LangChain introduces DeltaChannel to reduce memory use for long-running agents

LangChain released DeltaChannel in LangGraph 1.2, a new primitive for reducing memory use in long-running agents. Previously, the full state was stored at every step, leading to exponential growth. Now the system tracks only the changes (delta) between steps and periodically writes full state snapshots. As a result, storage grows linearly, and Deep Agents v0.6 uses it by default without reconfiguration.