Anthropic Latest Survey: Heavy AI Adopters Most Optimistic About Personal Income and Employment

According to Dynamis Beating monitoring, Anthropic has released its latest AI Economy Index report, combining survey data from 9,700 users with telemetry data to reveal the impact of artificial intelligence on work patterns and career prospects. The report found that the higher the value of a job, the more AI compute power it consumes: the average compute power consumption for high-paying positions is 2.07 times that of low-paying positions. For example, the compute power consumed by a Marketing Manager to write a proposal is 2.5 times that of an editor revising an article. However, high-paying positions also have exceptions, such as high-income Pharmacists whose daily AI usage compute power is only one twentieth of a Statistical Assistant's.

The survey shows that over a third of users expect AI to take over most of their work within a year. Surprisingly, those who are most willing to delegate their work entirely to AI for automatic execution are the most optimistic about their future income and job prospects, and are not concerned about their skills atrophying. When using different AI tools, users exhibit varying degrees of delegation: when writing articles on a web interface, users typically engage in back-and-forth edits with AI for an average of 13 rounds; however, when using the terminal tool Claude Code, users usually issue a single command and let AI generate the output in one step.

In terms of autonomy ratings, when users employ the command-line tool Claude Code, they are more inclined to let AI make decisions autonomously, scoring an average of 0.37 points higher than the web interface (0.26 points higher for the same model). The only exception is in handling data tables: web interface users mostly engage in intellectually demanding financial modeling tasks, while Claude Code users mainly use it for mechanical data extraction. Therefore, in table-related tasks, the autonomy of AI on the web interface is actually 0.35 points higher. The responses generated by AI are typically more insightful than the users' initial queries; in design and game development tasks, the educational understanding threshold of AI responses averages nearly two years higher than that of user queries.

Telemetry data has also depicted a profile of overtime among different groups: during non-working hours such as nights and weekends, high-paying positions are more likely to use AI for overtime work, with the proportion of high-paying job tasks increasing by 8%, while mid-low-paying positions see a decrease of 4% to 11%. Gender analysis has also shown distinct collaboration preferences: female users tend to prefer iterative collaboration, with the proportion of using the command-line tool and tasks under automation decreasing by 6.3 and 7.3 percentage points, respectively, but the total time spent interacting with AI through chat is longer, indicating a higher propensity for discussion.