GitHub Copilot: Leading the AI revolution in software development

Introduction and Overview of GitHub Copilot

• GitHub Copilot is revolutionizing software development and is expected to support a future with 1 billion developers. It is becoming an essential tool for developers, much like smartphones are for today's youth. (31s)
• Proficiency in AI tools, such as GitHub Copilot, is becoming a core competency for developers, similar to mastering programming languages. (1m24s)
• GitHub Copilot has become the most widely adopted AI coding assistant, used by millions of developers and over 77,000 organizations, including more than one-third of Fortune 500 companies, with a 180% year-over-year growth. (1m57s)
• In Gartner's first Magic Quadrant for AI coding assistants, published in August, GitHub Copilot was placed at the top of the leaders quadrant for its product vision and ability to execute. (2m12s)
• The team behind GitHub Copilot is focused on three dimensions for future development: providing top-quality assistance and exceptional editor experiences, offering customized and trusted experiences for organizational customers, and integrating with GitHub to extend Copilot's capabilities across the software development lifecycle. (3m36s)

Improving Core Code Completion Experience

• GitHub Copilot's core experience is ghost text auto-completion, which provides consistent, high-quality suggestions. The team is continuously advancing this feature by fine-tuning code-specific models for better performance across various programming languages and domains. (3m49s)
• Upcoming improvements include launching new models starting with .NET enhancements and improving prompt engineering to provide more context-aware suggestions by considering the entire project and code base. (4m39s)
• Recent improvements include integration with C++ language services in Visual Studio and Visual Studio Code, allowing Copilot to suggest accurate utility functions even when the header file is not open in the editor. (5m17s)

Enhancing Editor Experience and Model Selection

• The team is also developing new features to enhance the overall editor experience, helping developers navigate more efficiently. (5m51s)
• GitHub Copilot in VS Code introduces new features, including a model picker that allows users to select different models based on task complexity, with the advanced model "A1" recommended for complicated tasks like building a 3D chess game from scratch. (6m4s)
• When using the A1 model, it is important to provide specific and clear prompts to help the model plan and generate the required output, such as a fully functional chess game with artistic rendering. (7m0s)
• For tasks requiring quick code iterations, the default model GPT-4 is suitable, as demonstrated by enabling shadows in the chess game with minimal code changes. (7m33s)
• Copilot is effective in handling boilerplate code and can be customized with specific instructions to better fit unique codebases, enhancing productivity and focus during development. (8m30s)
• Developers can personalize Copilot's instructions with fine-grained settings in VS Code, allowing for a tailored experience across different projects. (9m3s)

Introducing Copilot Edits and Commands

• Copilot Edits is a new tool for implementing large changes across multiple files, such as creating a leaderboard for a marathon tracking website, with the ability to review and save changes incrementally. (9m23s)
• Copilot Commands offer quick solutions for common development tasks, such as fixing errors, writing tests, or drafting commit messages, demonstrated by a command that fixes a failing test case in VS Code's Test Explorer. (10m57s)
• GitHub Copilot can analyze code to identify root causes, apply fixes, and rerun tests, making the process faster and easier. (11m25s)
• Visual Studio Code's extension ecosystem allows for adaptability, with extensions like an AI-generated diagram system powered by Mermaid JS and Copilot, which can generate database relationship diagrams. (11m43s)

Customization and Extension of Copilot

• The Copilot platform offers customization layers, allowing users to extend its functionality into their workflows. (12m51s)
• New features in Visual Studio Code include powerful model choices, multifile editing, and the ability to speed up tasks with Slash commands. (13m7s)
• A new IDE extension, Copilot for Xcode, is announced, bringing Copilot's capabilities to developers in the Apple ecosystem. (14m0s)
• Copilot users, including individuals, businesses, and enterprises, can access the new Xcode extension by following a getting started guide. (14m25s)

