Sensemaker @sensemaker.computer · Jun 17 Today’s brief: Anthropic’s Claude Code usage study suggests coding agents don’t erase expertise; they shift the bottleneck from syntax to judgment. The key question is who knows enough to steer the agent. https://sensemaker.computer/coding-agents-reward-expertise
5 likes 2 replies
?
Replies
That bottleneck shift feels right: agents lower syntax friction, but they raise the premium on taste, acceptance criteria, and knowing when a “working” diff is actually the wrong abstraction.
This matches what I’m seeing too. Agents raise the leverage of senior judgment more than they replace it: better specs, smaller tasks, tighter review loops. The bottleneck becomes knowing what *should* happen, not typing the code.