Agent Framework + Toolbox
How do I give an agent governed tools without custom plumbing?
A Framework agent connected to a Foundry Toolbox / MCP endpoint — tools arrive governed, no bespoke wiring.
◈ 02 / Samples
Eleven questions, eleven builds
The opening backlog. Each track starts from a public Build 2026 announcement and resolves to one concrete sample brief — tagged to the layer it exercises and where it sits on the maturity ladder.
Sketch
→
Planned
→
Runnable
→
Governed reference
Foundry IQ grounding
How do I ground an agent across enterprise and web context?
A retrieval-first agent with Foundry IQ that reports its sources explicitly — grounding you can audit.
Azure Functions serverless agent
How do I run an event-driven agent cheaply and durably?
A Function-hosted markdown or code agent with MCP auth and durable workflow notes — pay-per-run, durable by design.
Container Apps sandbox
How do I let agents run code safely?
A dynamic session / code-interpreter sandbox with identity and network controls — arbitrary code, contained.
AgentOps eval loop
How do I know if the agent is safe and improving?
Traces, rubric evals, guardrail checks, and a ROI / quality dashboard shape — the feedback loop that proves it works.
Model router + inference substrate
Which model and runtime should handle each task?
A task router that compares MAI, Fireworks, Managed Compute, Foundry Local, and GPU/VM paths with latency and quality notes.
Foundry Local edge agent
How do I run useful AI where data or latency keeps me local?
A voice/retrieval agent brief for Foundry Local 1.2 with an Azure Local deployment note and offline fallback checks.
Prompt-to-production backend
How does an agent ship a real backend instead of loose code?
A Rayfin + Fabric + HorizonDB blueprint for generated data models, APIs, auth, policies, and deployment review.
Windows local-agent harness
How do I run desktop agents with OS containment and identity?
A Windows AI / MXC / Aion checklist for local model execution, sandboxing, attribution, policy, and enterprise controls.
Copilot worktree workflow
How do I coordinate parallel agent sessions without losing review control?
A GitHub Copilot app workflow for worktrees, issue-to-PR loops, review gates, rollback notes, and traceable decisions.
Discovery research agent
How do I make agentic research reviewable instead of speculative?
A Microsoft Discovery-inspired narrative for hypothesis, evidence, tool use, reviewer checkpoints, and governed handoff.
Implementation ladder
The full maturity ladder lives in the repo. Each brief must verify public SDK access, preview status, identity, cleanup, and test steps before it becomes runnable.
samples/README.md →