TL;DR
Anthropic engineers report 50% productivity gains using Claude, with 27% of work being entirely new tasks, but face concerns about skill atrophy and changing collaboration patterns.
Key Points
- Claude usage increased from 28% to 59% of daily work in 12 months; productivity gains jumped from +20% to +50%
- 27% of Claude-assisted work wouldn't have been done otherwise—scaling projects, documentation, exploratory work previously deemed too costly
- Engineers can 'fully delegate' only 0-20% of work; most collaborate iteratively with Claude rather than hand off tasks without verification
- Claude Code now handles ~20 autonomous actions before needing human input (up from ~10 six months ago); complex task usage grew from 1% to 10% for design/planning
Why It Matters
This research provides concrete data on how AI transforms developer workflows at scale—revealing both productivity multipliers and structural risks like skill erosion, reduced peer mentorship, and uncertainty about long-term career trajectories. For engineering teams adopting AI tools, it highlights the need for intentional strategies around knowledge preservation, code review rigor, and junior developer mentorship in an AI-augmented workplace.
Source: www.anthropic.com