The Short Leash AI Coding Method for Beating Fable
The Short Leash AI method advocates for tightly controlling AI coding outputs through detailed specs, incremental steps, and frequent human oversight to reduce errors and improve reliability in complex projects like game development.
Background
- **Fable** is an AI coding agent (similar to Cursor, Devin, or GitHub Copilot) that autonomously writes and edits code from natural-language prompts. It was released by the startup Fable AI in 2025 and quickly became a standard benchmark in the AI-assisted programming space.
- **"Short Leash"** refers to a workflow where the human developer keeps tight, step-by-step control over the AI — approving or rejecting each change, writing very specific prompts, and never letting the agent run fully autonomously. This contrasts with the "long leash" approach where you give the AI a high-level goal and let it figure out the implementation.
- The article's author (from the okturtles blog, a small independent software consultancy) argues that the Short Leash method consistently outperforms letting Fable run free, especially on complex or nuanced codebases where the AI's assumptions can introduce subtle bugs.
- This debate mirrors a broader tension in AI-assisted development: maximum autonomy vs. tight human oversight. The post contributes practical, opinionated advice from a team that claims to have beaten Fable's own advertised success rates by using this method.