Anthropic’s Economic Index just dropped the most comprehensive data on how AI changes work: 9,700 survey respondents linked to actual Claude usage data. 86% report productivity gains. 35% expect AI to handle most tasks within 12 months. And a paradox: the more automated your job, the MORE optimistic you are about your future.
Anthropic Economic Index — Key Numbers
86%
Report productivity speed gains
35%
Expect AI to handle most tasks in 12 months
68%
Report learning more because of AI
9,700
Respondents linked to real usage data
The Paradox: More Automation = More Optimism
The most counterintuitive finding: people whose work is more automated by AI are MORE optimistic about their job prospects, not less. Higher automation share correlates with more positive expectations about future employment.
The self-reported productivity data is overwhelmingly positive:
86% report productivity speed gains
82% report expanded scope of work
69% report quality improvements
68% report learning more because of AI
57% feel their skills are growing more valuable
But there’s a dark side: 33% expect a junior colleague’s job loss probability exceeds 60%. People are optimistic about themselves — and pessimistic about everyone else. The same pattern the Pew data showed: trust AI for yourself, fear it for society.
How People Actually Use Claude
What Claude Produces (% of conversations)
93% of conversations produce identifiable artifacts. 3,000+ unique work tasks identified.
The wage-token correlation: Marketing managers ($80/hr) generate ~2.5x more tokens than editors ($37/hr). Higher-wage occupations produce 1.34x more output per turn, use 1.53x more turns, and 34% use extended thinking. The more you earn, the more you use AI — and the more autonomously you use it.
The key insight: Anthropic’s data shows augmentation (user learns, iterates, gets feedback) accounts for just over half of conversations. OpenAI’s data shows 60 hours of delegated agent work per day. Two companies. Two data sets. Two different stories about the same transition: augmentation is the present. Delegation is the future. The question is how fast the switch happens.
The Structural Read
OPENAI vs ANTHROPIC — TWO VISIONS OF WORK
OpenAI’s report: 60 hours/day of agent work, 85% through Codex, full delegation. Anthropic’s report: augmentation is 50%+ of conversations, users learn and iterate. OpenAI sees the future as delegation. Anthropic sees it as collaboration. Both are right — for different users at different stages.
THE JUNIOR EMPLOYEE PARADOX
57% feel their skills are growing more valuable. But 33% think their junior colleagues have a 60%+ chance of losing their jobs. Everyone thinks AI makes THEM better — and makes the person below them replaceable. This is the middle management compression seen from the inside.
GEOGRAPHIC UNEVENNESS IS 2-5 YEARS FROM EQUALIZING
US, India, Japan, UK, South Korea lead Claude usage. Within the US, states with more tech workers adopt faster. Anthropic estimates equalization in 2-5 years. The McKinsey $1.9T European opportunity depends on this gap closing.
The Bottom Line
Anthropic just published the most rigorous study of AI’s economic impact — 9,700 respondents linked to real usage data. The headline: 86% productivity gains, 68% learning more, 57% feeling more valuable. The subtext: everyone is optimistic about themselves and pessimistic about juniors. The data point that matters most: augmentation is still 50%+ of Claude conversations. The transition from augmentation to delegation is the transition from the present to the future — and both OpenAI and Anthropic published data this week showing where we are on that curve.