Anthropic Claude survey: half of users say AI handles 50%
Survey of roughly 9,700 Claude users finds about half say AI can already do half or more of their tasks.
TL;DR
- 01Survey of roughly 9,700 Claude users finds about half say AI can already do half or more of their tasks.
- 02Anthropic's survey of roughly 9,700 Claude users finds about half say AI can already handle half or more of their work.
- 03The responses cover Claude Chat, Cowork and Code, and users reported both current task coverage and expectations for the next 12 months.
Anthropic's survey of roughly 9,700 Claude users finds about half say AI can already handle half or more of their work. The responses cover Claude Chat, Cowork and Code, and users reported both current task coverage and expectations for the next 12 months.
How much work do Claude users say AI can handle today and in a year?
About half of respondents place AI's current usefulness at the level of handling roughly half their work, with specific breakdowns: 33 percent of users say AI is usable for 30 to 60 percent of their tasks, 14 percent say 60 to 90 percent, and around 4 percent believe Claude could already do their entire job. Looking 12 months ahead, about 26 percent expect AI to take over most of their work.
Those numbers come directly from Anthropic's survey of roughly 9,700 users and distinguish between present-day assessments and a one-year projection. The study also notes that many users judge AI capability by concrete tasks, like writing a text, and cautions that whole jobs often involve stitching tasks together.
What do people actually use Claude for at work?
Anthropic's data shows the most work-related uses are concrete deliverables through Claude's Artifacts feature: marketing content is used for work by 80 percent of respondents, blog or article writing by 81 percent, and database queries by 82 percent. Artifacts are outputs such as documents or interactive graphics rather than chat replies.
The survey also breaks down Chat and Cowork uses: academic papers, presentations and database queries skew almost entirely to work or coursework, while recipes, creative writing and games skew personal. Marketing content, documents and blog posts rank among the most work-driven categories across respondents.
How do users feel about AI affecting careers?
Responses vary by experience and intensity of use: early-career workers see the highest share of AI-capable tasks and express the most worry about job risk, while the heaviest Claude users report greater optimism and believe their skills are becoming more valuable. Anthropic says expectations about the pace of progress are "strikingly consistent" across experience, location and profession.
Most respondents hope to work alongside AI, not be replaced by it. They want AI to take on routine or boring tasks and expect the gains to be shared widely.
Why it matters
If roughly half of users already see AI handling a large fraction of routine tasks, job definitions and daily workflows could shift quickly toward AI-plus-human collaboration. Heavy users reporting increased optimism suggests tools that integrate deeply into work may boost perceived productivity and skill value, while early-career anxiety signals uneven effects across the workforce.
The survey's split between task-level capability and whole-job coverage matters: even high percentages for specific tasks do not automatically translate into fully automated roles, because knowledge transfer across tasks remains a human coordination problem.
What to watch
Monitor the one-year expectations metric: Anthropic found about 26 percent expect AI to take over most of their work in 12 months, so any follow-up showing that share rising or falling will indicate whether optimism is matched by adoption. Also watch usage of Artifacts outputs versus chat replies, since the survey ties high work-use percentages to Artifacts' deliverables.
Anthropic's survey gives an early snapshot of how thousands of Claude users judge current capability and near-term change, with clear hotspots in marketing, blogging and database queries and mixed career sentiment depending on experience and intensity of use.
Written by The Brieftide · Source: The Decoder
The Brieftide Daily · 06:00
Briefs like this one, in your inbox every morning.
Continue reading
More in Enterprise AI AdoptionNVIDIA Confidential Computing: 98% performance, Blackwell GPUs
NVIDIA’s Confidential Computing secures models and data on Blackwell (HGX B300) while adding typically under 8% throughput or per‑token.
Teleperformance AI: Achieving Operational Excellence Now
Teleperformance says firms with Lean Six Sigma or BPM discipline can better translate AI investments; a sponsored report cites $113B market.
Microsoft Frontier Company launches with $2.5B investment
The unit will deploy 6,000 industry and engineering experts to deliver enterprise AI projects using Microsoft’s existing tools.
Multi-Agent Orchestration for Enterprise AI: arXiv Paper
An arXiv paper (18 Jun 2026) evaluates DAG Plan and Execute versus ReAct across 208 enterprise scenarios and adds a Task Manager that cuts.