Recent LangSmith updates point to a broader shift in AI agent engineering: model selection is giving way to evaluation, issue detection, pre-release testing, and operational control. Task environments, agent harnesses, and correctable memory are also becoming part of the production stack.