The best part is when their own tools and the application itself cause issues, without taking responsibility for the errors themselves.
AG-859
The Agent can’t return/respond unless it literally maxes out the response. Once this starts, it’s like a snowball, and unless you catch it within the first few steps, start a new conversation, and pray that the new Agent doesn’t try to look up any conversation history before it starts working…
Meanwhile, I’m doing everything I can to save on tokens… Disabling autocomplete, forcing a max of four tabs open at a time, preventing the Agent from reading the open tabs directly, even starting new conversations to try to be more efficient and not drag a history with me when moving to a new task….
But then I learn there is a bug where when you start a new conversation, it can consume a ton of tokens…
Not to mention the Gemini degradation, where you ask for a single task to be completed, and the Agent has decided that it wants to change your application to be something completely different. The whole “One step forward, two steps backwards” is the status quo for Gemini this week. Except, quite literally, I’m making one step forward and two to six steps backwards.
So, while I was trying to be a nice player, and use Gemini Pro and Gemini Flash primarily, because I understand this is easier/cheaper for Google, I am forced to use Claude when I have available tokens, as Gemini is just completely broken right now.
I’ve done quite literally everything that I can do, including building a system which is able to store my conversation live, on the fly. This means I have highly detailed insight for the steps taken, commands, etc, as well as historic records that can clearly show these errors, and some of the crazy stuff going on, which in no way show any evidence the User is doing anything wrong, instead, Antigravity is just hobbled and if they don’t have the bases working, then I can’t really expect them to be able to pick up on an account being impacted, and automatically providing a token refund, while Google takes ownership of the issues until they are addressed.
Over a month after launch, https://antigravity.google/ is still a joke, with no account page, platform health / notifications, and laughable “Changelog”, while the app itself still links to it for “Help” or “Troubleshooting”, as well as a black-hole for a Feedback system. More people may provide feedback if they felt that in any way it would actually induce change; Without being able to see my Feedback History, or any other Antigravity-specific account information… its just a bit hopeless.
So, yeah, it’s up to Google, and how much they want to see this succeed… and we’re all waking up to other options…