Three Years from GPT-3 to Gemini 3

Google has launched Gemini 3, its new AI model, along with a tool called Antigravity, according to a post on the One Useful Thing publication. The author wrote that they have been testing Gemini 3 and called it "very good."

Rather than present benchmark numbers, the author said they asked Gemini 3 to demonstrate how far AI has come in roughly three years since ChatGPT's release, supplying a screenshot of an earlier post with the prompt "show how far AI has come since this post by doing stuff." According to the post, Gemini 3 built a fully interactive, playable "Candy-Powered FTL Starship Simulator" game, which the author described as a tiny game about using candy power to escape otters, featuring small poems and ongoing updates. The post quotes Gemini 3 as saying that in 2022 AI could describe an engine, while in 2025 it can code the engine, design the interface and let the user pilot the ship.

The author wrote that along with Gemini 3, Google launched Antigravity. They described it as similar to Claude Code and OpenAI Codex, specialized tools that can be given access to a computer and which can autonomously write computer programs with guidance. The post argues that coding ability extends beyond programming because, in the author's framing, everything done on a computer is ultimately code, making coding-capable agents general purpose tools. The author wrote that Antigravity embraces this idea with an Inbox, a place to send AI agents on assignments and where they can ping the user when they need permission or help.

The author also recalled writing before ChatGPT's release that AI would disrupt analytic, creative, and writing- and communication-heavy jobs sooner and more drastically than expected, and wrote, "I think I got that one right."