A San Francisco startup called Instaplay has attracted 130,000 players by late September to a platform that converts plain-text descriptions into fully functional browser games. The company, which recently emerged from Y Combinator's latest cohort, had more than 25,000 creators by late September.
The pitch sounds almost too simple: type what you want to play, wait a few minutes, then share a web link with friends. No coding knowledge required. No game engine to master. Instaplay's AI handles the logic, graphics, and mechanics behind the scenes, partnering with what the company describes only as "frontier AI labs" to refine its generation capabilities.
Whether that simplicity holds up under sustained use remains to be seen. But the early numbers suggest demand for a fundamentally different approach to game creation, one that bypasses Unity tutorials and asset stores entirely.
From arcade clones to multiplayer experiments
The platform accommodates both 2D and 3D projects across a range of genres. Users have built arcade games, puzzle experiences, and multiplayer titles that support local or online play. Each creation exports as a shareable URL that works across devices without requiring downloads or installations.
That distribution model matters more than it might initially appear. Instaplay's founders argue their games spread organically through group chats and Discord channels, sidestepping the friction of app store approval processes and the competition for downloads. The company offers native apps for Android and iOS alongside browser access, though Google Play data shows modest traction so far—just over 1,000 installs as of a September 6 update.
Who's building this
Akhil Aggarwal, the CEO, previously invested at Bain Capital Ventures and worked as an engineer at Amazon Web Services. Frank Liu, who serves as CTO, spent time as a software engineer at Instagram. The third cofounder, Gary Wu, worked as an associate product manager leading reinforcement learning environments at Salesforce and worked on the company's Agentforce product. Wu holds computer science degrees from Harvard.
The trio incorporated as General Context Labs Inc., listing a San Francisco address on app store pages. Y Combinator and Bain Capital Ventures back the venture, according to the company's LinkedIn profile, though neither funding amounts nor valuation details have been disclosed. The team remains lean at three people.
Growth in tight windows
When Instaplay announced on Y Combinator's Launch platform roughly a month ago—likely in mid-August—it reported 20,000 creators and 100,000 monthly active users. By late September, those figures had climbed to 25,000 creators and 130,000 players, based on the accelerator's directory listing accessed on September 23.
The pace of that growth raises questions about retention and engagement depth that won't be answerable for months. Still, the startup has begun testing applications beyond pure entertainment. A September 14 blog post pitched teachers on using Instaplay to generate quiz-style review games with scoring mechanics and controlled answer reveals.
The AGI angle
In early September, Instaplay published its first research piece: a benchmark evaluating how well leading AI models generate 3D assets using the Three.js library. The company frames this work as part of a broader inquiry into "the role of games in achieving AGI," according to its research hub.
Y Combinator's directory describes the startup as an "AI-native gaming platform turning every game into an RL environment." That phrasing suggests Instaplay sees its creations not just as entertainment products but as reinforcement learning training grounds for AI systems—a dual-use case that could appeal to researchers as much as to casual creators.
A crowded field
Instaplay enters a market already populated by competitors pursuing similar visions. Rosebud AI offers browser-based game building with chat-driven editing tools. MakePlay generates games from text prompts using PixiJS and Three.js, the same libraries Instaplay appears to leverage. GDevelop embeds an AI assistant within a traditional editor interface, while Buildbox 4 advertises prompt-based creation.
The startup drew an explicit comparison to Roblox Studio in an August 13 blog post, positioning its instant browser links as faster for distribution than Roblox's installation requirements. The company acknowledged, perhaps more candidly than most early-stage ventures would, that larger projects might eventually migrate to Studio for expanded feature sets.
Gary Wu framed the company's mission in characteristically blunt terms during the launch announcement. "The world is burning trillions on AI to make us more productive," he wrote, "but almost none of it has made our lives more fun." Whether converting game ideas into playable prototypes within minutes qualifies as "fun" in a sustainable way—or just novel for now—will depend on how well the platform handles complexity as creators push its boundaries.
For the moment, at least, enough people seem willing to find out.
