AI Coding Agents and Software Robots in Professional Software Engineering

This dataset contains 205 responses to the survey AI Coding Agents and Software Robots in Professional Software Engineering. The responses were collected between 20 July and 4 August 2026. The questionnaire comprised 40 numbered questions, including several multi-item matrix questions. It examined the use of AI-enabled tools, coding agents, and software robots in professional software development. The topics covered include professional experience, technical responsibilities, workplace characteristics, and the frequency of AI-tool use. The survey also addressed multi-agent workflows, orchestrator agents, task planning, code review, testing, and verification practices. Further questions investigated AI-service costs and availability, workforce changes, junior-developer employment, training, and hiring expectations. Data collection was conducted primarily through LinkedIn, with particular emphasis on reaching experienced software developers. Participation was voluntary, and the questionnaire was anonymous. No names, email addresses, or other direct personal identifiers were requested. The published dataset contains all 205 non-duplicate submissions retained after the removal of 25 exact duplicate records; no other responses were excluded. Within the professional networks through which the survey was distributed (primarily the authors’ LinkedIn networks) the responses provide broad coverage of the intended respondent group, particularly experienced software developers. The dataset can therefore be regarded as broadly reflective of the active professional community reached through these networks, although it is not claimed to be statistically representative of the software-development profession as a whole. An unusual feature of the dataset is that the questionnaire was completed not only by humans but also by AI systems, presumably following human-initiated requests. Two submissions explicitly identified the respondent as an AI system answering without role-playing a human. One respondent reported that an AI system had generated or substantially rewritten some of their answers, while two respondents preferred not to disclose whether they were human. These responses have been retained so that researchers can define and document their own inclusion and exclusion criteria. The processing and analysis of the questionnaire are ongoing, and the research team intends to publish findings based on the dataset in several scientific articles. The dataset is now openly available. Researchers are welcome to use and analyse it but are requested to cite the dataset and its Zenodo DOI in any resulting publication or research output. The same dataset is provided in XLSX, CSV, and ODS formats.