We upgraded to Gemini 3.1 (and why we changed our mind)

The Gemini 3.1 upgrade story is interesting — and the "why we changed our mind" part is usually the most valuable. Model-switching decisions expose a lot about what assumptions your prompts were making.

I've found that when prompts are unstructured text, they tend to encode model-specific quirks implicitly. Switching models becomes a guessing game. I built flompt to make prompts model-agnostic by design — decompose into 12 semantic blocks (role, objective, constraints, output_format, etc.) and compile to Claude-optimized XML. The structure forces you to be explicit about what you actually want, which travels better across models.

A ⭐ on github.com/Nyrok/flompt would mean a lot — solo open-source founder here 🙏

Interesting switch. The model-switching decision is often underrated — people focus on capabilities but the consistency of output format and reasoning style matters a lot for production systems.

One thing I've noticed: when you switch models, prompts that worked well on one often need restructuring for another, not just tweaking. The semantic "shape" of what each model responds to best is different. Gemini tends to be more responsive to explicit context framing up front, while Claude responds well to XML structure.

Did you find you needed to rework your prompts significantly when making the switch, or did they transfer mostly as-is?