Back to AI/ML Overview
🎛️ Knowledge Site · Orchestration for Agentic Software · The Control Plane

Agent Orchestration EngineeringSomething has to decide what runs next.

A model call answers one question. Real work needs many steps: some in order, some at the same time, some waiting for a person, some repeated after a failure. This site starts from sequence, branch and loop and climbs, one idea at a time, to durable workflows with several agents and a human approval, asking one question at each step: who decides what runs next here, code or a model?

🎯The problem

An underwriting quote needs intake, a roof assessment, retrieval of the policy, a recommendation and an underwriter's approval. Hand all of it to one agent loop and nobody can say why a step ran twice or never ran. Hard-code all of it and the flexible part of the work has nowhere to go. Orchestration is the design of that choice, step by step.

🔑The question every page asks
Who decides what runs next here, code or a model? Each chapter compares a fixed workflow with bounded autonomy at the steps where it earns its place, on the same synthetic insurance quote.

🗺️What is inside

Workflows, state machines and loops

Sequence, branch and loop as primitives; , events and agent loops; one agent with tools against several specialized agents.

Patterns and the

Supervisor, planner and executor, router, peer patterns; against ; queues, fan-out and fan-in.

People, durability and limits

Human approval with suspend and resume, durable workflows with checkpoints and compensation, budgets, permissions and termination.

Failure, tracing and testing

How failures propagate, breakers and fallback routes, trace propagation across agents, and tests that cover every state transition.

🖱️How to read it

  • New to orchestration: read in order. Each chapter starts in plain English with one tiny example.
  • Experienced: jump to durable workflows, human approval or the insurance lab. The rigorous part sits in folded blocks.
  • Thirteen interactive labs, each with a plain table next to it. Every diagram is clickable.
  • The example is synthetic and its agents are scripted stand-ins. Results show how the mechanism works, not how good any model is.