David Park
Tried out Gemini 3.0 today. I was previously a power user of Gemini 2.5. I coded primarily with Claude 3.5 sonnet but I’ve found that Gemini 2.5 was really good at doing audits of large codebases. My biggest issue with Claude 3.5 is that it always wants to change something. If you do an audit of a perfectly fine codebase, Claude will try to make stylistic changes. It NEEDS to change SOMETHING. If it can’t find something to fault objectively, it will instead find it subjectively.
Gemini doesn’t seem to have this problem as much. Gemini does have a tendency to fall into existential spirals but that’s another topic for another time.
Vibe coding feels like slapping together an app just well enough to present like a functioning app. Some times it feels like a $10/hour dev shop + a decent designer: Don’t look under the hood if you don’t want to be disappointed.
Gemini seems to be better at the more computer science-y stuff. I had Gemini do an audit of my codebase and gave it a list of problems that I had. I specifically asked it to look for ways to optimize concurrency for server calls and to search for blocking server calls that were unnecessary. I also asked it to find middleware issues that were causing latency issue.
The benefit of coding from Vietnam with servers in the US is that latency issue gets amplified. It’s a lot easier as a tester to FEEL the sticky points in the user experience.
Over the last year, so many new LLM models, AI frameworks, and AI tools have released nonstop, not to mention updates and upgrades to existing tools. At first, every new release was met with hype and fanfare and I fell into the excitement and wanted to try everything out. I had this feeling that if I got ahead of the curve and learned new tools before everyone else, I would have an edge for a small window of time. Very quickly, I realized that it's impossible to keep up with everything. The point of all of these AI tools is to increase our productivity, but ironically, trying to keep up with all of them is the ultimate productivity killer.
However, Gemini has been a LLM product that I've been bullish on for a long time. Trying to talk to Gemini is like trying to talk with an autistic savant. Most people will write off Gemini, because on the surface it seems strange. But the longer you talk with it, the more you realize that there are certain things that it is just incredibly better at than other models.
Strange metaphors aside, I'm glad that I'm unstuck now and excited to keep working on SponsorRadar.