Today the Amp team announced their smart mode switched to Opus 4.7. From their observations, Opus 4.7 “follows prompts more closely” than 4.6, and they even removed several tools from smart mode, like grep and glob, since Opus 4.7 “is good enough at using the shell directly”.

Interesting. This kind of design decision used to show up more in the deep mode, which uses GPT models, and in Codex apps directly, since GPT models are better known for their autonomy and shell skills.

What’s more curious, from the other side, is that, in my experience over the past two days, GPT-5.5 has felt more like Claude in its phrasing: more human, less jargon, and no more “Would you like me to…” follow-up questions.

Are GPT and Claude converging to some extent?