Ok, see a lot of user reporting errors or issues they are having but was kind of hoping to get some inspiration on this forum.
Kind of sounds I am the only one quite happy with Anti Gravity 
Ok, let me share how I use it, I have an architecture.md that explains my solution and agents.md explaining what I want I don’t want. And noticed that is very important to have a whitelist that you allow Antigravity to run, in my case I develop in .Net / C# so I whitelisted: dotnet build, dotnet watch (handy for interactive browser sessions), dotnet test (for running unit tests) and dotnet run.
Meaning it can always run them in Agent mode without asking explicit permission, which is fine for all these commands. You don’t want to be bothered when it wants to check if the solution builds and unit tests are still passing.
I blocked: az, sudo, ssh, ansible-playbook. Blocked means it still can run it but need to ask for permission which I want in these cases.
So not really a pro tip maybe but seen this not mentioned a lot but settings these correctly already helps a lot.
Also make sure that you solution is GIT controlled, YOU control GIT I typically have VS 2026 running alongsite where I do the GIT because it has a nice GUI for git control. When agent make changes that work but you are not finnished and dont want to push the changes then already stage them.
So if in your next Agent run something breaks you can see clearly the difference with the last run you already commited (but not yet pushed).
Also I typically use Opus to generate a Task + Implementation plan out of the description I made of what I want. And after reviewing / changing when I am happy with Task and Implementation plan switch to Google Pro 3 for executing the plan.
Also cant hurt in Plan mode to mention that you do not want the system to create any code until you approve it. Otherwise if often starts implementation when I am not yet finished creating or brainstorming the Implementation plan.
I typically work only on 1 plan at a time. Also when a session becomes long ask it to create a handover that you can use in a new session with fresh context.
Ok, what are your tips for getting the most out of Antigravity, lets beside sharing bugs also share what works for you and things that work for you and other might not know.
Looking forward to reading your replies, also if you think what I described earlier can be improved I’m interested in hearing it.
Also I typically work on on 1 thing at a time, I read that Antigravity supports running multiple agents at a time but not sure when to use it and if it makes things faster and easier or more complex and perhaps slower due to more need for bugfixes?