The Cartesian Cut in Agentic AI
2026-04-09 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors explain that Large Language Models (LLMs) learn by predicting words, which can be used to make them perform tasks when connected to systems that act on their outputs. They compare how brains control actions internally with how LLMs separate the learning part from the decision-making part, allowing for easier oversight and updates but sometimes causing delays or errors. The paper describes three ways to manage this control: limited services, separated agent systems, and fully integrated agents, each balancing independence, reliability, and supervision differently.
Large Language Modelspredictioncontrol systemsfeedback controllersCartesian agencyruntimemodularityrobustnessgovernance
Authors
Tim Sainburg, Caleb Weinreb
Abstract
LLMs gain competence by predicting words in human text, which often reflects how people perform tasks. Consequently, coupling an LLM to an engineered runtime turns prediction into control: outputs trigger interventions that enact goal-oriented behavior. We argue that a central design lever is where control resides in these systems. Brains embed prediction within layered feedback controllers calibrated by the consequences of action. By contrast, LLM agents implement Cartesian agency: a learned core coupled to an engineered runtime via a symbolic interface that externalizes control state and policies. The split enables bootstrapping, modularity, and governance, but can induce sensitivity and bottlenecks. We outline bounded services, Cartesian agents, and integrated agents as contrasting approaches to control that trade off autonomy, robustness, and oversight.