Abstract:

In organisations, the use of AI is shifting from systems that support human decision making to systems that make decisions. Agentic AI is a type of AI that acts independently to achieve goals, requires planning to perform multi-step tasks, calls tools/other applications, and carries out actions with minimal human interaction. Information Systems (IS) scholarship has begun to theorise this shift as a change in the locus of agency: rather than humans using IS artefacts, work is increasingly delegated to agentic IS artefacts, which in turn may delegate back to humans (Baird & Maruping, 2021). Rather than just asking, “Should organizations use AI tools?” the real question now is: “How, when, and how much control?”

Initial research indicates that this shift is challenging, even for well-resourced companies. Gartner (2025) forecasts that by the end of 2027 over 40% of agentic AI projects will be cancelled, driven by rising costs, uncertainty about business value or insufficient risk management measures, and suggests that there will be a vast amount of "agent washing" – this is when traditional chatbots and automation tools will be rebranded as agents. Meanwhile, the impact of AI on the labor market is unevenly spread globally. Cazzaniga et al. (2024) estimate that on average about 40 per cent of global jobs are susceptible to AI integration, rising to about 60 per cent in advanced economies, compared with 40 per cent in emerging markets and 26 per cent in low-income countries. Although these estimates concern generative AI and employment rather than agentic AI adoption, they suggest that the opportunities and risks of AI autonomy are likely to differ systematically between developed and developing economies.