AI Agent Development and Security Tools Gain Momentum

New Relic announced the general availability of AI observability features aimed at reducing what it calls “agent debt,” the operational risk that arises when enterprises ship AI‑generated code without line‑by‑line verification. The company highlighted that while 94% of leaders rate AI‑generated code as higher quality, 78% see more incidents post‑deployment, prompting the need for monitoring tools that integrate telemetry into IDEs and production environments.

A panel hosted by OpenInfra Live discussed the growing security challenges of AI agents, noting that traditional container models are ill‑suited for non‑deterministic workloads. Experts argued that hardware‑virtualized isolation, such as Kata Containers, provides a hard boundary against kernel‑sharing attacks and memory‑persistence threats, recommending sandboxing at the infrastructure level.

The AI SDK 7 release standardizes agent development across providers by introducing typed tool context, runtime context, file/skill uploads, and Model Context Protocol (MCP) apps. It also adds production‑ready features like durable execution, approval gates, and multimodal support, though it now requires Node 22+ and is ESM‑only. These updates aim to simplify building, securing, and observing AI agents in production.