The Terminal Live - Day 2 at Microsoft Build

Key Concepts

• Intelligent Terminal: An experimental terminal experience with integrated AI agents that detect errors and suggest resolutions.
• Copilot CLI & Agents: Tools for automating developer workflows, including "Fleet" (agent swarms) and "Autopilot" (autonomous execution).
• WinUI 3 & WinDev Skills: A plugin/framework for building native Windows applications with AI-assisted design, testing, and packaging.
• GitHub Copilot SDK: A GA-released toolkit for embedding agentic loops into any application, supporting multiple languages (TypeScript, C#, Go, Java, Rust).
• Canvases: An agent-driven UI surface for building custom, interactive workflows and dashboards.
• Work Trees: A Git feature used by agents to isolate tasks and prevent merge conflicts during parallel development.
• Foundry Hosted Agents: A cloud-based sandbox technology for running agentic workloads in isolated environments.
• Aspire: An orchestration framework for building and deploying multi-service applications, now supporting TypeScript and Kubernetes.

---

1. Main Topics and Key Points

• Windows Development Evolution: Microsoft is focusing on "delighters"—small, impactful improvements to Windows (e.g., vertical taskbars, Git-integrated File Explorer) to remove friction for developers.
• Agentic Workflows: The shift from simple AI assistance to "force multipliers" where agents handle end-to-end tasks like UI testing, packaging, and deployment.
• Intelligent Terminal: Features an agent pane that monitors terminal output. If a command fails, the agent automatically analyzes the error and provides a fix.
• GitHub Copilot SDK: Now generally available, it allows developers to embed the same agentic core used by GitHub Copilot into their own applications. It supports "Bring Your Own Model" (BYOM) and runs in secure, isolated environments.

2. Real-World Applications

• Weather App Development: Kayla Cinnamon demonstrated building a native WinUI 3 weather app from scratch using AI agents. The agent handled API integration, UI design, and packaging, even adding a "Carl the Fog" personality (a San Francisco local lore reference).
• Automated UI Testing: The agentic workflow included automated UI testing where the AI moved the mouse, clicked elements, and implemented visual fixes based on screenshot analysis.
• Project Management: The use of "Canvases" to create custom Kanban boards and triage dashboards that interact directly with the agent to move tasks and update status.

3. Methodologies and Frameworks

• Fleet Mode: An agent swarm methodology where multiple sub-agents work on different parts of a codebase simultaneously. It uses dependency-aware planning to ensure tasks are executed in the correct order without merge conflicts.
• Autopilot Mode: An autonomous execution mode where the agent continues working until a task is complete, bypassing individual permission prompts (YOLO mode).
• Plan Mode: A structured interaction where the agent asks clarifying questions to define architecture and requirements before writing code.
• Aspire Orchestration: Uses a TypeScript or C# "App Host" to define application architecture, managing service discovery and randomized port allocation to prevent conflicts.

4. Key Arguments and Perspectives

• From Assistance to Orchestration: The speakers argued that the industry is moving away from simple code completion toward "agentic systems" where developers act as managers, delegating complex, multi-step tasks to AI.
• Clarity over Prompting: The presenters emphasized that "prompting" is evolving into "collaboration." Using voice and natural language to iterate with agents is becoming more effective than rigid, manual coding.
• AI-Ready Codebases: To move fast, codebases must be "AI-ready"—meaning they should have robust test suites, clear architectural layers, and automated CI/CD guardrails that agents can navigate.

5. Notable Quotes

• "GitHub is turning AI from a tool into a force multiplier."
• "I don't think I've manually written code since November." — Evan Bole
• "The goal of plan mode is getting to a clear understanding of the architecture... the agent will literally pull it out of you." — Evan Bole

6. Synthesis and Conclusion

The video highlights a paradigm shift in software development where the terminal and the IDE are becoming "agent-first" environments. By leveraging tools like the GitHub Copilot SDK, Aspire, and Intelligent Terminal, developers can offload repetitive tasks—such as environment setup, UI testing, and CI/CD management—to autonomous agents. The focus is shifting from writing individual lines of code to orchestrating complex systems, with the ultimate goal of increasing developer velocity while maintaining high quality through automated verification loops.