Do agents need to sleep?
Abstract
LLM-based agents, which increasingly operate persistently across sessions and accumulate memory through interaction, already exhibit the familiar symptoms of unchecked plasticity: semantic drift, degraded retrieval precision, propagation of stored errors, and saturation of memory capacity. Drawing on the selectionist "ringing, resonance and renormalization" (RRR) account of sleep, we propose that each of its three phases maps onto an operation feasible in digital memory architectures. Engineers building stateful agents, meanwhile, are independently rediscovering the problems sleep evolved to solve: what to forget, and when to do the forgetting. From this we derive the falsifiable prediction that selectionist consolidation will outperform instructionist alternatives as memory stores grow large. The core claim concerns the nature of continuous learning itself: plasticity has a price, and for any system that learns continuously under finite capacity while coupled to its environment, that price is sleep.
Related
- The need for pediatric sleep education to enhance healthcare for children and adolescents: A global perspective from a survey of members of the international pediatric sleep association
- All you need to know about testosterone results: caution and feedback from laboratories
- What science does biodiversity action still need? Lessons from a decade of IPBES
- Biodiversity science and policy need more model intercomparisons
- Response to “radiomics of mandibular osteoradionecrosis with recurrence: The need for etiologic and temporal validation”
BibTeX
@misc{haber2026agents,
title = {{Do agents need to sleep?}},
author = {Ido Haber and Giulio Tononi},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.02216},
eprint = {2611.02216},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.02216},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Do agents need to sleep?
AU - Ido Haber
AU - Giulio Tononi
PY - 2026
DA - 2026/11/01
PB - arXiv
DO - 10.48550/arxiv.2611.02216
UR - https://arxiv.org/abs/2611.02216
AB - LLM-based agents, which increasingly operate persistently across sessions and accumulate memory through interaction, already exhibit the familiar symptoms of unchecked plasticity: semantic drift, degraded retrieval precision, propagation of stored errors, and saturation of memory capacity. Drawing on the selectionist "ringing, resonance and renormalization" (RRR) account of sleep, we propose that each of its three phases maps onto an operation feasible in digital memory architectures. Engineers building stateful agents, meanwhile, are independently rediscovering the problems sleep evolved to solve: what to forget, and when to do the forgetting. From this we derive the falsifiable prediction that selectionist consolidation will outperform instructionist alternatives as memory stores grow large. The core claim concerns the nature of continuous learning itself: plasticity has a price, and for any system that learns continuously under finite capacity while coupled to its environment, that price is sleep.
ER -