We Made Claude Narrate an AI Model Race Like a Sports Commentator | Loop Lab

What if you didn't have to stare at logs while your AI agent worked? In this Loop Lab experiment, Ryan Hamilton built Claude Livecaster, a tool that gives Claude a live voice to narrate long-running agentic processes like a sports commentator.

The demo: six AI models (GPT, Gemini, and Claude variants) race through a CI/CD benchmark, and Claude calls the whole thing play-by-play. Rate limit hits, comeback stories, photo finishes, all of it, out loud.

Whether you're running a long vibe coding session or an overnight eval suite, Livecaster keeps you in the loop without keeping you at your keyboard.

Learn more about Claude LiveCaster: https://circle.ci/4bVpecG
Use Claude LiveCaster yourself: https://github.com/CircleCI-Research/claude-livecaster
See more CircleCI Loop Lab experiments: https://circle.ci/3PSAb7z

About Loop Lab:
Loop Lab is Ryan Hamilton's experimental research series at CircleCI, where the team ships real AI and CI/CD experiments in the open. No polished product demos, just genuine research, weird ideas, and whatever happens when engineers get curious.