I shouldn't need to disable a plugin like that. In my opinion, it's something that I should need to explicitly enable if I want it. All it takes is one person on a project accepting the prompts to agree to the ToS for everyone. They know it's problem too because commercial accounts can disable the integration for everyone.
And most of what I said was relating to the direction Jetbrains is moving. As soon as they start adding service based integrations, it's a slippery slope and we could end up with IDEs that depend on calling out to Jetbrains' services to function.
That's the exact opposite of what I want from Jetbrains and keeping that kind of stuff out of the IDE is the reason I pay them every year.
I guess the main reason the plugin is enabled by default is to promote it. AI tools are becoming an important part of the development workflow, with a majority of developers using a chat LLM or Copilot (https://www.jetbrains.com/lp/devecosystem-2023/ai/). So the functionality offered by the plugin might be interesting to a large part of the user base. A non-bundled plugin would not be discovered by most users. And for organizations it looks like the AI assistant is disabled by default.
Regarding the direction - LLMs are usually used as a service right now, so there's not much choice. At the same time, JetBrains is also working on local smart completion workflows like https://plugins.jetbrains.com/plugin/14823-full-line-code-co.... Hopefully at some point there'll also be an option to use a local LLM with the AI Assistant.
They call it a plugin, but it's an ad. It wouldn't be any different if they put an ad in the IDE and installed the plugin when you sign up.
> Hopefully at some point there'll also be an option to use a local LLM with the AI Assistant.
I would love something where I could pick the inputs or at least give it inputs that skew results. Right now I think the big problem with AI is bad input. Most of the dataset that's publicly available is junk. A lot of "answers" are people guessing at stuff they don't understand.
Honestly, I don't even know if (or how) LLMs work if you train them with a smaller, more accurate dataset.
GitHub copilot is not enabled by default or installed on vscode, yet tons and tons of people use it. Mostly because it's actually pretty good compared to Jetbrains implementation. You can't force stuff onto your users and not expect some backlash though. Especially for a paid product that could've just as easily made it easy to disable or remove.
Copilot is a 2-year-old product, it's expected that many people use it. I agree it currently may be better than AI Assistant. But LLM functionality is here to stay in IDEs, and most people will use it in one way or another.
Nobody forces stuff onto you. You can disable everything with just one checkbox.
And most of what I said was relating to the direction Jetbrains is moving. As soon as they start adding service based integrations, it's a slippery slope and we could end up with IDEs that depend on calling out to Jetbrains' services to function.
That's the exact opposite of what I want from Jetbrains and keeping that kind of stuff out of the IDE is the reason I pay them every year.