Three AI coding agents walked into the same prompt. Only one walked out with something worth a damn—and the results tell you everything about who's building real tools and who's building walled gardens.

XDA Developers ran Claude Code, OpenAI's ChatGPT Codex, and Google Antigravity through an identical test: build a 3D time-trial racing game that runs in a browser, with three tracks, medal times, and a ghost of your best lap. Same prompt, one shot, no follow-ups. Claude Code, running Anthropic's Sonnet 5.5, built the game the tester kept playing for over an hour. Codex, running GPT-5.6 Terra, finished in 3 minutes and 16 seconds—and it showed. The track floated in the sky, the "car" was a dashboard-shaped wedge with a red stripe, and the DualSense controller couldn't even start a run without reaching for a mouse. Google Antigravity, running Gemini 3.8 Flash, took 12 minutes and delivered features but not fun. As XDA put it, Antigravity built a functional car, but not a game.

XDA framed the test as a creative showdown—deliberately writing the prompt like a player, not a developer, to see which agent could make something worth replaying rather than just ticking boxes. That's fair. But the real stakes aren't which bot makes a better browser racer. The real stakes are concentration.

Three companies—Anthropic, OpenAI, and Google—now own the AI infrastructure that will increasingly shape how Americans write code, organize information, and communicate. And Google's behavior in particular shows exactly why that should alarm anyone who cares about an open digital public square.

Android Police reported this week that Google Keep, the company's widely used note-taking app, still has no Model Context Protocol integration—the open standard that lets AI assistants access external apps and services. Google has rolled out MCP support for Drive, Docs, Gmail, and Calendar. Keep? Nothing. The reason is obvious: Keep talks to Gemini, Google's own AI assistant, and Gemini only. As Android Police noted, the writer uses ChatGPT and Claude for brainstorming and research, but can't point either of them at notes stored in Keep without manually copying and pasting. Google built the wall, and your data lives on one side of it.

That's not a technical limitation. It's a business decision. MCP is an open standard. Nothing stops Google from supporting it in Keep except the desire to lock you into Gemini. When one company controls the note-taking app, the AI assistant it deigns to talk to, and the search engine that surfaces both, you don't have a tool. You have a cage.

Now scale that up. If three firms control the AI models that write code, synthesize information, and generate content—and two of them are actively walling off their ecosystems to steer you toward their own assistants—then the question of which model builds a better racing game is the least important question on the table. The important question is who decides what the models are allowed to build, what they're allowed to say, and what data they're allowed to see.

Claude Code won this round on merit. But merit alone doesn't keep markets open. The same concentrated power that locks your notes behind Gemini's garden can just as easily lock your speech behind a content filter. The founders didn't fight a revolution to hand the printing press to three companies and hope for the best.