![Main image](


alt="Article illustration 1"
loading="lazy">

)

The Experiment

In December 2025, Anthropic released a comprehensive research report titled *How AI Is Transforming Work at Anthropic*. The study combined a survey of 132 engineers, 53 in‑depth interviews, and an analysis of 200,000 Claude Code transcripts. The goal was to move beyond the high‑level economics of AI and look at the *human* side of the transformation.

Productivity in Numbers

The data show a dramatic uptick in Claude usage. Engineers now report using the model in 59 % of their work, up from 28 % a year earlier, and a self‑reported productivity gain of +50 %—more than double the previous year’s +20 %.

“I use Claude for debugging on a daily basis,” one engineer noted, echoing the survey’s finding that 55 % of respondents rely on Claude for debugging.

![Figure 1: Proportion of daily users for various coding tasks]() The survey also highlighted that 27 % of Claude‑assisted work would not have been undertaken otherwise. These include scaling projects, building interactive dashboards, and tackling “papercuts”—small quality‑of‑life fixes that were previously deprioritized.

Skill Shifts and the Full‑Stack Paradox

One of the most striking themes is the *full‑stack* effect. Engineers who traditionally specialized in backend logic are now building front‑end UIs, while researchers are creating data visualizations.

“I can very capably work on front‑end, or transactional databases, or API code, where previously I’d been scared to touch stuff,” an engineer explained.

However, this breadth comes with a cost. Many respondents voiced concerns about skill atrophy:

“I worry that my hands‑on coding skills are atrophying because I rely on Claude to write the code.”

Some engineers countered that the shift mirrors historical transitions—from assembly to high‑level languages—suggesting that the industry may be moving toward higher‑level abstractions where AI becomes the primary code generator.

Changing Social Dynamics

Claude has become the *first‑stop* for many questions that once went to colleagues. This filtering effect reduces routine interactions but also risks eroding mentorship opportunities.

“I like working with people and it’s sad that I ‘need’ them less now,” one engineer lamented.

Meanwhile, a minority of teams reported unchanged collaboration patterns, indicating that the impact varies across organizational cultures.

Career Uncertainty and New Roles

With AI taking on more routine tasks, engineers are increasingly framing themselves as *AI managers*—overseeing Claude instances, reviewing AI output, and ensuring alignment with project goals.

“My work has shifted 70 %+ to being a code reviewer/reviser rather than a net‑new code writer.”

The study found widespread uncertainty about the long‑term trajectory of software engineering. Some fear obsolescence, while others see new opportunities for higher‑level design and strategy.

Usage Trends: A Data‑Driven Lens

Claude Code usage data corroborate the survey findings. Over six months, the average task complexity rose from 3.2 to 3.8 on a 1‑5 scale, and the average maximum consecutive tool calls jumped by 116 %.

“Claude now chains together 21.2 independent tool calls without need for human intervention versus 9.8 six months ago.”

The distribution of tasks also shifted: feature implementation grew from 14 % to 37 %, and design/planning from 1 % to 9.9 %. ![Figure 4: Changes in Claude Code usage between August 2025 and February 2025]() Team‑level analysis revealed that different groups leverage Claude for distinct purposes—security teams focus on code understanding, while pre‑training teams build new features. ![Figure 5: Team‑specific Claude Code usage]()

Looking Forward

Anthropic is treating this research as a living laboratory. Planned next steps include:

  1. Refining best practices for AI‑augmented work through an AI fluency framework.
  2. Extending studies beyond engineers to other roles in the organization.
  3. Collaborating with external partners (e.g., CodePath) to adapt curricula for an AI‑assisted future.

The organization acknowledges that these shifts mirror past transitions in software engineering, but the pace and scale of AI’s impact may create a new paradigm.

Takeaway

Anthropic’s internal data paint a complex picture: AI is boosting productivity, expanding skill sets, and reshaping collaboration, yet it also introduces uncertainty about skill preservation and career paths. The story underscores the need for thoughtful strategies—both organizational and individual—to navigate an AI‑augmented engineering landscape.

Source: Anthropic Research (2025). https://anthropic.com/research/how-ai-is-transforming-work-at-anthropic/