Unlock Editor's Digest for free
FT editor Roula Khalaf has chosen her favorite stories in this weekly newsletter.
Financial analysts use chatbots to investigate possible risks and returns from investment programs. A task that would take 50 minutes can be completed in 10 minutes.
This is one real-world example that AI company Anthropic recently claimed to have discovered when it asked its Claude chatbot to analyze how it is used in the workplace. But showing how these one-off tasks translate into real business value for employers is less straightforward.
This is expected to be one of the key fronts in the battle between the big AI companies in 2026. OpenAI CEO Sam Altman recently said the company is shifting its focus to enterprise customers in an effort to grow revenue. This is a market where Anthropic currently leads.
But even if people start leveraging generative AI in the workplace, most companies won't be able to measure whether this technology makes individual employees more effective at their jobs, much less track productivity gains at an enterprise-wide level.
Exploring the impact of new technologies like AI on the economy as a whole is even more difficult. It is well known that the impact of IT on overall labor productivity is difficult to determine from official data. The impact of digital technology was not visible in U.S. economic data for many years, until productivity growth began to rise steadily starting in the late 1990s. However, by the beginning of this decade, growth had returned to its previous rate of about 1.5% per year.
The encouraging news for AI companies and their investors is that many people are starting to find uses for generative AI in their work. Summarizing long reports, drafting marketing presentations and analyzing financial data are the types of tasks employees are attempting for the first time this year. If any of these use cases become widely established, the implications for the use of AI models could be significant.
So far, generative AI has one “killer app” working in the form of a coding assistant used by software developers. The effect was explosive. According to OpenRouter's AI model usage study, in May of this year, 11% of all tokens generated by large language models were related to coding. By November, that percentage had risen to about 50 percent.
Employees themselves certainly believe that AI is starting to make them more efficient. Earlier this month, OpenAI announced that after surveying its employees, they were saving 40 to 60 minutes per day thanks to AI. That's up from the 2.2 hours a week workers believed they were saving in a similar study conducted by the St. Louis Fed a year ago.
Because such self-reports are highly subjective, Anthropic's study of real-world tasks could be more revealing. Based on 100,000 work-related conversations, Claude estimated that the average task could save 65 minutes from 85 minutes.
But, as Anthropic is the first to admit, demonstrating virtuosity at individual tasks does not directly translate into business advantage for customers. This figure doesn't tell you, for example, how much extra work goes into checking the output from the chatbot or how the overall quality affects the results.
Additionally, a single job may result in multiple chat sessions. The ability to easily and quickly get results from chatbots can lead to employees writing more reports and emails, leading to a cascade of unproductive “work gaps.” Claude also doesn't know how workers are using the time that technology has saved them.
Another drawback is that the task-based analysis that is central to most research on the impact of technology on productivity does not capture the realities of working life. For most people, work does not fall into discreet, self-contained segments. As Anthropic acknowledges, looking at single tasks in isolation doesn't capture the tacit knowledge, personal relationships, or connections between different tasks that influence how work gets done.
This would seem to explain the counterintuitive results of one study this year, which found that a group of experienced developers took 19% longer to complete tasks when using AI coding tools.
The full benefits of generative AI will only be realized if companies redesign their entire work processes to take full advantage of the technology and overcome the cultural barriers that stand in the way of this type of change. But the race continues, as workers begin to experiment with AI.
richard.waters@ft.com
