Processing, memory, and contextualisation as an analytical framework

Public and institutional discussion of artificial intelligence (AI) often turns on simplified oppositions: use versus non-use, benefit versus harm, replacement versus augmentation. This report argues that such framing is increasingly inadequate for understanding large language model (LLM) systems in organisational settings. The key question is not only whether AI is used, but how cognitive work is distributed between human and model across tasks and workflows. To address this problem, the report proposes a simple analytical framework organised around three functions: processing, memory, and contextualisation. It argues that human–LLM collaboration is best understood not as a human using a tool, but as a form of redistributed cognition in which these functions are arranged differently across the human, the model, and the surrounding infrastructure. On this basis, the report makes three main claims. First, memory architecture is a decisive mediating variable in the practical usefulness of human–LLM systems. Second, different organisations are unlikely to converge on a single model of AI use, because appropriate configurations depend structurally on task type and on how contextualisation is distributed across the organisation. Third, the redistribution of processing does not erase human thinking work, but repositions it upstream and downstream of model use, in framing, allocation, and evaluation. The report's contribution is therefore conceptual and analytical rather than prescriptive: to provide a clearer vocabulary for analysing how different forms of human–LLM collaboration fit different organisational problems.

VOLLBRACHT Ian; 

2026-07-16

Publications Office of the European Union

JRC146572

978-92-68-40908-4 (online),   

1831-9424 (online),   

EUR 40761,    OP KJ-01-26-267-EN-N (online),   

https://publications.jrc.ec.europa.eu/repository/handle/JRC146572,   

10.2760/1580746 (online),