Event in Spanish: Jueves de Quack con Gisela Torres

Aquí tienes un resumen detallado del transcript del video de YouTube, manteniendo el idioma original y la precisión técnica:

Key Concepts

GitHub Universe 2025 Announcements: Key product updates and future directions for GitHub.
AI Agents: Introduction and demonstration of custom AI agents for software development.
GitHub Copilot: Integration of AI assistance within the development workflow.
Visual Studio Code (VS Code) Integration: Demonstrations of AI agent functionality within VS Code.
Plan Mode vs. Agent Mode: Two distinct modes for interacting with AI agents in VS Code.
Custom Agents: Ability to create and configure personalized AI agents.
MCP Servers (Microsoft Connected Platform Servers): Tools and services that AI agents can leverage.
AI Controls (Enterprise Level): New organizational controls for managing AI features and MCP servers.
CodeQL: Integrated code analysis engine for security and quality.
Code Quality Analysis: New features for assessing code maintainability and reliability.
GitHub Copilot Chat: Enhanced chat interface for interacting with AI agents.
GitHub CLI: Command-line interface for GitHub, with AI integration.
GitHub.com Features: Updates and new functionalities available on the GitHub platform.

GitHub Universe 2025 Announcements and AI Agent Integration

The broadcast focuses on the significant announcements from GitHub Universe 2025, emphasizing the evolving role of AI agents in software development. The core theme is the integration of AI agents as collaborative partners within development teams, working alongside human developers.

Local Development with GitHub Copilot and Custom Agents in VS Code

The demonstration begins by showcasing how GitHub Copilot and custom agents can be utilized locally within Visual Studio Code (VS Code). The presenter uses a "Guardians of the Galaxy" theme to build a personal project, illustrating the practical application of these tools.

Project Setup: The presenter is building a web application using Astro and leveraging GitHub Copilot for assistance. The project aims to create an "Awesome Mix dashboard" inspired by the movie's cassette tapes, with future plans to link songs to streaming services.
New "Plan Mode" in VS Code: A key announcement is the introduction of a "Plan Mode" in VS Code, previously available only in Insiders. This mode allows developers to collaborate with Copilot to define project tasks and requirements.
- Process: The presenter asks Copilot to outline the steps for creating the dashboard. Copilot, after a conversation, generates a list of issues (ISUs) on GitHub.com, which are then used to structure the development process.
- Issue Creation: Copilot can create detailed issues based on the initial plan, which can then be managed within GitHub Projects.
- "Import Items from Repository" Feature: A new feature allows importing existing issues or tasks into a project, significantly saving manual effort.

Custom Agents and Their Configuration

The discussion moves to the creation and utilization of custom agents within VS Code.

Custom Agents vs. Chat Modes: Previously known as "chat modes," these are now referred to as "custom agents."
Agent Scope: Custom agents can be configured at two levels:
- User Data (Global): Agents available across all projects opened in VS Code.
- Repository Level (Local): Agents defined within a .github/agents directory in a specific repository.
Agent Configuration: Agents are defined using Markdown files, specifying their description, allowed tools, and guiding principles.
Astro Expert Agent: The presenter demonstrates creating a custom agent specifically for Astro development. This agent is trained using documentation from Astro's "Build with AI" section, ensuring it follows best practices.
Hand-offs: A concept allowing agents to pass tasks to other agents upon completion, facilitating complex workflows. For example, an Astro agent could generate a component, and then a Playwright agent could handle testing.
Agent Generator: A meta-agent that can create other agents, streamlining the agent creation process.

MCP Server Integration and AI Controls

The session highlights the integration with MCP Servers and the new enterprise-level AI controls.

MCP Server Issues: The presenter encounters an issue where MCP Servers are not accessible due to organizational policies.
AI Controls for Organizations: A new feature in the enterprise settings allows administrators to control the use of MCP Servers and agent modes.
- Policy Management: Organizations can decide whether to allow MCP Servers, restrict them to specific registries, or disallow them entirely.
- Private MCP Server Registries: The possibility of creating private registries for approved MCP Servers is mentioned, enhancing security and control.
Authorization with OAuth: The presenter demonstrates the ease of authorization with OAuth when using GitHub's remote server.

Orchestration of Agents and GitHub.com Features

The discussion expands to how agents can be orchestrated across different environments and new features on GitHub.com.

Agent Orchestration Tab: A new tab in VS Code allows users to manage and interact with agents running locally, via GitHub CLI, or as cloud agents on GitHub.com.
Cloud Agent Integration: Issues assigned to Copilot on GitHub.com can be handled by custom agents, allowing for seamless local and cloud development workflows.
Custom Agents in Pull Requests: Custom agents created locally can be used directly when assigning tasks to Copilot for pull requests on GitHub.com.
Enterprise-Level Custom Agents: Agents can be managed at the enterprise level, making them available across multiple organizations.
AI Controls Dashboard: An enterprise dashboard provides administrators with insights into AI usage, including session history, agent execution, and model enablement.
Metrics and Insights: A new "Copilot User" dashboard offers metrics on Copilot adoption, model usage, and other relevant data.

Code Review and Code Quality Analysis

The session delves into advanced AI-powered code review and quality analysis features.

Copilot Code Review: The presenter demonstrates how Copilot can review pull requests, offering suggestions for code improvements and adherence to organizational standards.
- Integration with CodeQL: Copilot's code review now leverages CodeQL for deeper analysis, going beyond simple security checks to assess code quality, maintainability, and reliability.
- Actionable Fixes: Copilot can suggest and even implement fixes based on review feedback.
Code Quality Analysis (Preview): A new preview feature, "Code Quality," analyzes code for maintainability, reliability, and potential improvements, providing visual feedback and suggesting fixes.
Dependency Alerts and Secret Scanning: Integration with Dependabot and secret scanning capabilities are highlighted as part of the broader security analysis.

Future Outlook and Community Engagement

The presenters conclude by emphasizing the rapid pace of AI development and encouraging community engagement.

Continuous Learning: The presenters acknowledge the challenge of keeping up with new releases and encourage a gradual adoption of new features.
Organizational Enablement: The importance of organizations enabling preview features for developers to experiment with is stressed.
Upcoming Features: Mentions of integrations with Microsoft Teams, Slack, and Azure Boards are made, indicating further AI integration across the development ecosystem.
Interactive Sessions: The presenters express enthusiasm for future collaborations and community events.

The session concludes with a demonstration of interrupting a running coding agent to provide real-time feedback and adjustments, showcasing the dynamic nature of AI-assisted development. The presenter also shares their personal approach to managing extensive plans by breaking them down into smaller, manageable tasks, mirroring human development workflows.