Agent infrastructure: plugins, stateless MCP, evals, context, cost

Agent Plugins package your skills, tools, and more announces Agent Plugins 1.0.0, a portable plugin format for packaging agent skills, tools, and MCP servers for cross-vendor use. This matters because outcome engineers can now treat skills as reusable, versioned artifacts that travel between IDEs and runtimes, simplifying the Graph and Map work of composing reliable agent behavior.

Scaling AI Agent Infrastructure with the MCP Stateless Updates publishes a stateless MCP spec and SDKs that replace stateful agent servers with cloud-native, horizontally scalable primitives and MRTR for long-running tasks. This shifts operational design: you can build serverless agent fleets, reduce state management complexity, and adopt CI/CD-friendly deployment patterns that map to principle-driven orchestration.

Agent and Model Evaluations in Gemini Enterprise Agent Platform are now GA launches a GA evaluation service with unified metrics, adaptive rubrics, and CI-integrated testing for agents and models. Outcome engineers get a production-ready evaluation loop to catch regressions and automate quality gates, supporting Validation (audit the outcomes) and an immune-system approach to reliability.

Agent context layers: Enterprises governing their AI data are catching twice as many bad answers as the ones who aren’t reports that companies with governed semantic/context layers detect twice as many confident-but-wrong agent answers. That demonstrates a practical lever for reducing hallucinations: invest in retrieval architecture and context engineering so agents operate on auditable, curated inputs rather than brittle, ad-hoc context.

Agentic orchestration: Enterprises govern agents but can’t meter costs finds organizations running multiple orchestration platforms and governance controls yet lacking real-time cost metering to prevent runaway spending. For outcome engineering this is a warning: governance without metering lets agent-driven processes blow budgets, so build cost controls and observability into your orchestration and Gate design.