About half of Claude users say AI can already handle 50 percent or more of their work tasks, according to a new Anthropic survey of roughly 9,700 people who use its Chat, Cowork, and Code products. The findings, published on June 27, 2026, offer one of the larger self-reported snapshots of how heavy AI users see automation reshaping their day-to-day jobs — and the picture is more nuanced than the headline number suggests.
What the Anthropic survey found about AI at work
Beyond the headline figure, responses spread across a wide range of automation levels. Crucially, the survey asked about concrete tasks — such as writing a piece of text or drafting a query — rather than entire job roles. That framing matters, because a task is a much smaller unit than a profession.
- Around 33 percent see AI as usable for 30 to 60 percent of their tasks.
- About 14 percent put that figure at 60 to 90 percent.
- Roughly 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.
Anthropic stresses that work is usually more than the sum of individual tasks. The knowledge transfer between steps — deciding what to do next, carrying context forward, judging when an output is good enough — is often the hardest part to automate. Respondents were judging discrete pieces of their workload, not whether a model could run an entire role end to end.
Where Claude is used most
The most common work uses tie to the Artifacts feature, where the output is a concrete deliverable rather than just a chat reply. The heaviest categories were narrow and production-oriented:
- Database queries — 82 percent
- Blog or article writing — 81 percent
- Marketing content — 80 percent
Anthropic also split uses into work-driven and personal categories. The most work-skewed tasks were academic papers, presentations, and database queries, while recipes, creative writing, and games leaned personal. For more on how these tools are evolving, see our AI news hub.
Collaboration over replacement
Expectations about the pace of progress were “strikingly consistent” across groups, Anthropic says, regardless of experience, location, or profession — a pattern the company describes as a “rising tide” of broadly improving capability. In other words, the optimism is not confined to a small circle of power users.
Attitudes diverge most by career stage. Early-career workers report the highest share of AI-capable tasks and worry the most about their jobs, while the heaviest Claude users are the most optimistic, believing their own skills are becoming more valuable rather than obsolete. How someone reads their prospects depends heavily on both AI experience and how far along they already are professionally.
Most respondents say they hope to work alongside AI rather than be replaced by it, wanting the tools to absorb boring, routine work and the resulting gains to be shared widely. That split between opportunity and anxiety mirrors a wider debate across the tech business world, where the same capabilities read as a promotion to some workers and a pink slip to others.
Why this matters
Self-reported surveys come with obvious caveats: people who pay to use Claude daily are not a representative sample of the workforce, and perception is not productivity. But the data is still a useful early signal of how the most engaged users are recalibrating expectations. If even a quarter of them expect AI to cover most of their work inside a year, employers, training programs, and workers themselves face pressure to redefine what a “job” consists of — emphasizing judgment, oversight, and the connective tissue between automated tasks.
Frequently asked questions
Does the survey mean half of jobs will be automated?
No. Respondents rated individual tasks, not whole roles. Anthropic explicitly notes that coordinating tasks and transferring knowledge between steps is the hardest part to automate, so task-level capability does not translate directly into job replacement.
Who was surveyed?
Roughly 9,700 users of Anthropic’s Chat, Cowork, and Code products — a group of relatively heavy, self-selected AI users, which skews the results toward optimism compared with the general workforce.
