Proforma Global

How orchestration of enterprise agent systems should be designed. Where language-model-driven orchestration fails, the five canonical patterns, and the deterministic-first discipline required when agents touch financial systems.

What the series argues

Orchestration is the decision about what runs next. Most agent platforms hand that decision to a language model, which is the wrong place for it. A model asked to govern a multi-discipline enterprise workflow is being asked to be reliable at the thing it is worst at.

The first paper sets out where that breaks and the four-layer split that replaces it. The second names the five orchestration patterns available, gives a four-axis framework for choosing between them, and identifies the two defaults the industry reaches for by reflex. The third narrows to financial systems, where rules verify what they can, the model judges only the gaps, and the risk rating of an action is a property of the architecture rather than a setting someone toggles later.

Together they describe an orchestration layer you can audit.

Papers

Listed in reading order.

  1. 01 Where Agent Orchestration Breaks Why language models should not orchestrate multi-discipline enterprise workflows, and the four-layer architectural split that does.
  2. 02 The Five Orchestration Patterns The five distinct orchestration patterns for agent systems, the four-axis framework for selecting among them, and the industry's two default mistakes.
  3. 03 Enterprise Agents in Financial Systems Why business agents that touch money have to be deterministic-first: rules verify what they can, the model judges only the gaps, and risk rating is an architectural property.