Back to glossary

Agent orchestration

Ajan orkestrasyonu

D3

Agent orchestration is the coordination of which agents run in an application, in what order, and how their next actions are chosen.

Review status: 2026-12-04

Technical explanation

The flow can be determined by application code or by an LLM. In documented multi-agent patterns, one agent can call another as a tool or hand control to another agent.

Conceptual boundaries

Orchestration is workflow coordination, not a proof of autonomy, reliability, domain competence, or correct delegation.

Provider-neutral example

A research workflow routes a retrieval request to a search tool, asks a model to summarize returned evidence, logs the trace, and sends high-impact claims to human review.

Limitations

Prompt injection through user input or retrieved content can redirect a generative AI system’s behavior; coordinating agents does not itself establish protection against that risk.

Related concepts

Atomic claims and evidence

  1. 1.1OpenAI Agents SDK, Agent Orchestration
    Source
    OpenAI Agents SDK, Agent Orchestration
    Source role
    Authoritative source
    Exact locator
    Overview: flow of agents and next-action decisions
    Supported claim
    OpenAI defines agent orchestration as the flow of agents in an application, including which agents run, in what order, and how the next action is decided.
    Last verification
    Review due
    Scope limitation
    This is an SDK-specific workflow-design description, not evidence that an orchestrated system is autonomous, reliable, competent, or correct.
  2. 2.1NIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
    Source
    NIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
    Source role
    Authoritative source
    Exact locator
    Printed pp. 13-14: direct and indirect prompt injection
    Supported claim
    NIST identifies prompt injection as an attack in which input is modified so a generative AI system behaves in unintended ways, including indirect attacks through retrieved data.
    Last verification
    Review due
    Scope limitation
    This is a threat description; it does not imply that every orchestrated workflow is compromised.
    2.2OWASP GenAI Security Project, LLM01:2025 Prompt Injection
    Source
    OWASP GenAI Security Project, LLM01:2025 Prompt Injection
    Source role
    Authoritative source
    Exact locator
    LLM01:2025 Prompt Injection risk overview
    Supported claim
    NIST identifies prompt injection as an attack in which input is modified so a generative AI system behaves in unintended ways, including indirect attacks through retrieved data.
    Last verification
    Review due
    Scope limitation
    This is a threat description; it does not imply that every orchestrated workflow is compromised.
  3. 3.1OWASP GenAI Security Project, LLM01:2025 Prompt Injection
    Source
    OWASP GenAI Security Project, LLM01:2025 Prompt Injection
    Source role
    Authoritative source
    Exact locator
    LLM01:2025 Prompt Injection risk overview
    Supported claim
    OWASP's 2025 prompt-injection risk entry independently treats prompt injection as a material LLM-application security risk.
    Last verification
    Review due
    Scope limitation
    OWASP risk guidance is not a probability estimate for a particular system or configuration.