What happens when your coding agents can delegate to each other?

A developer is experimenting with multiple AI coding agents, each specializing in tasks like understanding, researching, implementing, reviewing, and debugging. The core challenge lies in how these agents communicate and delegate to each other, rather than their individual functionalities. The developer is exploring various architectural approaches, including hardcoding relationships, using a central orchestrator, exposing agents as tools, employing A2A, or enabling dynamic discovery among agents, and seeks insights from others who have built similar systems.

This report uniquely focuses on the inter-agent communication and delegation challenges, unlike others that primarily address individual AI agent capabilities.