Anthropic's Claude Code lead says a single agent run rewriting the Electron Claude desktop app in Swift has been going about 15 days and is still running

Sources: Daring Fireball link post, HN discussion
Summary: John Gruber relays a statement by Boris Cherny, Anthropic's Claude Code lead, that a single agent run rewriting the Electron Claude desktop application in Swift has been going about 15 days and is still running. The quoted prompt builds its own verification loop, running the Electron app in a macOS runner, taking screenshots, and comparing them pixel by pixel against the Swift build. Cherny names verification as the single thing practitioners most often get wrong. No shipped artifact and no code from the run has been published.
Why it matters: The claim of interest is not the run length but the verification design, because a run of that duration only means anything if something other than the agent decides whether the output matches.