Top Uses of Claude AI for Work Tasks Revealed by Anthropic

As AI continues to weave its way into our daily lives, the big question on everyone's mind is: what's going to happen to the job market? Well, Anthropic, the folks behind the Claude chatbot, are diving deep into this mystery with their newly released Economic Index. Instead of just guessing about the future, they're taking a closer look at how people are actually using AI in their work right now.

What's unique about Anthropic's approach is that they're not just looking at job titles. They're focusing on the tasks people do and how AI fits into those tasks. It's like saying, "Hey, whether you're a designer, a photographer, a security screener, or a radiologist, you all use visual pattern recognition in your work." By zeroing in on these common threads, Anthropic hopes to get a clearer picture of AI's impact.

To gather their data, Anthropic used their own Clio system to analyze a whopping 1 million anonymized conversations between users and the Claude chatbot. They then matched these conversations to the Occupational Information Network (O*NET), a US Department of Labor database that lists 20,000 work tasks. This helped them figure out which tasks the AI was most involved in and how these tasks relate to different job categories.

So, what did they find? Well, it turns out that software engineering tasks were the most common, making up 37.2% of the conversations. This includes things like debugging code and network troubleshooting, which makes sense since Claude is designed with coding in mind. The next big category was writing and editing, at 10.3%, which falls under jobs like copywriting in the arts and media sector.

Interestingly, these two job categories only make up a small slice of the US economy—3.4% for software engineering and 1.4% for arts and media. Yet, they're using AI at a much higher rate than other sectors like office admin and sales. Jobs in science and education also showed a higher use of AI compared to their overall presence in the economy. Within these categories, tasks like producing for entertainment, conducting research, and creating educational materials were among the top uses for AI.

Augmenting vs. Automating Work

The report also sheds light on whether AI is replacing human workers or just helping them out. It turns out that in 57% of cases, AI is augmenting human capabilities—think of it as a helpful sidekick that double-checks your work, helps you learn new skills, or assists with brainstorming. On the other hand, AI directly performs tasks for people 43% of the time, which is what we call automation.

Now, Anthropic admits there's a catch: they can't be 100% sure that these conversations were all about work. They're just going off the fact that the queries match up with occupational tasks. Plus, they don't have data from API, Team, or Enterprise users, which might give a different picture. And there's another twist—whether users took Claude's suggestions as is or tweaked them outside the app can blur the line between augmentation and automation.

Other Findings

Using O*NET's median salary data, the study found that AI is more commonly used in mid-to-high wage jobs like data science. Meanwhile, people in the lowest and highest wage bands, like salon workers or doctors, are less likely to use AI. This might be because these roles involve more hands-on work, or because current AI tech just isn't up to the task yet.

The study also found that about 4% of jobs use AI for at least 75% of their tasks, while 36% of jobs use AI for at least 25% of their tasks. This gives us a sense of how deeply AI is integrated into different roles.

Takeaways

Anthropic plans to keep an eye on these trends by regularly updating their analysis. They're curious to see if certain roles are becoming more automated over time. "If AI continues to be used for specific tasks and only a few jobs rely heavily on AI, we might see most current jobs evolve rather than disappear," they note.

The report doesn't dive into policy recommendations, but Anthropic is open to feedback from economists, policy experts, and other researchers. They've even open-sourced the conversation data for anyone who wants to take a closer look or add their own insights.