Hands-on lab for the WeAreDevelopers 2026 (Berlin) workshop "Context > Models: How to make your agents truly intelligent."
You'll ask GitHub Copilot to plan a cupcake order for this workshop, then add context in stages and watch a generic guess turn into a grounded, costed, budget-safe plan — proving that the right context matters more than which model you use.
- Code → Codespaces → Create codespace on
main. - Wait for the devcontainer to build (Node 20, Azure CLI, Copilot extensions).
- Open Copilot Chat, switch to Agent mode, and follow docs/workshop.md.
Stuck on setup? git switch mcp-ready for a preconfigured .vscode/mcp.json.
The participant guide (docs/workshop.md) is formatted for
MOAW — a clean, paged, navigable view that's ideal to project and
for participants to follow along. Once this repo is on GitHub, open:
https://moaw.dev/workshop/gh:<owner>/<repo>/main/docs/
(MOAW loads docs/workshop.md.) Edit the authors/contacts in the file's frontmatter
before the event.
| Path | What it is |
|---|---|
docs/workshop.md |
The participant guide — the 4 rounds + the finale |
docs/facilitator.md |
How to run the session and provision the backend |
docs/fabric-iq.md |
Optional: use the real Fabric IQ data agent |
knowledge/ |
Org knowledge → ingested by Foundry IQ (rules, venue, suppliers) |
data/ |
Business data → served in Round 3 (attendees, budget, past consumption) |
mcp-server/ |
The cupcake-context MCP server (business data + place order) |
infra/ |
Scripts to provision the shared Azure backend |
- Model only — Copilot guesses.
- + Foundry IQ — it learns the org's rules.
- + business data (Fabric IQ-style) — it learns the numbers.
- Change the model — the answer barely changes. Context > Models.
Then place the order and claim a real cupcake. 🎉
GitHub Copilot Agent mode · MCP servers · Foundry IQ (Azure AI Search agentic retrieval) · Fabric IQ · grounding & context engineering.
Licensed under the MIT License.