Branch: squad/dozer-maf-upgrades-2026-04-08
Requested by: Bruno Capuano
Date: 2026-04-08
Microsoft Agent Framework v1.0 GA was released on 2026-04-02 (GitHub release). All 26 MAF samples in this repo currently reference preview packages (1.0.0-preview.260108.1 / 1.0.0-preview.251219.1). This plan covers upgrading to stable v1, updating docs, adding new Hosted Agent scenarios, and improving the overall learning experience.
- Introducing Microsoft Agent Framework (blog) — 2025-10-01
- MAF Reaches Release Candidate (blog) — 2026-02-20
- Foundry Agent Service GA (blog) — 2026-03-16
- MAF v1.0 GitHub Release — 2026-04-02
- Hosted Agents Docs
| Package | v1 Version |
|---|---|
Microsoft.Agents.AI |
1.0.0 |
Microsoft.Agents.AI.Abstractions |
1.0.0 |
Microsoft.Agents.AI.Foundry |
1.0.0 (replaces Microsoft.Agents.AI.AzureAI) |
Microsoft.Agents.AI.OpenAI |
1.0.0 |
Microsoft.Agents.AI.Workflows |
1.0.0 |
Microsoft.Agents.AI.Workflows.Generators |
1.0.0 |
Microsoft.Agents.AI.AzureAI→Microsoft.Agents.AI.Foundry— package rename, likely namespace changesOpenAIAssistantClientExtensionsremoved — any samples using this need alternativeServiceStoredSimulatingChatClient→PerServiceCallChatHistoryPersistingChatClient— rename in persisting samples- RC5 cleanup: kwargs across agents/chat clients/tools/sessions changed
Azure.AI.Agents.Persistent— check if still beta or if there's a v1 equivalent
| Package | Current | Action |
|---|---|---|
Azure.AI.OpenAI |
2.8.0-beta.1 | Check for stable release |
OpenAI |
2.8.0 | Check for updates |
Aspire.Azure.AI.OpenAI |
9.5.2-preview / 13.1.0-preview | Check for stable |
OllamaSharp |
5.4.18 | Check for updates |
ModelContextProtocol |
1.0.0 | Already stable ✅ |
elbruno.Extensions.AI.Claude |
0.1.0-preview.6 | Check for updates |
Microsoft.Extensions.AI |
10.3.0 | Already stable ✅ |
| # | Sample | Key Change |
|---|---|---|
| 1 | MAF01 | Microsoft.Agents.AI preview → 1.0.0 |
| 2 | MAF02 | Microsoft.Agents.AI + Workflows preview → 1.0.0 |
| 3 | MAF-Ollama-01 | Microsoft.Agents.AI preview → 1.0.0 |
| 4 | MAF-FoundryClaude-01 | Microsoft.Agents.AI preview → 1.0.0 |
| 5 | MAF-FoundryClaude-Persisting-01 | Microsoft.Agents.AI preview → 1.0.0 |
| 6 | MAF-MultiModel | Microsoft.Agents.AI.Workflows preview → 1.0.0 |
| # | Sample | Key Change |
|---|---|---|
| 7 | MAF-AIFoundry-01 | AI.OpenAI preview → 1.0.0 |
| 8 | MAF-AIFoundry-02 | AI.OpenAI preview → 1.0.0 |
| 9 | MAF-BackgroundResponses-01-Simple | Abstractions + AzureAI + OpenAI → v1 (AzureAI→Foundry) |
| 10 | MAF-BackgroundResponses-02-Tools | Same as above |
| 11 | MAF-BackgroundResponses-03-Complex | Same as above |
| 12 | MAF-Persisting-01-Simple | Abstractions + AzureAI + OpenAI → v1 |
| 13 | MAF-Persisting-02-Menu | Abstractions + AzureAI + OpenAI → v1 |
| 14 | MAF-ImageGen-01 | AI.OpenAI preview → 1.0.0 |
| 15 | MAF-ImageGen-02 | AI.OpenAI preview → 1.0.0 |
| # | Sample | Key Change |
|---|---|---|
| 16 | MAF-AIFoundryAgents-01 | AzureAI.Persistent + Workflows → v1; check Azure.AI.Agents.Persistent beta compat |
| 17 | MAF-MicrosoftFoundryAgents-01 | Same as above |
| 18 | MAF-MicrosoftFoundryAgents-02 | Same as above |
| 19 | MAF-MultiAgents | AzureAI + AzureAI.Persistent + Workflows → v1; 3-agent orchestration |
| 20 | MAF-MultiAgents-Factory-01 | Declarative + Workflows → v1; factory pattern |
| # | Sample | Key Change |
|---|---|---|
| 21 | MAF-AIWebChatApp-Simple | Hosting + OpenAI alpha/preview → v1; Aspire update |
| 22 | MAF-AIWebChatApp-Middleware | Hosting + Abstractions → v1; middleware pipeline |
| 23 | MAF-AIWebChatApp-Persisting | Hosting + Abstractions → v1 |
| 24 | MAF-AIWebChatApp-MutliAgent | DevUI + Declarative + Hosting → v1; multi-agent web |
| 25 | MAF-AIWebChatApp-FoundryClaude | Hosting → v1; Claude integration |
| 26 | MAF-AIWebChatApp-AG-UI | AGUI + Hosting.AGUI.AspNetCore → v1; AG-UI integration |
Based on the Hosted Agents documentation, create 2 new sample projects:
What it does: A containerized .NET agent deployed to Foundry Agent Service that answers time/date questions using a local tool function (inspired by the official Python sample).
