Donnie Flood

Finally trying out Wispr Flow. I'm not usually someone who talks at their devices, but I figured if this is going to be normal, I should learn how to do it and challenge myself. Side benefit is you can talk faster than you can type.

screenshot of a diff with content: ## Visual Consistency - Sections that present the same concept (e.g., numbered step cards) should use the same visual treatment site-wide. - The standard step/feature card style: `rounded-xl border border-border bg-white/[0.03] p-6` with number + title + description. - Don't create alternate layouts (centered open text, timeline dots, etc.) for content that's structurally the same as an existing card grid.

Here's a diff that was suggested by Claude for design rules to ensure better website design outcomes in the future. Will it work? Who knows??? Vibes-based context engineering isn't measurable!

SWE-Bench is not YOUR codebase - ContextBridge Blog SWE-Bench and other public benchmarks are better than vibes but they don't tell you how these models will perform on YOUR codebase. Here's why you need a benchmark for your codebase and what it can do...

The real lever for AI coding performance isn't just which model you pick — it's context engineering. Your rules, your docs, your code. But without a benchmark on YOUR codebase, every change is just guesswork.

Wrote more about this: contextbridge.ai/blog/swe-ben...

Two major model releases in 24 hours. Everyone's comparing SWE-Bench scores and feeling the vibes. But you wouldn't hire an engineer based on LeetCode alone — why pick AI coding tools that way? Your codebase is unique. Generic benchmarks can't capture that.