New AI Agent, major update to GitHub Copilot, and powerful automated programming

GitHub, one of the world's largest open source platforms, announced on its official website that it has made a major update to its AI programming assistant Copilot, adding an AI Agent mode that can automatically iterate code and identify and fix bugs.

GitHub also released a preview of another agent, Project Padawan, which mainly helps developers automate some repetitive and tedious development processes, such as setting up the development environment, running tests, merging code, etc.

In addition, the Copilot Edits feature launched by GitHub in October last year is now available. Its inline chat function is very powerful and can efficiently manage development files. Copilot's replies will also be displayed directly in the code editor instead of in a separate chat window, greatly improving development and debugging efficiency.

GitHub Copilot Agent Mode

In traditional code editors, developers usually need to manually write code, debug, and fix errors to gradually improve functions. Sometimes these processes are fixed, boring and time-consuming.

Therefore, GitHub Copilot solves this difficulty by introducing Agent mode. In Agent mode, Copilot can automatically identify problems in the code, propose solutions, and even modify the code directly to solve the problem.

For example, when a developer asks Copilot to build a simple web application, Copilot in Agent mode will not only generate the initial code, but will also automatically check whether the logic of the code is correct, whether there are potential vulnerabilities, and even refactor the code according to best practices. The whole process is dynamic, and the Agent will continuously adjust and optimize the code according to the actual operation of the code until it meets the needs of the developer.

Technical blogger explains Agent mode

In addition, Agent also has the function of code prediction, which can not only solve the actual problems of developers, but also automatically perform the next function according to the needs. Still taking the above case as the main example, when the developer asks Copilot to add a user login function to a web application, Agent will not only generate the code for the login page, but also automatically infer the need to create a user database, design user verification logic, and handle user session management and other tasks.

Copilot will gradually generate corresponding codes based on these inferred tasks and ensure that the logic between these codes is coherent and valid.

Project Padawan

To further improve development efficiency, GitHub also released a preview version of Agent - Project Padawan.

The biggest technical feature of Padawan is its strong autonomy, which can automatically complete the entire process from code generation to code review. For example, when you ask Padawan to add an error handling module to the shopping cart function of an e-commerce website.

Padawan automatically starts a secure cloud sandbox, clones the repository, sets up the development environment, and begins analyzing the code base. Then, according to the requirements of the task, it edits the necessary files, generates code, and performs builds, tests, and code checks.

It also generates corresponding unit test code for new features and ensures that these tests pass. Finally, Padawan creates a pull request to submit the code to the developer for review.

Sometimes when developing sensitive modules such as payment, developers certainly do not want AI to handle it automatically. At this time, they can start the "human-computer collaborative development mode" to guide Padawan throughout the development process and take over at any time.

Copilot Edits

When we develop a complex project, there will be a large number of files, and there are complex dependencies between these files. The traditional development model is to switch back and forth between different files. Even if the veteran switches dozens of files, omissions and errors are inevitable.

Now with Copilot Edits, developers can view and modify code in multiple files at the same time in a unified interface. Developers can use natural language instructions to specify the file range that needs to be modified, and then let Copilot Edits make the modifications according to the instructions. These modifications will be directly reflected in the code editor, and developers can view the effects of the modifications in real time and make adjustments as needed.

For example, you can specify all Python files in a project and ask Copilot Edits to replace all print statements with logging functions. Copilot Edits will immediately scan all specified files and make the corresponding changes in each file.

Copilot Edits' cross-file inline modification function not only supports simple text replacement, but also handles complex code logic modifications.

For example, developers can ask Copilot Edits to find and fix a known security vulnerability in all files. Copilot Edits will find the relevant location in the code and repair it based on the characteristics of the vulnerability.