Anthropic Publishes Internal Research: How Claude Is Changing Engineering Work
Anthropic has published new research today examining how Claude is transforming work for its own engineers and researchers — giving an unusually candid look at the real-world effects of AI on a highly technical workforce. The study surveyed 132 Anthropic engineers and researchers and compared findings to a similar internal survey conducted a year earlier. The results show dramatic adoption growth alongside honest concerns about skill atrophy and job satisfaction that are worth taking seriously as the industry grapples with AI's effects on knowledge work.
Key findings
- Adoption has nearly doubled: Engineers now use Claude in approximately 60% of their work tasks, up from 28% a year prior. Self-reported productivity gains have grown from 20% to 50% over the same period.
- New work unlocked: A striking 27% of completed work "would not have been done at all" without Claude — suggesting the productivity gain is not just speed on existing tasks but the enablement of entirely new initiatives that lacked sufficient time or confidence to begin.
- Full-stack expansion: Engineers report becoming more comfortable working outside their primary expertise. One engineer noted: "I can very capably work on front-end, or transactional databases…where previously I would've been scared to touch stuff." Claude lowers the barrier to tackling unfamiliar parts of the codebase.
- Claude Code autonomy growing: Claude Code now executes approximately 20 consecutive actions autonomously, up from 10 six months prior — a doubling of agentic reach that is visible in everyday engineering workflows.
The concerns Anthropic is naming openly
- Deskilling anxiety: Some experienced engineers worry about losing hard-won proficiency. Several report deliberately practicing tasks without Claude specifically to prevent skill atrophy — a behaviour that should not be necessary in a healthy AI-augmented workflow.
- Satisfaction erosion: A subset of engineers reported that "spending your day prompting Claude is not very fun or fulfilling," missing the "zen flow state" of independent problem-solving. One 25-year veteran described it as "the end of an era."
- Reduced peer collaboration: With Claude available as an on-demand collaborator, some engineers report consulting colleagues less frequently — a potentially unintended consequence for team knowledge transfer and mentorship.
Publishing this data openly — including the concerns — is notable. Most AI companies highlight productivity wins and downplay friction. Anthropic is naming the deskilling problem directly, which gives the industry a more honest framework for thinking about AI-augmented work design.