunity for Xiao is to make games another task Manus can carry from idea to working result. Whether creators keep using it will depend on what happens after the demo: how much they can shape the game, how well it holds together and whether sharing it brings anyone back.