Key concepts demonstrated:
- Using
Azure.AI.AgentServer.CoreandAzure.AI.AgentServer.AgentFrameworkNuGet packages - Wrapping a MAF agent with the hosting adapter
- Local testing on
localhost:8088 - Dockerfile for
linux/amd64containerization agent.yamlmanifest for Foundry deployment- Deployment via
azd ai agent init+azd up - Managed identity + Azure Container Registry integration
Structure:
samples/MAF/MAF-HostedAgent-01-TimeZone/
├── Program.cs # Agent with timezone tool
├── Dockerfile
├── agent.yaml # Foundry agent manifest
├── MAF-HostedAgent-01.csproj
├── README.md # Step-by-step guide
└── test.http # REST test file for local testing
What it does: A containerized multi-agent workflow (researcher → writer → reviewer) deployed as a hosted agent on Foundry. Demonstrates enterprise-grade agent hosting with observability.
Key concepts demonstrated:
- Multi-agent workflow packaged as a single hosted agent
- OpenTelemetry tracing integration
- Scaling configuration (min/max replicas, scale-to-zero)
- Agent versioning and non-versioned updates
- Log streaming for debugging
- Publishing to channels
Structure:
samples/MAF/MAF-HostedAgent-02-MultiAgent/
├── Program.cs # Multi-agent workflow with hosting adapter
├── Agents/
│ ├── ResearchAgent.cs
│ ├── WriterAgent.cs
│ └── ReviewerAgent.cs
├── Dockerfile
├── agent.yaml
├── MAF-HostedAgent-02.csproj
├── README.md
└── test.http
Add entry for MAF v1 GA release:
### 🚀 Microsoft Agent Framework v1.0 GA (April 2026)
- All 26+ MAF samples upgraded from preview to **stable v1.0** packages
- **Breaking change:** `Microsoft.Agents.AI.AzureAI` renamed to `Microsoft.Agents.AI.Foundry`
- **2 new Hosted Agent scenarios** — deploy containerized agents to Foundry Agent Service
- Multi-agent workflows, streaming, persistence, and MCP all production-ready
- See: [Official GA Release](https://github.com/microsoft/agent-framework/releases/tag/dotnet-1.0.0) | [RC Blog Post](https://devblogs.microsoft.com/foundry/microsoft-agent-framework-reaches-release-candidate/) | [Foundry Agent Service GA](https://devblogs.microsoft.com/foundry/foundry-agent-service-ga/)- Replace all
--prereleaseinstall instructions with stable v1 package references - Add new section: "Hosted Agents" covering Foundry Agent Service deployment
- Add new section: "Observability & Tracing" covering OpenTelemetry integration
- Reference the 2 new hosted agent samples
- Update the sample reference table with all 28 samples (26 existing + 2 new)
-
01-building-first-agent.md— Remove--prereleaseflag from install commands (line ~47-51) -
02-agents-with-tools.md— Update package references -
03-multi-agent-workflows.md— Update package references -
04-model-context-protocol.md— Update package references
- Remove "Microsoft Agent Framework (Coming Later)" text (line ~186-190) — MAF is here now!
