Microsoft Foundry Assembled: Agents, Tools, Skills, Knowledge & A2A

Microsoft Foundry gives you a set of building blocks for agents, and the power comes from assembling them. This developer session builds that story end to end — and assembles a working multi-agent orchestration entirely through the Foundry UI, one live demo building on the last.

We start with prompt agents: an agent defined entirely by configuration — a model, its instructions, and the tools it can use. Then we add the pieces that make an agent capable. The Toolbox aggregates many tools behind a single MCP endpoint, pointing toward deferred, on-demand loading so an agent isn't flooded with every tool definition up front. A Foundry IQ knowledge base grounds agents with permission-aware, cited retrieval that several agents can share. And Skills bring reusable behaviour as plain Markdown SKILL.md files — an enterprise platform adopting the open skills format, a shift that hasn't had the coverage it deserves.

With those blocks we build several specialised agents — and then wire them together with A2A, now supported directly in the Foundry UI for Foundry-to-Foundry communication, turning individual agents into a coordinated orchestration without leaving the portal.

We close with hosted agents, the most powerful and most involved piece, where you bring your own containerised code and Foundry runs it. By the end you'll know what each block is, how they combine, and when to reach for each one.