Merlions, Agents & Copilot¶
A standalone guide to building trustworthy Python agents on Azure. Read it cover to cover, or jump to the chapter you need. Each chapter pairs a concept with clear, copyable instructions so you can build the same patterns yourself, no presentation required.
Everything here is grounded in current Microsoft Build 2026 guidance and in the working repository content under chapters.
How to use this site¶
- Developers: follow the Do this next sections. They reference the repo structure and chapter flow.
- Decision makers: read the Why it matters and Key terms sections. They give you the vocabulary to ask the right questions.
- Everyone: the Build 2026 grounding boxes point to the supporting sessions and ideas used throughout the talk.
Chapter index¶
| # | Chapter | Theme |
|---|---|---|
| 1 | Why trust? | Framing |
| 2 | What we'll explore | Framing |
| 3 | Meet our agents | Cast |
| 4 | Multi-agent systems | Architecture |
| 5 | GitHub Copilot to the rescue | Tooling |
| 6 | Trust is our architectural style | Pattern |
| 7 | Walkthrough: Copilot scaffolds the safe parts | Hands-on |
| 8 | Hawker Recommender agent | Agent |
| 9 | Haze Tracker agent | Agent |
| 10 | Merlion Wisecracker agent | Agent |
| 11 | From local to cloud: Azure | Deployment |
| 12 | Observe. Evaluate. Improve. | Operations |
| 13 | Where to go from here | Close |
The one-paragraph version¶
Build small, specialised agents instead of one mega-agent. Give each one least-privilege tools, grounded retrieval, and citations. Make transparency, safety, reliability, and observability the architectural style, designed in, not bolted on. Use GitHub Copilot to generate the boring, safety-critical scaffolding. Ship to Azure with OpenTelemetry tracing and continuous evaluation from day one. The Build 2026 mantra: Observe. Evaluate. Improve. Roll out safely. Repeat.