- Add brief MAF mention with link to Lesson 04
- Create
samples/MAF/README.md— unified index of all MAF samples organized by category:- Getting Started (MAF01, MAF02, MAF-Ollama-01)
- Azure AI Foundry (MAF-AIFoundry-01/02, MAF-AIFoundryAgents-01)
- Multi-Agent Workflows (MAF-MultiAgents, MAF-MultiAgents-Factory-01, MAF-MultiModel)
- Background Responses (01-Simple, 02-Tools, 03-Complex)
- Persistence (MAF-Persisting-01/02, MAF-FoundryClaude-Persisting-01)
- Web Apps (all MAF-AIWebChatApp-* samples)
- Image Generation + MCP (MAF-ImageGen-01/02)
- Claude Integration (MAF-FoundryClaude-01, MAF-AIWebChatApp-FoundryClaude)
- NEW: Hosted Agents (MAF-HostedAgent-01/02)
- Update
samples/MAF/CLAUDE-SAMPLES-README.md— update elbruno.Extensions.AI.Claude version
- Add detailed MAF v1 entry with migration notes
Languages requiring updates: zh, tw, fr, ja, ko, pt, es, de
- Root README.md What's New section
- Lesson 04 equivalents (agent docs exist in several locales)
- Any MAF-referencing docs in translations
- Add a visual learning path in root README: Setup → Lesson 1 (Intro) → Lesson 2 (Techniques) → Lesson 3 (Patterns) → Lesson 4 (Agents) → Lesson 5 (Responsible AI)
- Add a "Start Here" guide for absolute beginners with 3 tracks:
- 🏃 Quick Start (30 min): Run MAF01, see an agent work
- 🚶 Full Course (half day): All 5 lessons in order
- 🎯 Agent Deep Dive: Skip to Lesson 4, build hosted agents
- Lesson 4.5 — Advanced Agents: Hosted agents, persistent agents, telemetry/tracing, prompt injection defense, human-in-the-loop patterns
- Lesson 6 — Production Deployment: CI/CD for agents, Azure Container Registry, Foundry Agent Service, monitoring with Application Insights
- Lesson on A2A (Agent-to-Agent) Protocol: Cross-runtime agent collaboration, the emerging standard
- Capstone project: A complete end-to-end sample (e.g., "AI Research Assistant") that combines RAG + multi-agent + MCP + hosted deployment — the culmination of all lessons
- Hands-on checkpoints per lesson: "Run this sample → Change this prompt → Swap provider → See the difference"
- Interactive code tasks: Add
// TODO:comments in samples for learners to complete - "Diff" view: Show before/after for each concept (e.g., "chat without tools" → "chat with tools")
- Unified setup script:
setup.ps1could detect and configure both Ollama and Azure OpenAI, validate prerequisites, and report readiness - DevContainer improvements: Pre-pull Ollama models, pre-install MAF packages, include azd CLI for hosted agent deployment
- Sample runner: A top-level script that lists all samples and lets you pick one to run
- Package status badges: Add NuGet version badges to sample READMEs so students see if they're on the latest
- Observability: No samples currently show OpenTelemetry tracing for agents — critical for production
- Safety: No prompt injection defense or human-in-the-loop patterns in samples
- A2A Protocol: The Agent-to-Agent protocol is a key v1 feature but has no sample
- Declarative agents:
MAF-MultiAgents-Factory-01uses declarative config but it's not well-documented as a pattern - Agent memory strategies: v1 supports various memory patterns — no dedicated sample
- Add a "Contributing Samples" guide — encourage community to submit MAF samples
- GitHub Discussions: Enable for Q&A about specific lessons/samples
- Monthly "What's New" cadence: Keep the What's New section fresh with package updates, new samples, community contributions
- Upgrade Tier 1 samples (simple package bumps) — build & verify
- Upgrade Tier 2 samples (AzureAI → Foundry rename) — build & verify
- Update Lesson 04 install instructions (remove
--prerelease) - Update Lesson 01 "Coming Later" text
- Upgrade Tier 3 samples (persistent agents, workflows) — build & verify
- Upgrade Tier 4 samples (web/hosting apps) — build & verify
- Run
dotnet formaton all changed projects - Full solution build verification:
dotnet build samples/MAF/MAF-Demos.slnx
- Create
MAF-HostedAgent-01-TimeZonesample - Create
MAF-HostedAgent-02-MultiAgentsample - Create
samples/MAF/README.mdunified index - Add Hosted Agents section to Lesson 04
- Update root README.md What's New section
- Update
10-WhatsNew/readme.mdarchive - Update CLAUDE-SAMPLES-README.md
- Update Lesson 04 sub-pages (01-04)
- Update translations (8 languages)
- Commit, push, create PR with comprehensive description
- Tag Oracle to draft a changelog entry
| Risk | Impact | Mitigation |
|---|---|---|
| v1 API breaking changes beyond documented ones | High | Build each sample individually, fix as discovered |
Azure.AI.Agents.Persistent still beta |
Medium | Check NuGet for v1 equivalent; may need to keep beta with note |
| Aspire packages still preview | Low | Note in docs; Aspire follows its own release cycle |
elbruno.Extensions.AI.Claude still preview |
Low | Community package; note in docs |
| Hosted Agent samples need Azure subscription | Medium | Provide clear setup guide; include local-testing-only path |
| Translation sync across 8 languages | Medium | Prioritize root README; lesson translations can follow |