What AI Really Does at Work: Insights from Millions of Conversations with Claude

The rise of AI tools like Claude, ChatGPT, and Gemini has reshaped how many of us approach daily work. From debugging code to job hunting, generative AI is quietly becoming a digital co-worker.

But for all the buzz, there’s been a lack of clear data on how AI is actually being used in the workplace. That’s what makes a new study from Anthropic, in collaboration with academic partners, so exciting.

By analyzing over 4 million anonymized conversations with Claude, a popular AI assistant, the researchers mapped AI usage across the American economy. The result is one of the most detailed portraits yet of where and how AI is making a real impact at work.

Where AI is Working the Hardest

Let’s begin by answering the most important question: where is AI used the most?

The data reveals that AI is heavily concentrated in software development and writing. Together, these account for nearly 50% of AI usage in the study. If you’ve ever used an AI assistant to write code, brainstorm blog headlines, or fine-tune copy, you’re part of this growing trend.

What do these tasks have in common? They’re text-based, creative, and iterative, perfectly suited to how today’s language models work. But it’s not just engineers and writers who are getting AI help. Some unexpected roles also ranked high in AI usage.

For instance, people seek AI support for therapy advice (Claude is not a licensed therapist), educational tutoring, and business strategy. These interactions may not replace professionals, but they offer informal, lightweight help, like a second opinion or a sounding board.

It’s a sign that AI isn’t just a productivity tool; it’s also becoming a thinking partner.

AI Isn’t Everywhere… Yet

While AI is making inroads across many white-collar jobs, it’s absent in a surprising number of roles, especially those involving the physical world. Construction workers, nurses, and anesthesiologists showed very low levels of AI engagement.

However, this shouldn’t come as a huge surprise, considering that these roles involve hands-on work, sensory judgment, or safety-critical decisions, all of which are challenging for AI. Even so, the data show pockets of adoption in unexpected places.

According to the study, 36% of occupations use AI in at least 25% of their core tasks. That means even in roles where AI doesn’t dominate, it still supports important parts of the job. Rather than replacing entire roles, AI is chipping away at specific tasks.

This aligns with what many AI experts have long argued: the future of AI at work isn’t replacement—it’s augmentation.

Augmentation vs. Automation

So, what exactly are people using AI for?

The researchers found that 57% of AI usage supports augmentation, helping users brainstorm, revise, or learn. The remaining 43% involves automation, where the AI performs a task more or less on its own, like translating text, summarizing articles, or generating code from prompts.

Tasks that are highly structured and information-based are more likely to be automated. On the other hand, tasks that require judgment, iteration, or creativity are more often augmented.

This distinction matters. It helps demystify fears about AI taking over jobs wholesale. The reality is far more nuanced: AI is a flexible tool that adapts to the way people work, offering support where it makes sense and staying in the background where it doesn’t.

AI Usage by Wage and Training Requirements

One of the most interesting parts of the study is how AI usage correlates with wages and training. Higher-wage, higher-education jobs tend to use AI more.

These include roles like software engineers, lawyers, and marketers—jobs with complex, information-heavy tasks. However, even in middle-wage occupations, there’s meaningful AI usage, especially in roles that involve communication, scheduling, or document preparation.

Lower-wage, less-formal roles—like drivers, waiters, or janitorial staff—see much less AI involvement. This suggests that current AI tools may be reinforcing existing divides in the workforce, offering more leverage to knowledge workers while bypassing others.

Still, this also opens the door for upskilling. As AI tools become more accessible, even non-technical professionals could benefit from learning how to prompt, evaluate, and collaborate with AI.

What This Means for the Future of Work

This study reinforces a key trend: AI is not replacing jobs, it’s redefining them.

As AI becomes more embedded in the workplace, job descriptions may shift. Roles will likely evolve to focus more on judgment, communication, and creative oversight while routine tasks get delegated to machines.

For individuals, this means adapting to a world where knowing how to work with AI becomes as essential as knowing how to use Excel or Google Docs. For companies, it means rethinking training, productivity metrics, and workflows.

Upskilling will be critical. Knowing how to frame a good prompt, critically evaluate an AI response, and combine machine outputs with human insight may become baseline skills across many industries.

Conclusion

AI is already changing how we work, one task at a time. From writing and coding to learning and brainstorming, tools like Claude are becoming quiet partners in the background of millions of workflows.

But AI isn’t everywhere yet, and it doesn’t have to be. Its real strength lies in helping us think faster, communicate better, and get more done.

As we head into a future shaped by human-machine collaboration, understanding where AI fits today is the first step to preparing for tomorrow.
Want to read more about AI’s impact on work? Check our Artificial Intelligence articles.