[1/2] How AI is Transforming Work at Anthropic (AI column)
With great interest. Anthropic internal study about how to use their AI assistant (Claude) It affects the work of the company’s engineers. Of course, Anthropic is an AI company in itself, so its employees are in privileged conditions: they are among the first to access the most advanced AI tools and work in the field directly related to the development of AI. Therefore, the authors emphasize that the findings may not be fully generalized to other organizations.
In the first post, I will try to summarize the insights for engineers, and in the second, talk about the insights for managers.
Productivity increased Developers are now using Claude around the 60% of their work (against ~28a year ago) They estimate productivity gains themselves.50%. That's more than a double jump in a year. Measurements confirm the trend - for example, the number of successful daily pool requests for an engineer has increased by one year. 67Percentage of implementation of Claude Code
Claude helps with routine tasks Most often it is attracted for debugging bugs and parsing someone else's code ~55The percentage of engineers do it every day. Around 42% use AI to understand the code 37% to write new functions. Less often asked for architectural design or data analysis (These tasks prefer to do themselves.)
New challenges are now within reach 27The percentage of the work that engineers do with Claude would never have been done before. AI frees up time for things like internal “nice-to-have” tools (dashboard) Experiments that would be too expensive by hand. In addition, Claude takes on minor improvements:8.6% of its tasks are repairing small bugs and refactoring, which before the hands did not reach. These little things over time add up to a tangible gain in quality and speed of work.
Delegates, but with due regard Most people believe that without verification, you can only trust AI to 20Percent of tasks. Claude has become a regular collaborator, but not a standalone artist - the developer still checks and guides it, especially in important things. Engineers have developed an intuition that to entrust tasks primarily simple to verify, low-risk or boring. ("draft" code, routine parts). Gradually, they are entrusted with increasingly complex work, but the architecture and final design decisions are controlled by themselves.
Skills are wider, depth is in question With Claude, people are bolder at taking on tasks beyond their core expertise – everyone is gradually becoming more full-stack engineers. For example, a backend developer with the help of AI can go to both the frontend and the database, instead of calling specialized specialists. However, there is a downside: when AI does routine, engineers are less practised in the basics and basic knowledge can gradually “atrophy”.
Attitudes to writing code are changing Some are happy to be able to focus on concepts and outcomes rather than writing code. Experienced engineers compare this paradigm shift to a shift to higher-level languages. Many people are willing to put up with the loss of part of the pleasure, because productivity is now much higher.
Less live communication Claude is increasingly the first person to ask questions instead of colleagues. It saves time. (Don't bother your little partner.)But mentoring suffers. Experienced developers note that the Juns are less likely to seek advice, because Claude has a lot to teach them. Some people don’t like that the phrase “Did you ask Claude?” has become commonplace.
Career and future The role of an engineer shifts to managing AI systems instead of writing every line of code. Many already feel more like a teammate for a couple of AI-agents than just a developer: for example, one assesses that the company is not an AI-agent. 70Percent became an AI code reviewer/editor rather than an author from scratch. Productivity is off the scale, but in the long run people are not sure what will result in their profession. There is optimism for the near term, but then there is uncertainty.
Continuation in the next post.
#Engineering #Software #Processes #Productivity #Economics