ArXiv position paper argues agentic AI orchestration should be Bayes-consistent — control layer must maintain calibrated beliefs over task-relevant quantities
A position paper submitted to arXiv on May 4, titled "Agentic AI Orchestration Should be Bayes-consistent," argues that the control layer of an agentic system must be grounded in Bayesian principles. The authors contend that the orchestration layer — the system that decides which tools to call — should be Bayes-consistent, enabling systems to maintain calibrated beliefs over task-relevant quantities. The framing is the most explicit recent attempt to give agent-orchestration a principled theoretical foundation.
The Bayes-consistency framing is the substantive theoretical piece. Through 2023-2025 the dominant agent-orchestration pattern was empirical: the orchestration layer (whether implemented as LangGraph-style stateful flows, ReAct-style reasoning-acting loops, or proprietary frameworks) optimized for downstream task success without an explicit theory of how decisions about tool selection should be made. The Bayes-consistent framing argues that orchestration decisions should reflect calibrated beliefs over the relevant task quantities — meaning the system maintains explicit probability distributions over which tool will succeed, which action will produce the desired state change, and what the residual uncertainty over task outcome is.
The deployment-practice consequence is what connects the theory to current agent work. Devin's autonomous-engineer deployment at $492M ARR operates the orchestration layer at production scale; Cursor Composer 2.5's parallel-agent stack operates a different orchestration pattern at the pair-programmer end. Neither implementation today is explicitly Bayes-consistent — the orchestration decisions emerge from learned policies and engineered heuristics rather than principled probability calibration. The position paper's contribution is making the gap between deployed practice and principled theory legible, which is the precondition for the next research cycle's work on closing it.
ArXiv cs.AI — Agentic AI Orchestration Should be Bayes-consistent May 4 → · DevFlokers — AI News May 3-4 2026 Models Papers Code → · DevFlokers — AI Tech Breakthroughs May 3-4 2026 →