Working Professionals No Longer Just Want to Chat

Google Antigravity
,
1

It seems the community has been quite heated lately regarding the quota issues in Google Antigravity.

Honestly, it feels a bit ironic to complain about the reduced quota for the competitor’s Opus 4.6 model within Google’s own Antigravity platform. After all, the Gemini 3.1 Pro model provides us with a much larger quota.

However, people who use Google Antigravity aren’t looking for an LLM that just flatters them with conversational responses. There is really only one purpose for using Antigravity: Vibe Coding.

So, let’s rethink the core problem. Fundamentally, what we need isn’t an increased quota for the Opus 4.6 model within Google Antigravity, but rather a significant enhancement in the vibe coding capabilities of Gemini 3.1 Pro.

While I can’t speak for everyone, the era of using various AI models merely for simple chatting is long gone. Now, people are leveraging AI to build multi-agent systems, construct automated workflows, and develop applications.

I previously submitted a request regarding the quotas, but I’d like to revise my stance. I will be patiently waiting for an upgraded version of Gemini 3.1 Pro.

Until that time comes, I will temporarily downgrade my current Google Ultra subscription to Pro and take a trip to the neighboring platforms.

Of course, I don’t feel that Gemini falls behind Opus in every single aspect. When it comes to design tasks, such as utilizing HTML and CSS, my experience is that Gemini actually performs better.

However, when it comes to fixing issues or updating an app that has already grown extensive, Gemini 3.1 Pro has a high tendency to overcomplicate the problem or break existing features. Moreover, its unnecessarily overly user-friendly responses are actually quite frustrating; they waste tokens and make it feel as though the model is just trying to gloss over its own mistakes (this is, of course, strictly my personal opinion).

Since I am sharing my perceived experience rather than exact statistical data, please take this simply as a point of reference.

I will wrap up my post here. I always support your continuous development and progress!

1 Like
3

Indeed, you can clearly see Gemini 3.1 wasn’t really made for coding but more as a general chat-bot thingy. You will see that every prompt you send it now starts with:

”Prioritizing Specialized Tools

I’m focusing intently on using specific tools instead of general ones. I’m actively avoiding ‘cat’ for file operations and replacing general grep with ‘grep_search’. My goal is to use tailored tools, not generic commands, for all these tasks, improving efficiency and accuracy. I’m particularly interested in related tools for complex file manipulation.”

You can clearly see they have to coach the LLM more than they can with a base prompt. If anything this start has helped it from constantly spitting out commands that only work on linux or some other platform.

It is a bit funny that sometimes its tools will come up with wanting to run a console command like this: echo “Hello”

I am not kidding, it frequently comes up with these odd things. Like just now I wanted it to write a plan and it goes a head and wants to compile my project out of the blue. It seems that it has problems adding results to the output too, like plans made, suddenly it wont show up - you have to locate them yourself lol.

2 Likes
4

would that affect the tokens limit ? and how to fix?