TL;DR:
Harnessing the wisdom of the community, Visual Studio IntelliCode is revolutionizing developer productivity. We started with AI-assisted IntelliSense and are now expanding the application of artificial intelligence to significantly accelerate learning, radically improve development agility, and increase code quality by means of two exciting new capabilities: whole line completions and refactoring.
Technology is evolving so fast that every developer is constantly learning, whether you’re adopting a new programming language, API, or architecture (e.g. microservices). Amidst this rate of technological change, existing tools are no longer sufficient for achieving agility as development teams are trying to accelerate their time-to-market and increase code quality. As a result, development tools need to radically evolve to satisfy the productivity demands of modern teams.
At Microsoft Ignite, we showed a vision of how AI can be applied to developer tools. After talking with thousands of developers over the last couple years, we found that the most highly effective assistance can only come from one source: the collective knowledge of the open source, GitHub community. This is exactly what IntelliCode provides.
AI-assisted suggestions + whole-line code completions
IntelliCode now provides whole-line code completion suggestions mined from the collective intelligence of your trusted developer knowledge bases. This is like having an AI-developer pair-programming with you, providing meaningful, suggestions and whole-line code completions without disrupting your flow. To generate accurate suggestions and provide completion assistance as you code, IntelliCode extends the GPT-2 transformer language model for our machine-learning models to learn about programming languages and coding patterns.
The GPT model architecture, originally developed by OpenAI, has demonstrated strong natural language understanding, including the ability to generate conditional synthetic text examples without needing domain-specific training datasets. For our initial language-specific base models, we adopted an unsupervised learning approach that learns from over 3000 top GitHub repositories. Our base model extracts statistical coding patterns and learns the intricacies of programming languages from GitHub repos to assist developers in their coding. Based on code context, as you type, IntelliCode uses that semantic information and sourced patterns to predict the most likely completion in-line with your code.
IntelliCode has now extended our machine-learning model training capabilities beyond the initial base model to enable your teams to train their own team completions. Team completions are useful if your development team uses internal utility and base class libraries or domain-specific libraries that aren’t commonly used in open-source GitHub repositories. If you’re using code that isn’t in that set of GitHub repos, those recommendations aren’t as useful to you. By training on your team’s code, IntelliCode learns patterns from your code to make more accurate suggestions. This enables your team to accelerate learning and take advantage of the knowledge of your team and broader community.
AI-assisted Refactoring
IntelliCode watches code changes as they occur in the IDE and locally synthesizes, on demand, edit scripts from any set of repetitive pattern changes. It uses these edit scripts to produce suggestions, enabling you to apply repetitive changes quickly or create a pull request to apply the suggestion(s) for team review without distracting your current work. IntelliCode refactorings take the time-intensity and error-proneness out of routine tasks, such as introducing a new helper function. To do so, IntelliCode uses an AI technology called program synthesis, and more specifically, programming-by-examples (PBE).