Will Larson queried how leaders evaluate engineering managers for AI skills. Sashank Gondala shared lessons from building an AI-native org at Further AI over nine months. Key improvements include a 5x increase in daily code merges per engineer with revert rates under 0.5%.
The shipping process was rebuilt step-by-step: AI agents now write 100% of code, 91% of changes include tests (up from 15%), and AI reviews 99% of changes (up from 60%). CI runs 20x more tests in 17 minutes. Bug intake is automated via Claude Tag, filing 60% of tickets.
Tribal knowledge converted to 21 automated rules. Up to a third of changes merge without human involvement. Non-engineers, including designers, now ship code.
Engineering output jumped from 17-30 customer-facing changes per engineer monthly to 45 in September and 48 in October. The new bottleneck is decision-making, not coding speed. While coding accelerated 3.5x, planning and review lag, creating a mismatch in delivery pace.
Source: Hacker News · Summarized by HeadlinesBriefing