SAGE: Stochastic Prompt Optimization via Agent-Guided Exploration

2026-06-17

AI Summary

This work addresses the challenge of automatic prompt optimization (APO) in open-domain task-oriented dialogue, where stable black-box methods are lacking. The authors propose the SPO framework, formulating prompt optimization as a black-box search problem and introducing three strategies: error-informed random search, genetic algorithms, and a novel multi-agent guided exploration approach termed SAGE. SAGE innovatively integrates multi-agent systems with diagnostic code execution, leveraging a sequential optimization paradigm to consolidate noisy A/B testing results into robust performance gains. Experimental evaluation across three benchmark tasks demonstrates the efficacy of the proposed methods. Notably, in a real-world deployment involving a mental health chatbot, SAGE achieved a significant improvement in next-day retention after eight rounds of A/B testing, underscoring its practical utility in authentic conversational AI applications.

Abstract

Context engineering has emerged as a primary lever for improving AI systems without parameter updates. Recent work showing that textual gradients do not function as real gradients motivates treating automatic prompt optimization (APO) as black-box search. We introduce SPO (Stochastic Prompt Optimization), a framework for stochastic search over prompt space, and compare three strategies of increasing sophistication: error-informed random search, a genetic algorithm with evolutionary operators, and SAGE (SPO via Agent-Guided Exploration), a multi-agent pipeline with diagnostic code execution. Across three benchmarks, no single strategy dominates; effectiveness depends on the interaction of landscape structure with error type. We further deploy SAGE on a mental-health chatbot under a continuous optimization paradigm, where it compounds eight cycles of individually-noisy A/B tests into a statistically robust gain in next-day retention. We argue that coupling qualitative diagnosis with quantitative validation is what makes agentic optimization effective for open-ended task-oriented dialogue.

Problem

• prompt optimization
• black-box search
• task-oriented dialogue
• stochastic optimization
• context engineering

Innovation

• Stochastic Prompt Optimization
• Agent-Guided Exploration
• Automatic Prompt Optimization
• Multi-agent Pipeline
• Black-box Search