Purpose Organisations are increasingly delegating consequential decisions to autonomous software. This paper aims to develop agentic performance management (A-PM), a framework that reconceives performance management when agentic AI (A-AI) acts as a delegated organisational actor rather than a passive analytic tool. Design/methodology/approach Agency theory, intelligent sociotechnical systems (iSTS) and technology acceptance perspectives are integrated to explain how agentic capabilities (goal-driven autonomy, adaptive planning and hyper-contextual reasoning) reshape performance management processes and governance demands. From this synthesis, nine testable propositions are derived to guide empirical research. Findings Agentic deployment produces a distinct governance problem, termed specification risk, whereby agents faithfully optimise formalised objectives that may diverge from tacit organisational values. To contain this risk, the framework prescribes two constitutive governance pillars – procedural transparency (explainability, auditability, employee voice) and sociotechnical envelopment (culture, managerial capability, oversight structures). When matched with social readiness, A-PM improves alignment, agility and wellbeing; when ungoverned, it amplifies harm. Practical implications Managers and HR teams must treat A-AI rollout as organisational redesign: pilot within tight governance envelopes, build managerial AI literacy and embed contestability and audit trails before scaling. Originality/value This paper extends agency theory to non-human agents, introduces specification risk for performance management and offers an integrative, testable research agenda for scholars and practitioners concerned with algorithmic governance and organisational design.
Protik Basu (Mon,) studied this question.