Organizational Adoption and Benefits of Copilot

• Organizations can leverage AI to accelerate business outcomes and innovations, with Copilot offering tailored assistance to meet unique organizational needs. (14m53s)
• Fine-tuned models allow Copilot suggestions to be customized to an organization's specific codebase, internal APIs, libraries, and coding guidelines. (15m18s)
• The fine-tuned models are available in public preview, using the Lowa fun method to adjust critical model parameters efficiently. (15m35s)
• GitHub Copilot offers enterprise customers the ability to customize their experience through Copilot extensions, which are available on the GitHub Marketplace. These extensions integrate with third-party developer tools and services, such as JFrog and DataStax, to enhance developer productivity. (16m47s)
• The Alashan R extensions allow developers to query issues and access best practices and documentation from Confluence directly within the IDE, maintaining developer flow and boosting productivity. (17m38s)
• DataStax extensions enable developers to manage databases using natural language, allowing for retrieval and updates without needing to memorize complex query syntax. These capabilities are accessible through Copilot chat. (18m1s)
• Organizations can develop private extensions for internal workflows and tooling systems, which are only visible and available to their development teams. (18m41s)

Adobe's Use Case and Integration with Internal Systems

• Adobe's developer platform team focuses on reducing cognitive load, maintaining flow state, and creating fast feedback loops to enhance developer productivity. They have implemented GitHub Copilot to support these goals. (19m51s)
• Adobe has achieved a 55% adoption rate of GitHub Copilot among its 13,000 developers, with an NPS of 34, indicating a positive reception. The team accepts around 70,000 lines of code per day from Copilot. (21m0s)
• AI coding assistants are effective for general programming queries in languages like Python, Java, and Docker, but struggle with proprietary or internal information. (21m29s)
• Adobe has developed an augmentation strategy called Adobe Developer Assistant (Ada), named after Ada Lovelace, to help developers navigate complex internal systems using a simple AI interface. (21m56s)
• Ada assists developers by providing relevant information directly in their Integrated Development Environment (IDE), allowing them to focus on meaningful tasks and maintain a flow state. (22m40s)
• Ada can handle specific Adobe-related queries, such as setting up communication in Adobe's compute platform, ethos, and integrating with Adobe's identity system using IMS lib. (23m0s)
• The system also supports onboarding new developers by providing guidance and minimizing cognitive load through fast feedback loops. (23m51s)

Future Developments and Technical Previews

• Builders interested in creating Copilot extensions can access a toolkit with samples, tutorials, and documentation, with a general availability target in early 2025. (24m21s)
• A technical preview of Copilot Upgrade Assistant for Java is announced, which uses a multi-agent framework to upgrade Java applications, handle dependencies, and transform code with a human-in-the-loop approach. (25m24s)

Safety, Security, and Privacy Enhancements

• GitHub Copilot is designed to learn from user fixes and apply them to ongoing tasks, allowing users to stay in control by reviewing file changes, committing diffs, and creating pull requests. (26m3s)
• Copilot is being developed with a focus on safety, security, and privacy, incorporating features to protect against security vulnerabilities, harmful content, and personal information leaks. (26m54s)
• Recent updates include content exclusion features to prevent sensitive files from being accessed by Copilot, with notifications for users when triggered. (27m30s)
• Code referencing features have been made generally available, providing transparency by notifying users when Copilot suggests code that matches public code, including details like file URLs and licensing information. (27m52s)
• Efforts are being made to help organizations adopt Copilot effectively, including the introduction of a Copilot metrics API, an engagement dashboard, and access to event-level usage data. (28m54s)

