# Agentic Workflow **Domain:** Artificial Intelligence / Orchestration **Doc Type:** Concept Node **Maturity:** Developed ## Definition An **agentic workflow** coordinates model calls, tools, data, checks, and state across multiple steps. The path may be fixed in code, selected dynamically by a model, or hybrid. The useful distinction is control allocation. In a workflow, the designer determines the sequence or available branches in advance. In an [[wiki/AI Agent|AI agent]], the model determines which actions and tools are needed as work proceeds. Both are forms of [[wiki/Agentic AI|agentic AI]], and production systems often combine them: a model chooses reversible actions while deterministic code enforces permissions, approvals, budgets, and completion checks. ## Common Patterns - **prompt chaining:** one model step transforms the result of another; - **routing:** a classifier or model selects a specialized path; - **parallelization:** independent subtasks run concurrently and are aggregated; - **orchestrator–worker:** a coordinating model creates and assigns subtasks dynamically; - **evaluator–optimizer:** one component produces while another evaluates and returns feedback; - **agent loop:** the model chooses tools and next steps from environmental feedback until completion or a stop condition. ## XDO Context XDO provides a deterministic document-native version of this pattern: commands operate on scoped state, call registered capabilities, select projections, and send results to named targets. A modern model can propose the plan while the XDO-style runtime validates and executes it. ## Relationships - **parent category:** [[wiki/Agentic AI|Agentic AI]]. - **dynamic actor:** [[wiki/AI Agent|AI Agent]]. - **state:** [[wiki/Agentic Memory|Agentic Memory]]. - **tool boundary:** [[wiki/Tool Contract|Tool Contract]] and [[wiki/Model Context Protocol|Model Context Protocol]]. - **execution:** [[wiki/Durable Execution|Durable Execution]]. - **human control:** [[wiki/Approval Architecture|Approval Architecture]]. ## Sources / Provenance - [Anthropic — Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) - [[projects/Ten Years Building a Symbolic Language Engine|Ten Years Building a Symbolic Language Engine]] **As of:** 2026-09-28