Graduated Autonomy for Artificial Intelligence Agents in Human Capital Management: A Systematic Review and a Consequence-Calibrated Governance Framework
Year: 2026
Type: review
Study design: Systematic Review
Abstract: Human capital management platforms are starting to embed software agents built on large language models. Unlike earlier analytics tools, these agents can plan a sequence of steps, call system functions and change employee records. Artificial intelligence is therefore moving from advising managers to acting for them. Most governance guidance, however, was written for systems that recommend rather than act. This study asks how organisations should match the autonomy of such agents to the consequences of the people decisions they touch. A systematic review was conducted and reported using the PRISMA 2020 guideline. Eighteen structured searches of a multi-source scholarly index returned 180 records. After duplicate removal and eligibility screening, 55 peer-reviewed sources were retained, and four foundational and regulatory sources were added. Thematic synthesis produced seven findings. Algorithmic control of work is already extensive. Bias in hiring tools is persistent and partly organisational. Human review is weaker than assumed, because reviewers often defer to machine advice. Responsibility tends to diffuse as systems gain autonomy. Employee trust depends on fair process and visible human involvement. Language-model agents add new failure modes, such as fabricated content and hidden instructions planted in documents. Finally, current governance frameworks rarely specify controls for individual delegated actions. Building on these findings, the study proposes a consequence-calibrated governance framework. It sets an autonomy ceiling for each class of people decision and links every ceiling to five control layers. Six testable propositions and practical guidance for human resource and technology leaders are offered. Keywords: large language models; people analytics; algorithmic management; automation bias; meaningful human control
Author: Shivanand R Koppalkar
Institution: Walsh College
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W7223211518