Work

Generative AI and Early-Career Computing Pathways: A Scoping Evidence Map of Skills, Employability Perceptions and Workforce Change, with Coding by Country Income Group

Year: 2026
Type: preprint
Study design: Systematic Review
Abstract: Background. Generative AI coding tools spread rapidly after ChatGPT's release in November 2022, and there is concern that they change how early-career workers enter computing. Existing reviews address junior developers' perceptions, software practice or computing education, but, as far as could be read, none maps the evidence jointly by career stage, outcome type, evidence basis and country income group. Objective. To map the evidence on generative AI and skills, employability perceptions, labour-market change, task or role change and education response for computing students, recent graduates (up to 2 years after graduation) and early-career workers (up to 3 years in a software or IT role), and to show where that evidence comes from. Methods. A scoping evidence map guided by Arksey and O'Malley, Levac et al. and the Joanna Briggs Institute, and reported against PRISMA-ScR. On 7 November 2026 two free sources were searched: arXiv (1,293 records) and OpenAlex (13,005 records), supplemented by earlier purposive searching. After de-duplication and a keyword pre-filter, 702 records were shortlisted and 82 were assessed in detail. The protocol was not registered. Screening and charting were done in a single pass by an AI assistant under the sole author's direction and have not been independently verified. Findings were tagged by evidence basis, and countries were coded to World Bank FY27 income groups for context. Results. Twenty-seven studies were included (29 charted rows). Twenty were from 2026. Hiring evidence is unsettled: US junior software vacancies fell 14 to 15% relative to senior vacancies, while firms adopting GitHub Copilot were 3 to 5% more likely to hire software engineers each month, driven by entry-level hires. Student studies reported job concern alongside a belief that AI skills are necessary, including surveys from Egypt, India and Brazil. Nine studies used high-income data only, five included verified lower- or upper-middle-income countries, and none reported results by income group. No verified Pakistan-specific study met the criteria. Conclusions. The evidence base is small, recent, mostly preprints and concentrated in high-income countries. A fuller, registered, multi-database review with dual screening is needed before conclusions are drawn for lower-income settings, including Pakistan.
Author: Naeem, Dua
Language: English
Open Access status: green
OpenAlex ID: W7235493518
Cites:
Cited by:
Related to: 10