AutoDSPy

Overview

AutoDSPy is a 2025 system that frames DSPy pipeline construction as a reinforcement-learning problem. Its policy selects reasoning modules, signatures, and execution strategies, so the optimized object is a modular LM program rather than only a prompt string.

Why it matters

It matters because it shifts the RL target from prompt text toward workflow structure. For long-lived harnesses, that is much closer to the real object of control: reusable, inspectable program artifacts that can be evaluated, promoted, and rolled back.

Distinctive trait

Its distinctive trait is RL over DSPy program structure and configuration rather than RL over discrete prompt tokens.

Relationships

Read AutoDSPy with dspy, dspy-assertions, rlprompt, and prompt-optimizer-regimes-for-harnesses. It is a useful bridge between flat prompt optimization and the more general problem of optimizing modular language workflows.