Microsoft: agents as infrastructure

Microsoft is pitching AI agents as managed infrastructure, not prompt scripts, in a new Foundry Agent Service explainer published on YouTube and backed by fresh Azure documentation. (youtube.com) (learn.microsoft.com) In Microsoft’s docs, Foundry Agent Service is a “fully managed platform” for building, deploying, and scaling agents, with support for “any framework” and models from the Foundry catalog. The same material describes an agent as a persisted, versioned orchestration definition that combines models, instructions, code, tools, parameters, and optional governance controls. (learn.microsoft.com 1) (learn.microsoft.com 2) The video frames the product around production controls rather than model novelty: local deployment into the service, secure identity, scoped permissions, traceable actions, and monitoring for “every interaction.” Azure’s product page uses similar language, promising agents that are “governed, observable, and integrated” at scale. (youtube.com) (azure.microsoft.com) An AI agent is software that calls a model, decides when to use tools, and keeps track of work across steps. Microsoft’s documentation turns those moving parts into platform features: orchestration definitions, tool catalogs, hosted runtimes, memory stores, and monitoring dashboards. (learn.microsoft.com 1) (learn.microsoft.com 2) (learn.microsoft.com 3) That packaging reflects a wider shift in enterprise AI spending from chatbot demos to operating systems for agents. Microsoft Foundry’s top-level documentation now describes the product as an “AI app and agent factory” for building, optimizing, and governing apps and agents at scale. (learn.microsoft.com) Memory is one example of that shift from prompt engineering to infrastructure. Microsoft introduced memory in Foundry Agent Service as a public preview at Ignite 2025, and current docs describe it as a managed long-term store that can retain preferences and conversation history across sessions, devices, and workflows. (devblogs.microsoft.com) (learn.microsoft.com) Hosting is another. Microsoft’s hosted-agents documentation says developers can deploy containerized agents with custom code or a preferred framework while the service handles managed hosting, scaling, security, and observability. (learn.microsoft.com) Observability is a third pillar. Microsoft Foundry’s observability docs say the platform tracks latency, token consumption, error rates, quality scores, and agent-specific measures such as tool-call accuracy and task completion, with dashboards built on Azure Monitor Application Insights. (learn.microsoft.com 1) (learn.microsoft.com 2) Microsoft is also leaving room for developer choice while steering users toward its own control plane. The overview page says Foundry Agent Service works with many models and frameworks, but the service wraps them in Microsoft-managed identity, deployment, memory, and monitoring primitives. (learn.microsoft.com) (learn.microsoft.com) The result is a clearer product boundary: the model is one component, while the sellable system is the managed stack around it. Microsoft’s latest Foundry materials describe that stack in concrete terms — orchestration, tools, memory, hosting, and governance — and present agents as cloud infrastructure that enterprises can buy, not just prompts they can write. (learn.microsoft.com) (azure.microsoft.com)