Uploaded July 2026 | Updated September 2026, 3 hours ago
The Ralph Wiggum loop is a deceptively simple idea: point an agentic coding pipeline at a plan and let it cycle, implement, review and fix on repeat, with no hands on the keyboard. In this talk I'll walk through three real runs of Ralph on my own codebase, from a 24 minute one-shot that shipped subtle bugs to a two and a half hour run where the agents actually caught and fixed their own mistakes. You'll see what made the difference (spoiler: it wasn't the loop), and why the human still isn't going anywhere just yet.
The Ralph Wiggum loop is a deceptively simple idea: point an agentic coding pipeline at a plan and let it cycle, implement, review and fix on repeat, with no hands on the keyboard. In this talk I'll walk through three real runs of Ralph on my own codebase, from a 24 minute one-shot that shipped subtle bugs to a two and a half hour run where the agents actually caught and fixed their own mistakes. You'll see what made the difference (spoiler: it wasn't the loop), and why the human still isn't going anywhere just yet.
