production agents need to feel responsive in real time. two primitives make that possible: streaming and concurrent request control. on stre…

production agents need to feel responsive in real time. two primitives make that possible: streaming and concurrent request control. on streaming: streaming is a first-class primitive in LangSmith's agent server. partial output flows to the client as the agent produces it: token by token, step by step, or as custom application events. on concurrency: in a live session, users don't wait politely for a response before sending the next message. we call this double-texting, and your runtime needs an explicit position on what to do: 1. enqueue: finish the current run, process the new message after. safe default. 2. reject: block new input until the current run finishes. 3. interrupt: halt mid-run, preserve state, resume with the new input. right when the second message builds on the first. 4. rollback: halt mid-run, revert all progress, start fresh. right when the second message replaces the first. concurrent request docs: https://docs.langchain.com/langsmith/double-texting full breakdown of the runtime: https://www.langchain.com/blog/runtime-behind-production-deep-agents

Source: Harrison Chase (X) | 2026-05-04