How does this game *run forever* on Vercel, automatically retry failures, time out unresponsive LLMs, stream updates…? Enter: Workflows Each ♛ game is a "𝚞𝚜𝚎 𝚠𝚘𝚛𝚔𝚏𝚕𝚘𝚠" Each ♞ move is a "𝚞𝚜𝚎 𝚜𝚝𝚎𝚙" It's just JavaScript™. Opus picked up immediately, even with no exposure during training (given its recency). Great validation of its design. The local DX, testing story, and the built-in observability with tracing is *chef's kiss*
This is fun. Left the autonomous chess match (v0-chess-match.vercel.app) running overnight between @xai Grok 4 and @openai GPT 5.2. Grok's 𝚐𝚛𝚘𝚔-𝟺-𝚏𝚊𝚜𝚝-𝚛𝚎𝚊𝚜𝚘𝚗𝚒𝚗𝚐 won 19 of the last 20 matches 😅 What models should battle next?