JetBrains has unbundled its AI Assistant from the IDE, in a new beta of IntelliJ IDEA, which is also the basis for many other developer tools.
The company released its AI coding service in December but soon received objections that the tool was installed by default and hard to remove. “Your new update with the AI Assistant is impossible to uninstall. It keeps prompting me for updates and I cannot turn it off,” complained a developer on the JetBrains issue tracker.
One of the reasons for wariness is that “When you use AI-powered features, the IDE sends prompts to the LLM provider. In addition to the prompts you type, the IDE may add context information, such as pieces of code, file types, and frameworks used,” according to the FAQ.
JetBrains insisted that no data would be sent unless “the user has explicitly consented to use the AI Assistant and is using the trial or has purchased the corresponding subscription,” but also acknowledged that the plugin could not be completely removed.
That has been resolved in the first beta for IntelliJ IDEA 2024.1. “AI Assistant has been unbundled and is now available as a separate plugin,” says the introductory post. The plugin is now described as “freemium” though it is not yet clear what value a non-paying user gets since the FAQ refers only to a seven-day trial.
User reviews for the AI Assistant though are currently not good. Comments this month include “it completely hallucinates about what the library does and how it is used”, “currently it’s a terrible experience to use” and “Having tested it side-by-side with Github Copilot, I can definitely say that the AI Assistant is close to useless – at least when used in a big JS/TS project.”
There is plenty more in the new beta aside from making AI Assistant removable. IntelliJ IDEA 2024.1 includes Java 22 support, which is timely since this new JDK (Java Development Kit) is currently a release candidate with full release expected on March 19th.
There is also a new terminal window with code completion for commands and a new color scheme; sticky lines in the editor which keeps lines key to the code structure visible as the developer scrolls (similar to sticky scroll in Visual Studio Code); and full line code completion for Java which predicts entire lines of code based on the context.
Full line code completion uses machine learning baked into the IDE, but according to JetBrains developer advocate Matt Ellis: “The ML plugins are not related to the AI Assistant. They provide on-device models for code completion and search everywhere, do not send any data off-device, and are included in your existing subscription.”