Now loading...
Artificial intelligence is reshaping the daily routines of software engineers at Anthropic, the company behind the Claude AI models, according to new internal research from the firm. The study, based on surveys and interviews with employees in August 2025, shows that workers are accomplishing more tasks in less time, expanding their skills into unfamiliar areas, but also grappling with concerns over declining expertise and reduced human collaboration.
Anthropic, a leading developer of AI systems, turned its research focus inward to examine how its own tools are altering workflows among its engineers and researchers. The company polled 132 staff members, held detailed discussions with 53 of them, and reviewed usage logs from its Claude Code feature, which integrates AI assistance directly into coding environments. The results paint a picture of accelerated productivity tempered by unease about long-term professional changes.
Engineers reported relying on Claude for 60 percent of their daily tasks, up from 28 percent a year earlier, and estimated a 50 percent overall productivity increase compared to 20 percent previously. This boost stems largely from handling more work volume in categories like debugging and code comprehension, even as time per task slightly decreases. Daily use is highest for error correction, with 55 percent of respondents turning to the AI for that purpose, followed by 42 percent for understanding existing code and 37 percent for building new features.
One striking outcome is that 27 percent of AI-assisted efforts involve projects that likely would not have happened without the tool, such as enhancing data dashboards, conducting exploratory analyses, or addressing minor code inefficiencies often overlooked due to time constraints. However, full handover of responsibilities remains limited; most said they could completely delegate just 0 to 20 percent of their duties, preferring instead to actively oversee and refine the AI’s suggestions, especially in critical areas.
Interviews revealed evolving strategies for task assignment to Claude. Workers favor delegating verifiable, low-risk, or tedious jobs, like routine debugging or repetitive scripting, while reserving strategic planning and creative design for themselves. Boundaries are shifting as the technology improves, with some noting a growing comfort in using AI even for moderately complex implementations. Yet, this reliance raises fears of skill erosion. Participants expressed worry that quicker AI-generated solutions might reduce opportunities for deep learning, potentially hindering their ability to scrutinize outputs effectively.
Many described themselves as turning “full-stack,” tackling front-end designs or database work beyond their usual specialties, which fosters faster innovation and bolder project scopes. One engineer highlighted how Claude enabled building a sophisticated user interface that would have otherwise stalled the team. Still, others lamented less hands-on practice, likening it to skipping foundational steps in coding education that build enduring proficiency.
Social aspects of work are shifting too. Claude often serves as the initial resource for queries, cutting down on colleague interactions and mentorship moments. While some welcome the efficiency, others miss the collaborative spark, noting fewer chances for junior staff to seek guidance from seniors. Usage data supports these trends: Over six months, Claude handled more intricate assignments autonomously, chaining up to 20 tool actions without input versus 10 earlier, and required 33 percent fewer human interventions.
Team variations underscore the tool’s versatility. Security specialists use it heavily for dissecting unfamiliar code, while alignment researchers craft visualizations, and non-technical users troubleshoot basics like network glitches. Across the board, 8.6 percent of sessions fix small irritants that enhance overall efficiency, from refactoring to shortcut creation.
Looking ahead, Anthropic acknowledges its unique vantage as an AI pioneer with early access to advanced models like Claude Sonnet 4 and Opus 4, available at the study’s time. The findings may preview wider industry shifts, prompting discussions on preserving expertise, fostering teamwork, and adapting careers in AI-enhanced settings. The company is exploring internal adjustments, including professional growth programs and best practices for AI integration, while sharing survey details for broader replication.
Anthropic has outlined potential economic policies in related work, such as ideas for addressing AI’s labor impacts. Full details of the research are available on the company’s site.
