A hobbyist revived a half‑finished YouTube‑Music shim by letting Claude Code write the bulk of the service. The project, dubbed “Sub‑standard,” exposes YouTube tracks through the OpenSubsonic API so any Subsonic client—Navidrome, Feishin, Symfonium—can stream them. After a manual proof‑of‑concept, the author seeded a fresh FastAPI repo, added ytmusicapi and yt‑dlp, then handed the spec to the AI.

Claude received $50 in free credits, which the author used to run Opus 4.6 and generate code iteratively. Starting with a plan mode prompt, the model stubbed all JSON endpoints from the OpenAPI file, then refined them after each test failure. Errors like missing .view suffixes were fixed by feeding request logs back to Claude, and unit tests were added on the fly.

The final MVP supports license, user, genre and directory listings, a pass‑through getSong, basic search, streaming via yt‑dlp, and cover‑art retrieval. In‑memory caching and a SQLite cache keep API calls cheap, while streamed files are saved to disk to avoid re‑downloads. The author stresses that such “bucket‑two” projects prove AI can un‑brick abandoned ideas without replacing the need for deeper engineering practice successfully.