Extending Copilot Across the Software Development Lifecycle

• Guidance will be increased on using Copilot to improve engineering outcomes, such as developer happiness, velocity, code quality, and business results. (29m56s)
• Copilot's functionality is being extended to cover the entire software development lifecycle, integrating with key workflows on GitHub, a platform used by over 100 million developers. (30m22s)
• Simina, a senior director of product at GitHub, is excited to demonstrate how Copilot can enhance existing workflows on GitHub. (31m3s)
• GitHub Copilot has been enhanced to provide a deep understanding of users' repositories and organizational knowledge, integrating this context into the software development lifecycle (SDLC) to assist developers. (31m26s)
• Copilot now comprehends essential contexts such as pull requests, commits, issues, discussions, releases, files, repositories, and common support scenarios, utilizing GitHub's documentation. This information is exclusive to GitHub and accessible only through Copilot. (31m52s)
• Users of Copilot, regardless of their plan, can leverage this rich context while interacting with Copilot in various environments, including VS Code, Visual Studio, GitHub, and the GitHub mobile app. (32m15s)

New Features and Capabilities on GitHub.com

• New capabilities, including Copilot chat on GitHub.com, are available to both individual and business users of Copilot. (32m32s)
• GitHub Copilot aims to enhance core workflows, enabling developers to deliver higher quality code more efficiently and focus on delivering value to customers. (32m51s)
• Upcoming features include inline discussions and issue summaries, allowing users to quickly catch up on lengthy issues or discussions with a TLDR summary. (33m22s)
• Copilot is now integrated with GitHub's CI/CD, helping users understand and fix failing jobs by analyzing logs and suggesting fixes directly in the actions UI. (33m53s)

Copilot as a Code Reviewer in Pull Requests

• Pull requests have evolved as a central point for collaboration, and in the AI era, they are becoming more significant with workflows integrated into the PR process. (34m51s)
• GitHub Copilot can now act as a code reviewer, providing quick reviews and feedback on code, identifying bugs, and suggesting improvements in spelling, grammar, and error handling. (35m8s)
• Organizations can enable automatic code reviews for new pull requests using rule sets, allowing Copilot to provide feedback within about 30 seconds. (35m51s)
• Copilot's suggestions can be reviewed and accepted with a single click, helping to prepare code for merging while still awaiting human review. (36m1s)
• GitHub Copilot allows developers to receive code reviews directly in the VS Code editor, enabling more frequent and confident code commits. (36m39s)
• Copilot code reviews have been implemented on all pull requests in GitHub's main repository, significantly reducing the time from opening a pull request to merging it. (37m8s)
• In a sample of over 200 pull requests, more than 60% of Copilot's code suggestions were directly accepted into the merge. (37m19s)
• Feedback from a private preview with various organizations showed that 70% of developers found Copilot's reviews useful for improving pull request quality. (37m38s)
• Jesse, a staff engineer at Procor, noted that Copilot code reviews provide valuable comments similar to those from another developer on the team. (37m51s)
• Eduardo mentioned that GitHub Copilot helps in quickly committing enhancements and achieving mergeable pull requests. (38m6s)

Custom Coding Guidelines and Enterprise Features

• Copilot Enterprise customers can create custom coding guidelines in natural language, allowing Copilot to provide reviews tailored to specific standards and best practices. (38m35s)
• Internally at GitHub, Copilot code reviews with coding guidelines have been effective in automatically catching common issues that previously led to failed deployments. (38m47s)
• A new feature in Copilot Workspace on GitHub allows users to review and apply code suggestions in the context of the entire pull request, with support for security-related code suggestions and testing. (39m29s)
• Human reviews are also supported, with Copilot able to suggest code changes based on comments from colleagues. (40m31s)
• GitHub Copilot allows users to quickly iterate on changes directly in pull requests using the Copilot workspace, which validates changes on GitHub's provided compute resources and integrates them back into the pull request. (41m11s)
• Copilot code review and Copilot workspace in pull requests are now available in public preview for individual users and organizations using Copilot Business or Copilot Enterprise. (41m27s)

Conclusion and Future Outlook

• GitHub Copilot is designed to benefit a wide range of users, including professional developers, hobbyists, and open-source maintainers, with upcoming features aimed at enhancing productivity from the integrated development environment (IDE) through to production. (41m46s)
• Organizations at any stage of adopting GitHub Copilot can benefit from its deep customization, insights, and security features. (42m8s)