Agents League: AI-Assisted Development Workflow
Visual Studio Codego watch the original →
the gist
A demonstration of using the GitHub Copilot app and MCP servers to automate the creation of multi-agent web applications, emphasizing 'vibe coding' and cross-context automation.
The Shift to Agentic Development
The session introduces the 'Agents League' hackathon, a competition focused on building with GitHub Copilot, Microsoft Foundry, and M365 Copilot. The core philosophy presented is moving beyond simple code generation toward 'agentic' workflows—where developers act as architects and orchestrators rather than manual coders. The demonstration highlights how modern tools allow developers to manage multiple concurrent projects by offloading repetitive tasks to autonomous agents.
Automating the 'Work Around the Work'
Kyle Dagel demonstrates a workflow using the GitHub Copilot app to build two distinct applications simultaneously. Rather than manual coding, he utilizes 'skills'—pre-configured instructions and MCP (Model Context Protocol) servers—to handle tasks like database schema creation (SQLite), frontend design, and cross-platform information retrieval. A key insight is the use of 'autopilot' mode, where the agent iterates on code, runs tests via Playwright, and performs self-critiques using a 'rubber duck' model to verify architecture before the human developer even intervenes.
Contextual Intelligence and Multi-Model Orchestration
The presenters emphasize that the real power of these agents lies in their ability to access personal and professional context. By connecting the agent to external data sources like Obsidian vaults (for personal notes) and Work IQ MCP servers (for M365/Teams data), the agent becomes a personalized assistant that understands specific company lingo and project history. Dagel highlights the importance of multi-model orchestration: using high-tier frontier models for complex architectural decisions while defaulting to faster, cheaper models for routine bash commands or simple file manipulations.
The Role of the Developer
There is a clear distinction made between the GitHub Copilot app and traditional IDEs like VS Code. The app is positioned as a 'control center' for the developer's entire digital life, not just a code editor. It allows for asynchronous development—starting a task on a desktop, monitoring progress via a mobile device, and waking up to completed options. The developer's role evolves into defining intent, setting constraints, and performing high-level reviews of the agent's output.