How AI is transforming work at Anthropic

Source: Anthropic · 2025-12-02

Turning the lens inward, in August 2025 we surveyed 132 Anthropic engineers and researchers, conducted 53 in-depth qualitative interviews, and studied internal Claude Code usage data to find out how AI use is changing things at Anthropic. We find that AI use is radically changing the nature of work for software developers, generating both hope and concern.

Our research reveals a workplace facing significant transformations: Engineers are getting a lot more done, becoming more “full-stack” (able to succeed at tasks beyond their normal expertise), accelerating their learning and iteration speed, and tackling previously-neglected tasks. This expansion in breadth also has people wondering about the trade-offs—some worry that this could mean losing deeper technical competence, or becoming less able to effectively supervise Claude’s outputs, while others embrace the opportunity to think more expansively and at a higher level. Some found that more AI collaboration meant they collaborated less with colleagues; some wondered if they might eventually automate themselves out of a job.

More capable AI brings productivity benefits, but it also raises questions about maintaining technical expertise, preserving meaningful collaboration, and preparing for an uncertain future that may require new approaches to learning, mentorship, and career development in an AI-augmented workplace. We discuss some initial steps we’re taking to explore these questions internally in the Looking Forward section below. We also explored potential policy responses in our recent blog post on ideas for AI-related economic policy.

Anthropic engineers and researchers use Claude most often for fixing code errors and learning about the codebase. Debugging and code understanding are the most common uses (Figure 1).

People report increasing Claude usage and productivity gains. Employees self-report using Claude in 60% of their work and achieving a 50% productivity boost, a 2-3x increase from this time last year. This productivity looks like slightly less time per task category, but considerably more output volume (Figure 2).

27% of Claude-assisted work consists of tasks that wouldn't have been done otherwise, such as scaling projects, making nice-to-have tools (e.g. interactive data dashboards), and exploratory work that wouldn't be cost-effective if done manually.

Most employees use Claude frequently while reporting they can “fully delegate” 0-20% of their work to it. Claude is a constant collaborator but using it generally involves active supervision and validation, especially in high-stakes work—versus handing off tasks requiring no verification at all.

Skillsets are broadening into more areas, but some are getting less practice. Claude enables people to broaden their skills into more areas (of software engineering (“I can very capably work on front-end, or transactional databases... where previously I would've been scared to touch stuff”), but some employees are also concerned, paradoxically, about the atrophy of deeper skillsets required for both writing and critiquing code—“When producing output is so easy and fast, it gets harder and harder to actually take the time to learn something.”

Changing relationship to coding craft. Some engineers embrace AI assistance and focus on outcomes (“I thought that I really enjoyed writing code, and I think instead I actually just enjoy what I get out of writing code”); others say that “there are certainly some parts of [writing code] that I miss.”

Workplace social dynamics may be changing. Claude is now the first stop for questions that used to go to colleagues—some report fewer mentorship and collaboration opportunities as a result. (“I like working with people and it's sad that I ‘need’ them less now...

Claude is handling increasingly complex tasks more autonomously. Six months ago, Claude Code would complete about 10 actions on its own before needing human input. Now, it generally handles around 20, needing less frequent human steering to complete more complex workflows (Figure 3). Engineers increasingly use Claude for complex tasks like code design/planning (1% to 10% of usage) and implementing new features (14% to 37%) (Figure 4).

Employees self-reported that 12 months ago, they used Claude in 28% of their daily work and got a +20% productivity boost from it, whereas now, they use Claude in 59% of their work and achieve +50% productivity gains from it on average. (This roughly corroborates the 67% increase in merged pull requests—i.e. successfully incorporated changes to code—per engineer per day we saw when we adopted Claude Code across our Engineering org.) The year-on-year comparison is quite dramatic—this suggests a more than 2x increase in both metrics in one year. Usage and productivity are also strongly correlated, and at the extreme end of the distribution, 14% of respondents are increasing their productivity by more than 100% by using Claude—these are our internal “power users.”

Engineers and researchers are developing a variety of strategies for productively leveraging Claude in their workflow. People generally delegate tasks that are:

Many users described a progression in their Claude usage that involved delegating increasingly complex tasks over time: “At first I used AI tools with basic questions about Rust programming language... Lately, I've been using Claude Code for all my coding.”

People consistently said they didn’t use Claude for tasks involving high-level or strategic thinking, or for design decisions that require organizational context or “taste.” One engineer explained: “I usually keep the high-level thinking and design. I delegate anything I can from new feature development to debugging.” This is reflected in our survey data, which showed the least productivity gains for design and planning tasks (Figure 2). Many people described delegation boundaries as a “moving target,” though, regularly renegotiated as models improve (below, the Claude Code usage data shows relatively more coding design/planning usage now than six months ago).

The survey finding that 27% of Claude-assisted work wouldn't have been done otherwise reflects a broader pattern: engineers using AI to work outside their core expertise. Many employees report completing work previously outside their expertise—backend engineers building UIs; researchers creating visualizations. One backend engineer described building a complex UI by iterating with Claude: “It did a way better job than I ever would’ve. I would not have been able to do it, definitely not on time... [The designers] were like ‘wait, you did this?’

Engineers report “becoming more full-stack... I can very capably work on front-end, or transactional databases, or API code, where previously I would've been scared to touch stuff I'm less of an expert on.” This capability expansion enables tighter feedback loops and faster learning—one engineer said that a “couple week process” of building, scheduling meetings, and iterating could become “a couple hour working session” with colleagues present for live feedback.

At the same time, some were worried about “skills atrophying as [they] delegate more”, and losing the incidental (or “collateral”) learning that happens during manual problem-solving:

One reason that the atrophy of coding skills is concerning is the “paradox of supervision”—as mentioned above, effectively using Claude requires supervision, and supervising Claude requires the very coding skills that may atrophy from AI overuse. One person said:

Honestly, I worry much more about the oversight and supervision problem than I do about my skill set specifically... having my skills atrophy or fail to develop is primarily gonna be problematic with respect to my ability to safely use AI for the tasks that I care about versus my ability to independently do those tasks.

​​Engineers diverge sharply on whether they miss hands-on coding. Some feel genuine loss—“It’s the end of an era for me - I've been programming for 25 years, and feeling competent in that skill set is a core part of my professional satisfaction.” Others worry about not enjoying the new nature of the work: “Spending your day prompting Claude is not very fun or fulfilling. It's much more fun and fulfilling to put on some music and get in the zone and implement something yourself.”

Some directly addressed the trade-off and accepted it: “There are certainly some parts of [writing code] that I miss - getting into a zen flow state when refactoring code, but overall I'm so much more productive now that I'll gladly give that up.”

One of the more prominent themes was that Claude has become the first stop for questions that once went to colleagues. “I ask way more questions [now] in general, but like 80-90% of them go to Claude," one employee noted.