The human in the loop is tired

Sources: Pydantic article, HN discussion
Summary: A Pydantic article (dated 2026-02-18, resurfaced on the Hacker News front page 2026-07-17) argues that LLM-assisted programming automates the satisfying parts of coding and replaces them with supervisory work, so reviewers face fatigue and rubber-stamp large plausible-looking diffs as volume rises. It frames expertise as a quality gate and argues developers should distill judgment into clearer specifications rather than review every action.
Why it matters: It names a failure mode where human approval in agent workflows degrades into rubber-stamping once agent output outpaces review capacity.