Monday, February 2, 2026
Claude Code Skills and Impact
People are discussing Claude Code, particularly its "Skills" feature, its perceived performance compared to previous versions, and its potential to shift the bottleneck in software development from labor to defining product vision.
As mentioned in the article, the disparity is about to become absolutely insane… I’ve only just recently started getting into things like running Skills with Claude Code myself, but until just a little while ago I thought that simply being able to use ChatGPT well already put
Here’s the real, hard data proving AI is getting dumber: Claude Opus 4.5 This is from the Marginlab team. They run daily pass@1 tests using Claude 4.5 Opus (code mode) on 50 problems from SWE-Bench-Pro. From the numbers: Early January → ~60% Now → 54% → Clear
Check out the Claude code skills github. Just give Claude code that link and ask it to make a skill for the thing and use the GitHub as reference for how to format it etc. It will just do it as long as you give it all the ingredients