The 'bedside manner' point about Claude landed differently for me than it probably would for a VC.
I spent two months running both Claude Code and Codex on the same production codebase. The technical capabilities are closer than the internet discourse suggests. What actually made Claude stick was how it handled ambiguity - it defaults to asking rather than guessing, which matters enormously in complex projects with implicit context.
The 'specialized coding models failed' observation connects to something I saw too. Context quality beats model specialization, almost every time.
And what will power all of these... a Terminal. Just like it always has been!
... and $10m a year and the talent war feels inflated from all sides. LOLWHUT.
The 'bedside manner' point about Claude landed differently for me than it probably would for a VC.
I spent two months running both Claude Code and Codex on the same production codebase. The technical capabilities are closer than the internet discourse suggests. What actually made Claude stick was how it handled ambiguity - it defaults to asking rather than guessing, which matters enormously in complex projects with implicit context.
The 'specialized coding models failed' observation connects to something I saw too. Context quality beats model specialization, almost every time.
Full comparison with actual usage patterns: https://thoughts.jock.pl/p/claude-code-vs-codex-real-comparison-2026