Google recently released DiffusionGemma, and it's weird in the best way.
After months of testing local LLMs, I found that productivity depends on tools, not just models.