Position paper from 30 researchers: agentic AI orchestration must be Bayes-consistent — tool-routing as a decision-theoretic problem
A position paper signed by 30 researchers across industry and academia argued on May 4 that the "orchestration layer" in agentic AI systems — the component that decides which tool to call, when to escalate to a stronger model, and how much resource to invest — should be Bayes-consistent rather than heuristic. The argument lands in a moment when MCP 2.0 has just standardized the cross-runtime tool-call schema and orchestrators are becoming the active design surface.
The technical claim is that orchestration is a sequential decision-theoretic problem and that the heuristic orchestrators most production agent frameworks ship today produce systematically suboptimal action sequences. The paper proposes Bayes-consistent orchestration as a principled formulation — one that explicitly tracks posterior beliefs over tool outcomes and routes accordingly.
The political subtext is interesting: 30 named co-authors is a lot, and they cluster around teams at the model labs, agent-framework startups, and academic labs working on agentic decision theory. The signal is that the field has consensus on the framing — even if implementations vary. For agent runtimes looking to differentiate on routing intelligence rather than model raw capability, Bayes-consistent orchestration is now the language that frames the conversation.
DevFlokers — AI News May 2026 Models Papers Code → · DevFlokers — AI Tech Breakthroughs May 2026 → · Harness — Daily AI Agent News Roundup May 8 2026 →