"Winter Arc team on AMD AI League Hackathon"

CO-AI is a collaborative coding workspace where a team can review an AI-assisted code change, discuss it in context, and preserve a replayable record of the decision. Each coding request lives in a shared thread with the proposed plan, patch, discussion, and reviews. The verification workflow is designed to run tests against the exact proposed revision in a disposable sandbox and attach the result to that change. The agent’s own assessment does not count as test evidence. Teammates can approve or reject a change, and a rejection keeps the review gate closed. The guided demo uses a synthetic payment fixture about duplicate charges on retry. It focuses on a common problem with AI coding tools: explanations can sound convincing, but teams need to inspect the actual change and know what was tested. The current demo has verified live collaboration between two independent browser sessions, chat that persists across reloads, and replay export. Nebius model and sandbox access is still being completed, so end-to-end provider execution has not yet been verified.