I’ve reached the point where I feel I need to create another post, this time specifically about Gemini.
Since around Wednesday (at least that’s when I first noticed it), Gemini has become almost unusable for software development. The degradation has been significant enough that it no longer feels like the same models I’ve been using for months.
The most concerning issue is that it often behaves as if it only sees the current prompt. It appears to ignore previous conversation context, even in relatively short and recent chats. Context that was available just a few messages earlier is suddenly forgotten, making multi-step development workflows extremely unreliable.
I’m also seeing problems with the Antigravity ecosystem. Gemini frequently fails to recognize available tools or skills, such as /teamwork-preview, and behaves as if those capabilities don’t exist.
Jules has become another source of frustration. Running with Gemini 3.1 Pro, it no longer consistently fixes the requested issues, doesn’t automatically create a PR after completing the work, and often requires an explicit prompt just to perform the PR creation. Even worse, it regularly introduces new bugs, breaks existing features, bypasses TypeScript errors by adding things like @ts-ignore simply to make the project build, and ignores architectural patterns and coding standards that are already established in the repository.
What makes this especially confusing is that I’ve been using both Gemini 3.1 Pro and Gemini 3.5 Flash extensively for months, and they were genuinely excellent. They handled complex coding tasks reliably, respected project structure, and generally produced high-quality work. Now, both models feel noticeably worse, even on relatively simple tasks.
One particularly strange example happened while using Google’s own AI Mode in Search. I asked about these issues, and Gemini responded with a long apology that referenced problems with “Gemini 1.5 Pro,” as if it were relying on outdated training information or failing to use the search capabilities that would normally recognize we’re now using Gemini 3.1 Pro. That response alone made me question whether something deeper is currently broken.
Overall, the experience has changed dramatically. I’ve been on Google AI Pro for months with very few complaints, but over the past few days the quality has dropped to a point where these tools are actively slowing down development instead of helping.
At the same time, there has been very little communication from Google. Gemini 3.5 Pro is still delayed, while Gemini 3.1 Pro and Gemini 3.5 Flash currently feel unreliable for day-to-day development work.
I’m posting this because I’d like to understand whether others are experiencing the same regression. If this is a temporary issue, some acknowledgement from Google would be incredibly helpful. Right now, though, the current state of the ecosystem is frustrating enough that I’m seriously considering moving away from it.