LangSmith shifted focus toward agent observability and collaboration tooling. The platform expanded baseline comparison to persist across experiment runs rather than require manual selection each time, and introduced unified cost tracking across the entire agent stack—moving beyond just LLM call accounting. Agent Builder (renamed to LangSmith Fleet in March) added collaborative features like a central chat interface, file uploads, and a shared tool registry, while the Insights Agent graduated to scheduled execution on custom cadences, eliminating manual report triggers. Deep Agents v0.4 shipped pluggable sandbox support and smarter conversation history summarization as the foundation for more complex agentic workflows.