"Fake Done" vs "Undefined Done": The /goal Primitive for Agent Finish Lines
Summary
The AI Maker piece frames the core failure modes of AI agents: "fake done" (agent claims completion without verifying) and "undefined done" (agent stops because it ran out of instructions, not because it finished). The /goal primitive solves both by giving agents a verifiable finish line rather than just a list of instructions. Also introduces the "Ralph Wiggum loop" pattern — shipped by Anthropic as a plugin — where agents loop until a defined success condition is met.
Why this signal
Directly addresses Neo's #1 focus area: agent reliability. The "fake done" / "undefined done" framing is a crisp mental model for why agents fail silently — useful for team conversations about when to trust agent output and how to structure prompts for complex tasks.
Original email
AI Maker Wyndo · 6/4/2026
How I Use /goal To Stop Babysitting AI Agents