I've always taken the example that at any given moment openai and anthropic have models that are "far better and exceed any expectations", yet their coding tools suck ass and are outperformed by random people on the internet.