Amazon has launched CodeWhisperer, a pair programming AI tool similar to GitHub Copilot

This week, Amazon announced the debut of CodeWhisperer, an AI-powered pair programming tool that can autocompletion whole functions with just a remark or a few keystrokes, much to GitHub Copilot. The business used billions of lines of open-source code and its own codebase, as well as publicly accessible documentation and code on public forums, to train the system, which supports Java, JavaScript, and Python at now.

For developers who prefer Visual Studio Code, IntelliJ IDEA, PyCharm, WebStorm, or Amazon’s own AWS Cloud 9, it’s now available in preview as part of the AWS IDE Toolkit. The AWS Lambda Console will also be supported in the near future.

Today, Amazon announced a new AI service called Copilot, but the company’s vice president of AI services, Vasi Philomin, made it clear that this wasn’t just a clone of Copilot. For the business’s first public debut in a few years, he said, the company has already built the framework with AI code reviewer and performance profiler CodeGuru, as well as the operation problem finder DevOps Guru.

This decision was made because the technology was at a place where “we felt it was appropriate,” Philomin added. “It also blends in well with the rest of their decor. This is a trip that we’ve taken at various points in life,” he said.

A tiny group of Amazon engineers have been beta testing the service internally, mostly to keep the announcement a secret.

Even your own coding style and variable names are taken into consideration by the machine, according to the business. A bespoke code snippet is generated using the current context and the location of your cursor.

It’s important to note that CodeWhisperer differs from other tools like Copilot in many ways. First, although most of its generated code is completely new and unique, if it creates a piece of code that is extremely similar to an existing function in its training data, it will record that and indicate its licencing information. Finally, the developer may determine whether or not they want to utilise it. Using a tool like this may relieve some (but not all) of the copyright worries.

In addition, Philomin emphasised the importance of safety. When analysing the code, Codewhisperer will look for any possible security concerns. This is based Now’s own experience managing huge codebases and conducting debriefs after things go wrong (using their established ‘repair of mistakes’ method), as well as CodeGuru.

It’s critical to Philomin that the code he generates is safe since security is a top priority for AWS. In order to execute a scan on the source code, Codewhisperer may simply say, “Run a scan on the current source file.”” It will do a security check and alert you to any problems or security vulnerabilities in the created or updated code.”

Additionally, Philomin emphasised that the Codewhisperer team worked hard to guarantee that the code it creates does not generate any biassed code. There are filters in place to automatically delete the code if it occurs.

User and developer interaction is simple and intuitive. Developers may choose from a variety of code recommendations (in the demo I saw, there were usually at least two options). It has the ability to autocomplete comments and then recommend functions based on those remarks. Something like #See whether an integer is prime is an easy example of this kind of thing.

In addition, although the system will operate well for developers outside the AWS environment, Philomin emphasised that the team took great effort to guarantee that it would perform really well for developers who wish to use AWS services (think #Create an S3 bucket).

AWS API first-class support is just a gimmick. Our measurements and tests have shown it to be exceptional, he added, and he expects that when finished, the product will be cutting edge. Since the service is now available for testing by developers, we can expect to hear more about how well it works and how it compares to initiatives like Copilot in the near future. Having additional options in this area is a good thing (and somebody is surely going to hack up a project that will automatically provide suggestions from multiple pair programming AIs).