batch

You can reduce token consumption in multi-agent orchestration by grouping tasks into batches of 3-5 agents. This approach uses dedicated context windows per batch and executes sequentially to maintain cost-effective workflow efficiency.

Multi-agent batching is an orchestration technique that groups AI tasks into 3-5 agent batches executed sequentially with parallel execution within each batch. You should use it when coordinating complex workflows like full-stack feature development across frontend, backend, and tests.

You can orchestrate full-stack features by analyzing task complexity, planning batches of 3-5 agents with dedicated context windows, and executing them sequentially. The orchestrator synthesizes batch results and writes final outputs to work/batch1-results.md.

Yes, agents within a batch run in parallel while batches themselves execute sequentially. This batching approach maintains isolated context windows per batch and synthesizes results into a final output after all planned batches complete execution.

You should group tasks into batches of 3-5 agents for efficient multi-agent batching. This batch size allows parallel execution within each batch while maintaining dedicated context windows and sequential batch processing to optimize token efficiency.

No external dependencies are required to run batched multi-agent orchestration. You can start the batching process by running a task with the /batch command, which lets the orchestrator analyze task complexity and execute the first batch automatically.