The rise of large language models has been swift and loud, echoing from social media feeds to corporate earnings calls. Yet beneath the hype lies a more complicated reality: their impact on the workforce has been far from uniform.
Data from the Anthropic Economic Index shows how unevenly different industries are engaging with Claude, Anthropic's AI chatbot. The company compared each sector’s share of conversations with Claude to its share of the US workforce. Most industries are underrepresented—meaning they account for a smaller portion of AI interactions than their share of employment would suggest. Only a few are meaningfully overrepresented, including arts, design, entertainment, sports, and media. Computer and mathematical jobs, which include roles like software engineering and data science, represent less than 5% of the US workforce, but nearly 40% of the conversations with Claude.
Some of the disparity in LLM usage can be explained by the fact that tech employees often work at the bleeding edge. They are quick to find and experiment with new technologies before they gain traction in other industries. As LLMs make their way into new industries, companies like Anthropic and OpenAI could see a surge in usage by less technical occupations and roles.
But another, more consequential, explanation is that software development is uniquely well-suited to benefit from LLMs. First, start with the abundance of training data for code-related tasks. GitHub hosts tens of millions of public and open source repositories, representing the collective of work of over 100 million developers around the world. Coding patterns and standards are potentially easier for AI models to recognize and learn compared to other forms of language. Second, consider the vast amount of infrastructure most companies have to validate new code against clear acceptance criteria. Over the last decade, companies have invested heavily in QA and CI/CD automations, which developers can use to quickly test and iterate on code changes. Such systems can help steer and correct AI-generated code.
Unlike previous waves of automation, high paying knowledge workers are among the first to adopt these new tools. According the Bureau of Labor Statistics, the median pay for software developers in the US is $130,160 (as of 2023). Software development-related jobs also represent the largest percentage of conversations with Claude.
What does this mean for the future of software developers? The job market for developers in the US has slowed significantly since the pandemic. Many companies claim to be doing more work with fewer developers with the help of AI tools. Whether software developer salaries drop in response to these two trends is hard to forecast.
But wages aren’t shaped by automation alone. A range of other factors—some harder to quantify—are also at play. For example, AI can help semi-technical employees write simple scripts or automate routine tasks, reducing the need to rely on engineers for everyday technical work. While these workers won’t be building complex systems, they can take care of an increasing amount of low-level development work, freeing engineers to focus on high-priority, complex projects. This dynamic could even push wages upward: as AI handles more of the basics, expert developers become even more valuable, commanding a premium to work on the hardest problems.