Astra for prose, Fable for long-horizon work: a developer's multi-model division of labor, with Pi as best harness
seatedro · x · 2026-09-21
The author shares lessons on maximizing token value: Astra is far better at prose and human communication (raw behavior and instruction-following) but loses the plot on long-horizon research/programming, while Fable excels at long-horizon work but can't follow style guides. Their setup: Astra for comms, Fable to drive research, cheaper models for grunt work — with Pi as the best harness for staying close to raw model capability.
More from coding & agent
- jev-skill-suggester: a lightweight pre-execution skill router stops agents picking wrong tools — udmrzn · 2026-09-21
- OpenClaw adds FaceTime: your agent can now call you, and answer your calls — steipete · 2026-09-21
- Reverse-engineering Instinct's memory: beautifully simple, replicable in 60 lines — julianweisser · 2026-09-21
- 3 Weeks of Logs Show 40% of Messages to Coding Agents Are Overhead — Lucky-Group9525 · 2026-09-21
- Humanizer Academic: a one-file skill that de-AI-fies academic medical papers — udmrzn · 2026-09-21
- Build a 100% local decision engine from an open-source LLM, no retraining needed — udmrzn · 2026-09-21