Follow ZDNET: Add us as a preferred source<!–> on Google.
ZDNET’s key takeaways
- An Anthropic study aimed to quantify the economic impacts of AI.
- It found that AI could double the US economy’s annual growth rate.
- Time-saving with AI varies significantly across industries.
Tech companies have promised time and time again that the technology will boost human productivity, and by extension, economic growth. But accurately measuring the legitimacy of such claims is difficult, since individual employees and organizations are using AI for different purposes, and it’s not yet clear that it can deliver the ROI that businesses have hoped for.
A new study from Anthropic, however, shines some light on the economic impacts of AI that can be expected over the next several years.
Also: OpenAI tested GPT-5, Claude, and Gemini on real-world tasks – the results were surprising
Published last week, the study analyzed 100,000 anonymized user conversations with Claude to find out how long various tasks would take with and without the chatbot’s help. The researchers found that, on average, Claude helped users complete tasks around 80% faster than it would’ve taken without its assistance.
The researchers extrapolated outward from there, estimating that current AI models will boost total labor productivity in the US by an annual rate of 1.8% over the next 10 years – roughly double the current annual growth rate.
The findings
One of the driving goals behind the new Anthropic study was to understand the degree to which AI can practically assist with specific tasks across industries.
Also: Anthropic’s new warning: If you train AI to cheat, it’ll hack and sabotage too
It’s one thing to claim that Claude will transform legal professions, for example; it’s quite another to try to understand, in precise detail, how the day-to-day lives of lawyers and judges will be impacted by AI. It’s also insufficient to tally the number of tasks that AI helps someone complete in a particular day. As the Anthropic researchers write in their report, “a software developer might use Claude to write 10 pull requests in a day, but if nine are minor documentation updates and one is a critical infrastructure change, simply counting the number of these tasks performed with Claude misses the point.”
The study therefore aimed to get a concrete grasp of not only “which tasks Claude handles, but how substantial those tasks and time savings are.” To that end, the researchers prompted Claude to generate two estimates for each of the 100,000 transcripts: the length of time it would’ve taken that user to complete the relevant task on their own, and the length of time it took to handle it with Claude’s help.
Using data pulled from 2024 Occupational Employment and Wage Statistics (OEWS), the researchers then calculated the amount of time and money that Claude’s assistance helped save on specific tasks across a range of different professions.
Also: 3 ways AI agents will make your job unrecognizable in the next few years
For financial analysts making $58 per hour, for example, the study found that when Claude was used to analyze economic data impacting investment decisions – which without the use of AI would take a little under one hour – the chatbot reduced total completion time by 80% and saved $43. On a related note, another study published in September by the New York University Stern School of Business and wealth management platform GoodFin found that some industry-leading AI models could pass the Chartered Financial Analyst Level III exam, which is basically the bar exam for aspiring financial professionals.
The median amount of time saved on tasks across the full dataset of conversations with Claude was 84%, though the researchers note that there was “considerable variation across tasks and categories.” The study found, for example, that some types of vocations are more amenable to seeing productivity boosts from AI than others. Management, education, and construction were among the job categories for which Claude helped produce the highest time savings on specific tasks, while the savings for personal care, sales, and office support were significantly lower.
On the whole, Anthropic estimate that Claude’s findings translate to a 1.8% increase in annual US labor productivity, with software development jobs leading most of that growth (19%), followed by general and operations managers (6%), market research analysts and marketing analysts (5%), customer service representatives (4%) and secondary school teachers (3%).
Anthropic’s study was published just one day after the company released its latest AI model, Claude Opus 4.5, which it said outperforms other industry-leading models in software engineering tasks (though ZDNET’s David Gewirtz tried it himself, and it didn’t really go that way). The company also called Opus 4.5 “a preview of changes to how work gets done.” Anthropic’s CEO, Dario Amodei, predicted earlier this year that AI could replace humans in as many as half of all entry-level white collar jobs within the next decade.
Caveats and future impact
Anthropic based its study on the assumption that the capabilities of AI models (and humans’ ability to use them efficiently) will remain steady over the next 10 years. This is unlikely to be the case, however, as the rate of technological progress over recent decades strongly suggests that AI will become much more advanced in the coming years.
Also: Does the CEO doubt your AI savvy? CIOs must upskill in these 3 critical areas
You can already see it happening: ChatGPT turns three years old this weekend, and already we have systems that can do much more than respond to user queries in natural language. Many frontier models and AI agents can interact with digital tools like web browsers and documents, “reason” through multi-step tasks over extended time frames, speak in eerily human-like voices, and assist with online shopping, just to name a few examples.
“This estimate should be taken as an exercise exploring what might happen based on current usage patterns [emphasis in original], not a prediction of the impact on productivity that is actually most likely to happen,” the Anthropic researchers wrote.
The study also exclusively draws upon conversations with Claude, and could therefore be biased toward analyzing tasks which users feel are best suited for that particular chatbot. Findings from chatbots with differing strengths could reveal alternate takeaways.
Also: How AI-enabled autonomous business will change the way you work forever
Most importantly, when comparing the length of time it takes humans to complete tasks with and without Claude’s help, the study fails to account for the additional time committed to those tasks outside of the conversation with the chatbot. For example, a teacher using Claude to develop a lesson plan might take extra time to fact-check its recommendations.
Anthropic acknowledges each of these limitations and frames its study as an early effort to understand and quantify the impacts of AI on the US economy, one that could provide a blueprint for future studies.
Artificial Intelligence
–>