opencode-subagents

You can run parallel AI agents by creating specialized subagents that execute concurrently on distinct tasks. This architecture enables modular AI development by isolating agent context and skills, ensuring focused operations without context pollution.

Agent context pollution occurs when a single AI session accumulates irrelevant information, degrading performance. Isolated subagents prevent it by operating with clean context, maintaining their own independent tools and skills separate from the parent session.

You configure specialized agents by defining their specific expertise, configuring their available tools, and setting model preferences for targeted operations. This allows domain-specific task execution like code review or testing.

Yes, you can assign different tools and models to individual AI agents. Subagents maintain their own tools and skills independently, allowing you to configure specific model preferences and toolsets for targeted operations.

The best way to orchestrate multiple agents for complex code refactoring is invoking specialized subagents in parallel. You can run a code review specialist and a testing agent concurrently, ensuring clean context and skill isolation for each task.

You should not use subagents for AI development orchestration when tasks require shared session context or heavy inter-agent communication, as subagents are designed for isolated, parallel execution with clean, separated contexts.