Multi-View Encoders for Performance Prediction in LLM-Based Agentic Workflows

Researchers developed Agentic Predictor, a lightweight AI system that uses multi-view encoding to optimize LLM-based agent workflows without expensive trial-and-error evaluations. The system incorporates code architecture, textual prompts, and interaction graphs to predict task success rates and select optimal configurations across different domains.

• →Agentic Predictor reduces computational costs of optimizing LLM-based agent systems through predictive modeling.
• →The system uses multi-view representation learning incorporating code, prompts, and interaction patterns.
• →Cross-domain unsupervised pretraining enables high accuracy with fewer training evaluations required.
• →Benchmark testing across three domains shows superior performance versus graph-based baseline methods.
• →The approach addresses the challenge of vast search spaces in agent configuration optimization.