Closing the loop from emergent flow to explicit prompt graph

Determine whether the explicit-versus-emergent structure trade-off in prompt graph engineering is fundamental or can be resolved technologically by constructing systems that record emergent flows as execution traces, lift them into explicit prompt graphs, and support replay and refinement so that emergent flows become versioned, optimizable artifacts.

Background

The paper identifies a core design tension between explicit, authored graphs (as in graph-first frameworks) and emergent, runtime-decided flows (as in free multi-agent conversations or coding harness delegation). Explicit graphs enable inspection, validation, and optimization, while emergent flows provide adaptivity and unanticipated delegation. The authors ask whether these advantages can be reconciled by lifting emergent behavior into explicit artifacts.

They note that dynamic DAG construction (e.g., LLMCompiler) suggests this reconciliation may be possible, yet they are not aware of any system that fully closes the loop from execution trace to a persistent, versioned, and optimizable graph artifact.

References

The open question is whether the trade is fundamental or technological: can a system record the emergent flow, lift it into an explicit graph, and replay or refine it, making emergence a discovery mode for structures that then become artifacts? The dynamic DAG construction of LLMCompiler suggests the two ends can meet , but, to our knowledge, no system today closes the loop from trace to versioned, optimized graph.

— What makes prompts a graph: necessary and sufficient conditions for prompt graph engineering  (2607.27578 - Macedo, 30 Jul 2026) in Section 7 (Research Agenda), first axis: explicit vs emergent structure

EQ5 --- Decision artifact utility. Can the decision artifact support after-the-fact auditing, debugging, and policy refinement? Qualitative evaluation should assess whether practitioners can (a) determine from the artifact alone why a given candidate was selected or rejected, (b) replay the decision under a modified policy without re-running execution, and (c) detect policy conflicts from artifact analysis. This question evaluates the observability and governance value of the decision layer independently of its routing quality.

— Brain API: An Intent-Aware Control Plane for Policy-Governed Agentic Systems  (2609.21299 - Chernov, 18 Sep 2026) in Section 12.7, Evaluation Criteria (EQ5)