.NET AI Community Standup: .NET + AI at Build 2026: What’s Coming Next?
TL;DR
Build 2026 is overwhelmingly about AI — Bruno used Microsoft’s new open-source Build CLI skill to query the catalog and found 229 of 330 sessions are AI-related, which he called a clear signal of where the event is headed.
Hosted agents are the big .NET-to-cloud story — Bruno framed Azure AI Foundry hosted agents as the missing deployment piece for developers building locally with Microsoft Agent Framework, because you can package a C# agent into a container and deploy it with traces, monitoring, evaluation, and versioning built in.
Microsoft Agent Framework is maturing fast around context and workflows — Jose highlighted improvements like AI context providers, chat history management, compaction strategies, and durable workflows as the practical stuff that saves tokens and makes multi-step agents production-worthy.
A2A interoperability is becoming real, not theoretical — Bruno demoed an Aspire app where an NVIDIA NeMo agent and a Microsoft Agent Framework agent both expose A2A-compatible endpoints, then showed how a Blazor UI and Aspire dashboard let him inspect calls and telemetry across the system.
Aspire is turning into an observability control plane for agents — both hosts kept coming back to Aspire because it gives developers and coding agents a single place to inspect components, logs, traces, MCP integrations, and service health, which Jose said is “fantastic” for agentic workflows.
Model choice matters less than harness quality now — Bruno said the real differentiator is the “set of tools and frameworks and processes and workflows” around the model, while Jose tied that to token savings and precise context control rather than chasing raw model capability.
Summary
Bruno sets the stage: this isn’t a product recap, it’s a Build prediction market
Bruno Capuano opens by saying this standup is intentionally different: not a feature rundown, but an honest look at what he and Jose expect Microsoft to announce at Build on June 2–3 in San Francisco. He calls Build his favorite Microsoft event because it’s where the big product, security, and data reveals usually land, while also joking that if you’re attending online, the city almost doesn’t matter.
A quick Build tour, and why the labs sound unusually worth it
They walk through the Build catalog and session filters, with Bruno pointing out keynotes, breakouts, in-person labs, and the new digital labs format. Jose jumps in to say the labs are “fantastic” and recommends anyone attending — especially virtually — check them out because they’re more thoughtfully designed than people might expect.
The new open-source Build CLI turns your repo into a session recommender
Bruno then demos the GitHub Copilot-powered Build CLI plugin, which scans a real repo — in his case a .NET 10/Aspire/NVIDIA NeMo/Microsoft Agent Framework sample — and recommends relevant sessions. The fun part is how specific it gets: it maps the stack, searches terms like Aspire, OpenTelemetry, Azure OpenAI, and returns sessions such as Damian Edwards and David Pine’s Aspire-for-agents talk plus telemetry and production-agent demos; when Jose asks how many AI sessions exist overall, the tool answers 229 out of 330.
Hosted agents: the cloud deployment piece Bruno thinks was missing
The conversation shifts to Azure AI Foundry hosted agents, which Bruno describes as the bridge from local development to production. His pitch is simple: build your agent locally in C# with Microsoft Agent Framework, containerize it, push it, and now it shows up in Foundry with prompt-vs-hosted agent support, traces, monitoring, evaluation, and version history — the stuff teams actually need once a demo turns real.
Jose’s deeper take: context engineering is where the framework gets serious
Jose says one of the smartest parts of Agent Framework is how it sits on top of Microsoft.Extensions.AI and how much work has gone into execution changes, workflows, and especially memory. He calls out agent sessions, AI context providers, and custom compaction strategies as the real machinery behind “context engineering,” because they let you control exactly what gets passed back to stateless models instead of wasting tokens on the entire history every time.
CodeAct, durable workflows, and the new obsession with not wasting GPU time
From there, the conversation zooms out to the broader trend: saving tokens, shrinking context, and making multi-step agents resumable. Bruno mentions CodeAct as an emerging approach to collapsing multi-step plans into tighter context blocks, while also bringing up durable functions as a way to pause agent work at step 3, persist state, and resume later instead of forcing one giant HTTP request to carry the whole flow.
A live A2A + Aspire demo with NVIDIA and Microsoft agents in one app
Bruno closes the technical section by showing an Aspire app that orchestrates two agents: one built with NVIDIA’s NeMo toolkit and one with Microsoft Agent Framework, both exposing A2A endpoints and fronted by a Blazor UI. The demo’s real point isn’t the UI — it’s that Aspire gives him a graph of dependencies, traces for A2A calls, and enough visibility to immediately ask GitHub Copilot why one agent emits OpenTelemetry GenAI data while the other doesn’t.
Final predictions: agents everywhere, Foundry GA, and maybe a little .NET 11
In the wrap-up, both hosts make their Build bets explicit: expect lots of agent announcements, more Agent Framework news, Foundry Agent Service going GA, and tooling updates around Visual Studio, coding agents, debugging, profiling, and modernization. Bruno says .NET 11 may appear, but the real advice is simpler: if you’re building with AI, prioritize the agent sessions, because that’s where the energy clearly is.
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
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
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.