Opus 4.7 dropped, I checked the coding benchmarks, and by the end of the day I had completely rewired how I build SayMei. Until this week, my default coding model was Gemini 3.1 through Anti-Gravity. It is still excellent for design, but the new benchmarks made the call obvious: switch the harness, switch the model, do not stay loyal to a setup just because it has been working. Today I was dual-wielding Cursor Pro and Claude Code on the Max plan, sometimes with three Cursor agents running in parallel on top of that. My MacBook was loud enough that I could hear it from across the room. Here is the new stack: - Primary execution: Opus 4.7 in both Cursor and Claude Code - Design iteration: Claude Design The difference shows up on long, multi-step work. Earlier today Opus 4.7 ticked through a 17-step plan touching LiveKit and agent rooms in a single clean run. The same kind of task used to cost Gemini three or four shots with documentation open. That extra headroom let me ship the thing I have been wanting to do for weeks: a full rebuild of the SayMei design system from the ground up, using Claude Design. I turned it into a design skill, retired the old one, then went page by page tightening UI, copy, and responsive behavior for both mobile and desktop. The site finally feels consistent end to end. The takeaway: do not get too attached to one tool or one model. Switching the harness almost always unlocks another 30-50 percent productivity increase, and the setup that was right last month is rarely the right one this month. Attaching a shot of our new live globe section, built with Claude Design and Opus 4.7. Check out saymei.app to see the dynamic globe midway down the landing page. #DeveloperTools #Cursor #ClaudeDesign #Opus4.7
Things have come a long way since agent.ai 😊