This page may contain stale information. Last updated: 2026-08-14
Definition
Autonomous app maintenance is the use of coding agents on scheduled routines to perform mechanical software hygiene — crash fuzzing, dead-code removal, flaky-test fixes, abstraction unification — with humans still in the merge path.
Key Points
- anthropic internal deployment via claude-code + Slack (
proj-claude-maintains-apps) - Reported KPI: 388 PRs / 180 merged (~46%) after Claude + human review (boris-cherny)
- Specialized routines outperform generic “fix everything” prompts
- Review-burden risk: agent throughput can overwhelm human reviewers (ai-code-review-bottleneck)
- Complements safety research on multiagent-turf-war — autonomy ↑ requires orchestration controls