Anthropic interviews 1,250 professionals about AI

Anthropic used AI to interview 1,250 professionals about how they work with artificial intelligence. The results show a relationship that is less straightforward than headlines often suggest: most people value the time it saves, but they also fear losing autonomy, recognition or even their profession.

The tool is called Anthropic Interviewer and runs on Claude. It can plan interviews, ask follow-up questions in real time and analyze hundreds of conversations afterward. This first study included 1,000 workers from different fields, 125 creatives and 125 scientists.

AI saves time, but the concern remains

In the general group, 86% said AI saves them time and 65% were satisfied with their role at work. However, 55% expressed anxiety about its future impact and 69% mentioned the stigma that still surrounds its use in the workplace.

The tension appears in a recurring idea: people want to delegate routine work while keeping the tasks they feel define their profession. For many, the future is not about AI doing everything, but about supervising systems, reviewing results and handling the more human parts of the job.

48% imagined a transition toward roles focused on managing or controlling AI systems. In practice, that could mean a teacher automating some administrative work to spend more time with students, or a communications professional reviewing and guiding work generated by a model instead of producing every piece from scratch.

People's perception of collaboration also has some nuance. 65% described their main use as a form of support, while 35% called it automation. Anthropic notes that this view does not exactly match another analysis based on real Claude conversations, where automated and collaborative tasks appeared more evenly balanced. The difference may be because people adapt results after the chat or describe their use differently from how it happens in practice.

Creatives gain productivity and lose confidence

Among creative professionals, 97% said AI saves them time and 68% said it improves the quality of their work. Some use it to research, generate ideas, edit material or speed up repetitive tasks. One photographer, for example, said editing a project went from about 12 weeks to around 3.

But the benefit comes with serious concern. 70% mentioned negative judgment from colleagues or clients for using AI. Participants also worried about replacement, markets being flooded with cheap content and losing control over creative decisions.

The 125 creative participants said they wanted to retain control of their work. Even so, several acknowledged that AI already influences decisions they used to make themselves. That contradiction sums up their situation: the tools make it more viable to produce, but they can also change what it means to be an author and how much economic value that work retains.

Scientists still do not trust it for research

Scientists were more cautious. They mainly use AI to review literature, write manuscripts, program and correct analyses. They still do not usually trust it with the central parts of their work, such as formulating hypotheses, designing experiments or interpreting decisive results.

Reliability was an obstacle in 79% of the interviews. The problem is not only that a model can make mistakes, but that checking every answer can take as much time as doing the task without assistance. Data confidentiality, inconsistent answers and the difficulty of using AI in environments with security requirements are also concerns.

Despite these reservations, 91% of scientists wanted more AI assistance. Many imagine a system capable of finding relationships that are not obvious, proposing hypotheses and connecting large databases. They want a research partner, but current tools still do not reliably fill that role.

Automated interviews with human oversight

Anthropic Interviewer works in three stages:

• Planning: prepares a common script for all interviews while leaving room to deviate based on each response.
• Interview: talks with each participant for about 10 to 15 minutes and adapts the questions in real time.
• Analysis: groups topics, answers the research questions and extracts examples for human researchers to review.

The study does not automatically turn the answers into a general truth about the labor market. Participants were recruited through collaborative work platforms and came mainly from Western contexts. They also knew they were speaking with an AI about AI, which may have influenced their answers.

Anthropic also acknowledges that people may describe their use differently from how they actually work. The study offers a snapshot of the present, not a prediction of how these relationships will develop over time.

The next step is already underway: Anthropic is inviting some Claude users to take part in public interviews lasting 10 to 15 minutes about the role they want AI to have in their lives. The responses will be analyzed anonymously and could influence future products, policies and models.

The study's most important signal is not that AI will replace all professionals or improve every job. It is that adoption is advancing while each group negotiates its own boundaries: which tasks to delegate, which decisions to keep and how much error it is willing to accept. That discussion, more than automation alone, will define AI's place in everyday work.