Today's AI trends
AWS published a multi-account MCP reference architecture in which line-of-business accounts retain their data and selected MCP tools while a central AgentCore Gateway provides discovery, authentication, authorization, and observability. It matters because it offers a concrete alternative to either centralizing sensitive data or granting a single agent broad cross-account access. Teams designing agent platforms should model user identity, gateway identity, and local service authorization as separate concerns from the beginning. AWS’s reference architecture
Databricks introduced Unity Gateway CLI, which it says centrally configures approved coding agents, models, MCP servers, Skills, Smart Routing, and spending policies while developers launch supported agents through ug. The company also describes cohort rollouts, budget-aware session defaults, and centralized trace export; its reported 35% Smart Routing cost saving is an internal evaluation, not an independent study. The practical implication is to separate developers’ agent frontends from centrally managed policy, routing, spend controls, and telemetry. Databricks’ announcement
Langfuse v4.45.2 adds batch-inference telemetry per project and retries for transient failures in its AI gateway path. This matters because, without first-class telemetry, teams cannot reliably diagnose failures, retries, and Evaluation runs in asynchronous batch jobs. Teams should enable project-level telemetry for batch jobs and test retry rules for transient failures before production use. Langfuse v4.45.2