Building the AI Native Cloud¶
AgentCon Singapore 2026 · Michelle Sandford
AI-native is a way of engineering, not a feature. Build cloud systems where AI has useful context, bounded capabilities, and evidence behind every consequential decision.
Start the workshop Explore the chapters
Welcome¶
This technical companion turns the ten ideas from Building the AI Native Cloud into hands-on engineering practice. Follow the talk in order or jump straight to a chapter. Each slide is available full size; the text below it explains the same ideas without requiring you to read the image.
The running example is a read-only cloud incident-triage assistant. GitHub Copilot helps you specify, implement, test, and review it. The Python harness, not a model's confidence, decides which operations are permitted. No cloud account, paid API, credentials, or live model is needed for the exercises.
Get ready¶
Use Python 3.15 for the workshop and a current VS Code installation with GitHub Copilot access for the prompting exercises. The standard-library example also works with Python 3.12 or later. Copilot features and availability depend on your plan, editor, and organisation policies.
Clone this repository using its Code menu and open the folder in VS Code. In the integrated terminal, run python -m examples.cloud_triage, then python -m unittest discover -s tests -v. On Windows, use py if python is not available. The resources page links to the source repository and setup documentation.
Choose your pace
Follow the ten chapters alongside the 25-minute talk, or allow 60 to 90 minutes to complete the exercises. Each chapter includes a Copilot prompt and an observable checkpoint. You can read all the content without a Copilot subscription.
The Agentic SDLC¶
Specify → plan → implement → verify → review → release → observe → learn.
AI can contribute at every stage, but people own the intent, boundaries, and release decision. Treat repository instructions as guidance, tests as evidence, and permissions as enforcement. A prompt alone is never a security boundary.
Choose your chapter¶
00 · Welcome to AgentCon Singapore
Workshop setup, learning outcomes, and the Agentic SDLC.
01 · Build the AI-native cloud
Start with engineering intent, not a model integration.
02 · From AI-assisted to AI-native
Distinguish an assistant, a workflow, and a production system.
03 · Human judgment
Make direction, context, judgment, and accountability explicit.
04 · Context and specifications
Give Copilot the right context and measurable acceptance criteria.
05 · Tools, agents, and boundaries
Constrain capabilities in the harness, not just in the prompt.
06 · Inspectable engineering
Keep specifications, patches, tests, and review connected.
07 · Reliability and evaluation
Measure representative behaviour and safe failure.
08 · Least privilege by design
Protect data and require meaningful approval for higher risk.
09 · Repeatable team practice
Turn individual experiments into shared engineering capability.
10 · Build with intent
Make an evidence-based release decision and own what follows.
Continue learning · Resources
Official Copilot documentation, learning paths, and workshop references.