aiqbench analysis
This research introduces GTBP, a novel framework for automating prompt engineering in multi-LLM agentic systems by modeling workflows as directed acyclic graphs. The method addresses critical issues like inaccurate credit assignment and lack of convergence guarantees through backward propagation of target outputs to guide stage-wise updates without modifying model weights. Empirical results demonstrate that this approach consistently outperforms existing baselines while maintaining comparable computational costs, offering a robust solution for scaling complex agentic workflows.
Graph-based Target Back-Propagation for Context Adaptation in Multi-LLM Agentic Systems
arXiv:2606.14155v1 Announce Type: new Abstract: Context adaptation automates prompt engineering in LLM-based systems by iteratively revising tunable prompts from task feedback, without modifying model weights. Extending this paradigm to multi-LLM agentic systems is crucial: existing methods suffer from inaccurate credit assignment and lack convergence guarantees. We propose \textbf{G}raph-based \textbf{T}arget \textbf{B}ack-\textbf{P}ropagation (GTBP), a context adaptation framework for agentic workflows modeled as directed acyclic graphs. GTBP propagates local target outputs backward through the workflow graph and uses target--output discrepancies to guide a stage-wise prompt update mechanism. Theoretically, we show that GTBP's stage-wise prompt updates become stable over iterations, and that a sufficiently capable LLM optimizer can decrease the overall objective. Empirically, GTBP consistently outperforms strong baselines across three benchmarks while maintaining comparable computational cost.
Abstract:Context adaptation automates prompt engineering in LLM-based systems by iteratively revising tunable prompts from task feedback, without modifying model weights. Extending this paradigm to multi-LLM agentic systems is crucial: existing methods suffer from inaccurate credit assignment and lack convergence guarantees. We propose \textbf{G}raph-based \textbf{T}arget \textbf{B}ack-\textbf{P}ropagation (GTBP), a context adaptation framework for agentic workflows modeled as directed acyclic graphs. GTBP propagates local target outputs backward through the workflow graph and uses target--output discrepancies to guide a stage-wise prompt update mechanism. Theoretically, we show that GTBP's stage-wise prompt updates become stable over iterations, and that a sufficiently capable LLM optimizer can decrease the overall objective. Empirically, GTBP consistently outperforms strong baselines across three benchmarks while maintaining comparable computational cost.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2606.14155 [cs.LG]
(or arXiv:2606.14155v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.14155 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 215 words
Related coverage
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
AgentOps: Operationalize agentic AI at scale with Amazon Bedrock AgentCore
AWS · Jun 1, 2026
From Vector Search to Agentic RAG: Building an Enterprise Research Analyst with hipVS
AMD · Jul 15, 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · Oct 2, 2026