Anthropic's May 19 London event demonstrated developers shipping pull requests written entirely by Claude 4.6 and 4.7, with many reporting no manual review. Boris Cherny stated the default workflow now involves Claude prompting itself for error correction and testing. The new dreaming feature lets agents store and consolidate task notes across sessions to improve performance on specific codebases. Primary source: https://www.technologyreview.com/2026/05/21/1137735/anthropics-code-with-claude-showed-off-codings-future-whether-you-like-it-or-not/. Related coverage in Anthropic's official Claude Code updates from February and April 2026 confirms the same automation push. OpenAI's parallel Codex claims at I/O events show identical patterns of reduced human code authorship at scale. The event coverage omitted longitudinal data on code quality degradation or maintenance costs after initial deployment. Patterns from prior AI coding rollouts at Microsoft and Google indicate similar short-term velocity gains followed by rising technical debt within 12-18 months.