🧭 Economic Index: High-Tenure Users Are 10% More Effective — And the Gap Is Growing
Anthropic published the fifth instalment of its Economic Index on March 24, covering usage data from November 2025 through February 2026. The headline finding: users who have been with Claude the longest are measurably better at using it. High-tenure users show roughly 10% higher conversational effectiveness compared with new users — they write shorter prompts, get better first-pass responses, and tackle longer, more complex tasks. The report calls these “learning curves,” borrowing the economics term for how skill improves with accumulated experience.
What the data shows
- Coding dominates: 35% of all Claude conversations involve coding — unchanged from prior reports, but growing in absolute volume as the user base expands
- Diversification signal: The top 10 task types fell from 24% to 19% of total usage, meaning Claude is being applied to a broader range of work than a year ago
- Task complexity drift: Long-tenure users increasingly bring Claude complex, multi-step work tasks; new users skew toward lower-complexity, personal productivity queries
- Wage exposure: Higher-wage occupations account for a disproportionate share of complex tasks, while lower-wage groups are more likely to use Claude for single-turn lookups
What this means for teams onboarding colleagues to Claude
The 10% effectiveness gap between new and experienced users is not magic — it’s learned behaviour. Teams that invest in structured onboarding (example prompts, shared CLAUDE.md files, prompt libraries) compress that learning curve from months to weeks. If your organisation’s AI adoption is stuck at low-value tasks, it is almost certainly an onboarding problem, not a model capability problem.
Economic Index
user research
onboarding
productivity
learning curves
🧭 Anthropic Publishes First Methodology for Measuring AI’s Real-World Labor Market Effects
Alongside the Economic Index, Anthropic researchers Maxim Massenkoff and Peter McCrory published a formal academic paper introducing a new way to measure AI’s actual impact on the labour market — not hypothetical exposure scores derived from job descriptions, but real-world effects observable in employment data. The methodology cross-references actual Claude usage logs with US Current Population Survey employment data, allowing the team to test which occupations are seeing measurable hiring changes.
Early evidence — significant but narrow
- No broad unemployment signal: Across most occupations, there is no statistically significant drop in employment attributable to AI adoption so far
- Early-career hiring falls: Workers aged 22–25 in AI-exposed roles (finance, law, management, computer science) saw a 6–16% decline in new hire rates — a pattern consistent with AI absorbing entry-level task load rather than eliminating jobs outright
- Theoretical vs. actual gap: AI is theoretically capable of performing the majority of tasks in several high-exposure occupations; actual adoption is still a fraction of that potential, suggesting a large runway before structural employment effects become widespread
- Junior developer signal: The paper notes the tech sector shows the clearest early hiring slow-down, which aligns with the widely reported softening in junior software engineering roles since late 2025
For hiring managers in AI-exposed organisations
The 6–16% drop in early-career hiring is not a reason to stop hiring juniors — it is a signal that the nature of entry-level work is changing faster than job titles reflect. Teams that redefine junior roles around AI-assisted output (reviewing, testing, and directing generated code rather than writing all of it from scratch) will find early-career hires remain highly valuable. Teams that don’t redefine roles risk the worst outcome: paying for headcount that is under-utilised because the workflow wasn’t redesigned.
labor market
AI impact
employment
research
Economic Index
🧭 Claude Paid Subscriptions More Than Doubled in 2026 — Downloads Outpace ChatGPT
TechCrunch reported on March 28 that Anthropic confirmed Claude’s paid subscription base “more than doubled” in 2026, with external estimates placing the total paid consumer user base somewhere between 18 million and 30 million. That figure would represent a remarkable acceleration: Anthropic’s total consumer user base sat at roughly 10 million at the start of Q1 2026. The growth follows the viral success of Claude Code, a Super Bowl advertising campaign, and Anthropic’s widely reported decision not to allow Claude for mass domestic surveillance or fully autonomous weapons.
The download data is striking
- Claude mobile downloads (early March 2026): ~149,000 per day in the United States
- ChatGPT mobile downloads (same period): ~124,000 per day — Claude is now ahead
- Super Bowl effect: Anthropic ran its first major consumer advertising campaign during Super Bowl LX, targeting general consumers rather than just developers
- Subscription cohort mix: Growth is concentrated in the Pro tier ($20/month) and the newer Max tier; the free tier still represents the majority of total users but a minority of monthly active usage
What the download data means for API developers
Higher consumer awareness translates directly into easier enterprise sales — your Claude-powered product is pitching into an audience that already has a mental model for what Claude can do. If you’ve been hesitant to call out Claude by name in your product (rather than hiding it as “AI-powered by™”), the brand recognition data now argues for disclosure over obscurity.
subscriptions
growth
consumer
mobile
market share
