In a multi-agent Copilot Studio setup using generative orchestration, how does the parent agent determine when to route a user's message to a child agent versus handling it internally?

Generative orchestration is an AI-driven routing mode in Copilot Studio where the parent agent uses a large language model to reason about where each user message should be handled. Each child agent is registered with a name and description of its capabilities. When a user message arrives, the orchestrator evaluates: 'Does this message's intent best match one of my registered child agents, or should I handle it with my own topics and actions?' The correct approach is the one where the AI orchestrator evaluates the user's intent against the description and capabilities of each registered child agent, routing to the best match. This semantic matching is far more flexible than keyword triggers — it can route 'my laptop won't connect to the VPN' to an IT child agent even if the exact phrase 'IT Helpdesk' never appeared in the user's message, because the orchestrator understands the intent. The keyword-matching approach, which uses the child agent's name to decide routing, describes classic trigger-phrase routing — not how generative orchestration works. The round-robin routing approach makes no sense for a Q&A agent — routing should be intent-based, not arbitrary. The explicit 'Redirect to agent' node approach, which requires adding that node in every relevant topic, describes manual routing — also called classic multi-topic orchestration. Generative orchestration replaces this manual wiring with AI-driven intent matching. Exam caveat: Child agent description quality matters: write clear, specific capability descriptions for each child agent so the orchestrator can make accurate routing decisions. Vague descriptions lead to misrouting.