Minor Problems in Antigravity that can be fixed to vastly improve the user experience

Google Antigravity
1

I have been using Google’s Antigravity for the past couple of months, and I absolutely love it. However, there are certain issues which really degrade the user experience.

  1. Commit Messages: The commit message generator in the source control tab always assumes that the files I have staged are new even if I have edit them. This means that even when I make bug fixes or refactors on existing files, the commit messages generated convey that the changes implement a new feature. For this reason, I have to keep cursor open just for writing the commit messages or have to write them myself.

  2. Code search via grep: I am unaware if a tool for this exists in Antigravity, but it would randomly use terminal commands to search the code base which is annoying because I have to keep coming back to approve it. If it always used a code search tool, the dev experience would be far better.

  3. Documentation: Right now, if the agent doesn’t understand why a linter error is happening for a given library, it visits the node_modules filter instead of looking up the documentation for that library. While this fixes the linter errors most of the time, this approach is also prone to causing runtime errors.

  4. Linter errors: Unlike Cursor, Antigravity is unable to detect linter errors in it’s code unless I mention it, at which point it uses commands to detect those errors and keeps running the same command to detect if it has fixed the linter errors. This is bad for UX because I keep having to come back to approve the same command, besides that, if it was able to detect the linter errors automatically, many requests can be fulfilled in a single prompt.

  5. Allowlist for commands: As the previous issue suggests, sometimes users have to keep allowing the same “safe” commands repeatedly. It would be much more ergonomic if an allowlist existed that allowed users to pre-approve commands.

Once again, I really love the IDE and appreciate the improvements the team has been making. For instance, the “Proceed” button in the Implementation Plan page was a great recent addition.

2

i totally agree! its almost great but not quite, the allowlist is a big letdown, such a simple mechanism not working well at all.

also the problem with agent manager gets stuck loading agent forever.

if both of those are fixed it’ll be perfectly happy with AG

1 Like