Coding spreadsheet: generative AI, productivity and judgment among newly appointed government attorneys in Brazil (anonymised qualitative data)

This dataset contains the anonymised coding spreadsheet supporting the article "From AI's jagged frontier to the risk of technical complacency: productivity and judgment among newly appointed public attorneys". The study is an exploratory in-depth qualitative case study of how generative AI affects productivity and the acquisition of legal expertise among newly appointed federal government attorneys (up to two years of service) at Brazil's Office of the Attorney General of the Union (Advocacia-Geral da União, AGU). Data come from six semi-structured interviews, a corpus of about 24,700 words in Portuguese, analysed through categorical content analysis in the tradition of Bardin. The coding combines three theoretical categories defined a priori (jagged technological frontier, technical complacency and algorithmic complacency) with fifteen emergent categories identified inductively. The file (Coding_spreadsheet_interviews_EN.xlsx) has five sheets: Read me: study description, file structure, and methodological and anonymisation notes. Participant profiles: anonymised professional profiles of the six participants (E1–E6), with region, time at the AGU, previous career, area of work, work arrangement, AI tools used, frequency of use and satisfaction rating. Codebook: 18 categories with type, origin, definition, inclusion criterion and an automatic count of coded excerpts. Coded excerpts: 82 recording units, each assigned to a category and a participant, with the question context, an analytic note, an English translation and the anonymised original excerpt in Portuguese. Category x participant matrix: cross-tabulation of coded excerpts by category and participant, calculated by formula, for examining convergence, divergence and saturation. Anonymisation. Participants are identified only by the codes used in the article (E1–E6). Names, ages, exact dates, cities, unit names, gender, job titles not reported in the article, and case-identifying details such as monetary amounts and page counts were removed or generalised. Editorial insertions appear in square brackets. The English translations were prepared by the authors. Ethics. The study was approved by the Research Ethics Committee of the Federal University of Technology – Paraná (UTFPR), Opinion No. 8.278.895 (CAAE 94905726.1.0000.0177). All participants gave informed consent. The full interview transcripts are not shared, in line with the confidentiality commitment made to participants. This dataset belongs to the first stage of an ongoing research project; a later stage will examine experienced government attorneys at the AGU.