.NET AI Community Standup: Multi-Agent Apps with Aspire + MAF
TL;DR
Aspire 13.4 closes a major Foundry gap: Tomaso explains that earlier Aspire support stopped at Foundry resources and model deployments, but 13.4 adds support for Foundry projects, prompt agents, and hosted agents so modern Foundry apps map cleanly into the Aspire app model.
The architecture treats agents like microservices: The ski resort demo splits work into narrow agents like weather, traffic, and safety, echoing patterns .NET developers already know from 10-plus years of microservices design.
A2A is the glue for mixed-language agent systems: Tomaso uses A2A to expose Python and .NET sub-agents over standard endpoints like /.well-known/agent-card.json and /message-stream, then plugs them into a Foundry hosted orchestrator.
Foundry has become a real compute target, not just a model catalog: Bruno highlights that developers can now build hosted agents locally, containerize them, and deploy them into Foundry in a natural flow instead of stitching everything together manually with Azure Container Apps.
Simple code replaces painful Azure setup: A single line like addWebSearchTool replaces what Bruno describes as a very recent headache involving Bing resources, approvals, and fragile Bicep configuration.
Aspire hides the ugly cloud scaffolding: Even a small sample causes Aspire to generate supporting resources like managed identity, Azure Container Registry, Log Analytics, and App Insights, which Tomaso calls the kind of infrastructure developers usually forget until deployment breaks.
The Breakdown
A ski resort demo turns into a full-stack preview of where .NET AI is heading: with Aspire 13.4, Microsoft AI Foundry projects, hosted agents, prompt agents, A2A, voice, and Azure Container Apps can finally be wired together without the usual deployment mess. The big win is simple but concrete: things like web search, hosted-agent deployment, and even realtime voice now show up as a few lines of code instead of a pile of Bicep, Bing setup, and container plumbing.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.