Heavy AI users are most optimistic about their pay and job prospects, Anthropic finds
Anthropic's latest AI Economic Index report, combining survey data from 9,700 users with telemetry, reveals how AI is reshaping work and career outlook. The report finds that the higher the value of a job, the more compute it consumes: high-salary roles use on average 2.07 times the compute of low-salary ones. For example, a marketing manager writing a proposal uses 2.5 times the compute of an editor revising an article. But there are exceptions: a well-paid pharmacist uses just one-twentieth the compute of a statistical assistant.
More than a third of survey respondents expect AI to take over most of their work within a year. Surprisingly, those who delegate the most to AI are the most optimistic about their future income and employment — and aren't worried about their skills atrophying. How much control users hand over varies dramatically by tool: when writing on the web, users go back and forth with AI for an average of 13 rounds of edits; but when using the terminal tool Claude Code, they typically give one command and let AI produce the final result in one go.
On autonomy scores, users let Claude Code make more independent decisions, scoring 0.37 points higher than web use (0.26 points higher when controlling for model). The exception is spreadsheet work: web users tend to do financial modeling that requires planning, while Claude Code users mostly perform mechanical data extraction. So for spreadsheet tasks, the web-based AI actually scores 0.35 points higher on autonomy. The AI's final outputs are generally more sophisticated than users' initial prompts — in design and game development tasks, the educational level needed to understand AI responses is nearly two years higher than that of the prompts.
The telemetry also paints a portrait of overtime work: after hours and on weekends, it's high-salary workers who use AI more, with their share of tasks rising 8%, while mid- and low-salary workers see a 4% to 11% drop. Gender analysis shows different collaboration preferences: women are more likely to iterate collaboratively — they use command-line tools and fully automated tasks 6.3 and 7.3 percentage points less, respectively, but spend more total time chatting with AI in back-and-forth discussions.