Putting the processing, memory and contextualisation framework into practice

Large language models have moved rapidly into professional knowledge work, where their most immediate effect is an increase in processing speed. The dominant discourse around their adoption remains largely organised around this fact: faster drafting, synthesis, translation and analysis. This report argues that speed is an incomplete guide to effective LLM use. Building on the analytical framework developed in the companion report Human–LLM Thinking in Organisations: Processing, Memory, and Contextualisation as an Analytical Framework, it proposes two linked practical ideas. The first is the slow–fast–slow pattern: a model for human–LLM task structure in which users move deliberately at the beginning of a task to frame the work and supply relevant background, use the model’s processing speed in the middle, and slow down at the end to evaluate outputs, correct errors and update working memory. The second is minimum viable memory architecture: a lightweight set of artefacts — project brief, state-of-play note, decision log, terminology note, review checklist and failure note — that can be maintained in plain text without specialist infrastructure, and that give LLM-supported work continuity over time. The report is written primarily for non-specialist institutional knowledge workers using LLMs in constrained systems: environments without persistent session memory, deep file integration or unrestricted connectivity. It argues that such constraints make good practice more important, and not less possible. The central claim is that effective LLM use is not a matter of maximising speed, but of redistributing it: directing fast processing where it belongs, while preserving deliberate human judgement for the phases of the task cycle where it matters most.

VOLLBRACHT Ian; 

2026-07-16

Publications Office of the European Union

JRC146924

978-92-68-40907-7 (online),   

1831-9424 (online),   

EUR 40760,    OP KJ-01-26-266-EN-N (online),   

https://publications.jrc.ec.europa.eu/repository/handle/JRC146924,   

10.2760/6463566 (online),