LangChain: async subagents bring fire-and-steer paradigm for hundreds of parallel AI agents

LangChain released a new async subagent model on April 16, 2026 that fundamentally changes how AI agents coordinate work. Instead of the traditional sequential approach where a supervisor agent waits for each subagent to finish, the new model introduces asynchronous execution and the ability to intervene in real time.

How does fire-and-steer work?

The paradigm relies on five key tools: start_async_task for launching a subagent, check_async_task for checking status, update_async_task for changing instructions, cancel_async_task for cancellation and list_async_tasks for viewing all active tasks.

The key innovation is the fire-and-steer approach: the supervisor agent launches a subagent, continues with other work, and can come back and change the instructions of a subagent that is still running. This is fundamentally different from the fire-and-forget model because the supervisor retains control.

Why is this important for agentic systems?

Sequential agent systems have an inherent limitation: the total execution time is the sum of the time for all substeps. The async model enables parallelization across hundreds to thousands of subagent instances, dramatically shortening total time for complex tasks.

The system works on the LangSmith platform or self-hosted infrastructure via Agent Protocol, making it available for both cloud and on-premise implementations. LangChain positions this as the foundation for the next generation of enterprise agentic systems where a single orchestrator agent manages a fleet of specialized workers.