Haueis and Colaço provide a timely, original, and highly stimulating contribution by arguing that metabolic considerations should be incorporated into models of cognition. Evaluative metabolic considerations test whether a cognitive model is energetically plausible, while generative metabolic considerations use energy constraints to guide the construction of new models. For cognitive science, metabolic considerations provide an objective standard for evaluating models of mental effort. Rather than treating cognition as purely computational, this perspective emphasizes that cognitive processes are fundamentally energy-constrained and must be explained in relation to their biological costs. In an era increasingly shaped by comparisons between biological brains and artificial intelligence systems, their emphasis on energetic efficiency also highlights an often neglected but fundamental feature of cognition. Indeed, metabolism introduces energy efficiency as a central metric alongside performance. It complicates claims of parity between artificial and biological systems by highlighting their radically different energy regimes, and it encourages the development of more sustainable, biologically informed, humanized approaches to artificial intelligence design.

We strongly support the central thesis of the article. Nevertheless, several conceptual and metabolic issues may enrich the discussion further.

The argument depends critically on how cognition is defined. If cognition is conceived narrowly as symbolic or computational information processing, metabolism may appear as a supporting constraint. However, if cognition is understood more broadly as embodied, adaptive regulation of organism–environment interactions, metabolism may be constitutive of cognition itself rather than merely enabling it. Some processes that look purely metabolic – like blood flow – may actually contribute to information processing. Contemporary philosophy of biology and embodied cognition increasingly supports this broader view, emphasizing that informational and metabolic processes are deeply intertwined in living systems (Theparambil et al., Reference Theparambil, Kopach, Braga, Nizari, Hosford, Sagi-Kiss, Hadjihambi, Konstantinou, Esteras, Gutierrez Del Arroyo, Ackland, Teschemacher, Dale, Eckle, Andrikopoulos, Rusakov, Kasparov and Gourine2024).

If metabolism is best understood as a resource rather than a mechanistic component, then metabolic considerations for cognitive modeling should not be restricted to energy metabolism. Likewise, we proposed that cognitive fatigue is not due to energy depletion but to the accumulation of neurochemical byproducts, notably glutamate, during sustained cognitive control (Wiehler et al., Reference Wiehler, Branzoli, Adanyeguh, Mochel and Pessiglione2022). Using magnetic resonance spectroscopy, we showed that demanding tasks increase glutamate levels in the lateral prefrontal cortex, alongside behavioral signs of fatigue. Instead, metabolism is often reduced to neuronal energy production – primarily oxidative phosphorylation and glycolysis – and to the energetic costs of information.

While historically understandable, current neuroscience increasingly recognizes that glial cells – particularly astrocytes, but also oligodendrocytes and microglia – are active participants in cognition. Astrocytes regulate neurotransmitter recycling, lactate shuttling, ion buffering, blood-flow coupling, synaptic plasticity, and network synchronization. Several studies show that astrocytes actively contribute to fear memory retrieval, extinction, and memory engram modulation, challenging purely neurocentric models of cognition (Bukalo et al., Reference Bukalo, O’Sullivan, Tanisumi, Mendez, Weinholtz, Zimmerman, Offenberg, Carpenter, Bhagwat, Mosley, O’Malley, Lyons, Fang, Goldschlager, Ostroff, Penzo, Wake, Halladay and Holmes2026), as well as key regulators of signaling and behavior (Won, Patel, & Fields, Reference Won, Patel and Fields2024).

Cerebral metabolism is not a static variable but an extraordinarily dynamic and context-dependent process. It changes substantially across the lifespan: the infant brain displays metabolic patterns radically different from the adult or aging brain, with developmental windows of intense glucose use, synaptogenesis, and plasticity (Rava et al., Reference Rava, Cannone, Francolini and Taverna2026). Metabolism also differs across brain regions according to functional specialization, connectivity, and cellular composition. Furthermore, it varies by cell type: fast-spiking inhibitory interneurons have distinctive energetic requirements compared with pyramidal neurons (Szegedi et al., Reference Szegedi, Douida, Hutóczki, Novák and Lamsa2026). Even within a single cell, metabolism is compartmentalized, with relevant distinctions between soma, dendrites, axons, synaptic terminals, and mitochondria-rich subdomains. Single-cell transcriptomics further demonstrate the remarkable spatial and temporal heterogeneity of astrocytic metabolic states across brain regions (Serrano-Pozo, Reference Serrano-Pozo, Li, Li, Muñoz-Castro, Jaisa-Aad, Healey, Welikovitch, Jayakumar, Bryant, Noori, Connors, Hu, Zhao, Liao, Lin, Pastika, Tamm, Abdourahman, Kwon, Bennett, Woodbury, Wachter, Talanian, Biber, Karran, Hyman and Das2024). These layers of heterogeneity suggest that invoking “brain metabolism” as a unitary explanatory variable may be insufficient.

The most fruitful integration of metabolism into cognitive science may come not only from energy budgets but also from a broader biochemical network perspective (Marano, Mariano, & Novarino, Reference Marano, Mariano and Novarino2025), including brain–body interactions. Indeed, interoception suggests metabolism is not merely an external constraint but constitutive of cognition (Feldman, Bliss-Moreau, & Lindquist, Reference Feldman, Bliss-Moreau and Lindquist2024), because metabolic states are internally represented and used to drive processing. This means metabolism does not just limit computation – it becomes part of the causal and informational structure of cognitive mechanisms themselves.

Finally, metabolic considerations of cognition should be approached cautiously. Neurons show remarkable resilience and can preserve viability and partial function under low-energy or fluctuating metabolic states through homeostatic plasticity, adaptive prioritization, and network redundancy (Sobieski, Fitzpatrick, & Mennerick, Reference Sobieski, Fitzpatrick and Mennerick2018). This resilience suggests that metabolism may be one major dimension of cognition, but not necessarily a sufficient explanatory principle. Cognitive phenomena likely emerge from interactions among metabolism, cellular physical properties, circuit dynamics, developmental programs, and environmental inputs.

These considerations raise an important methodological question: how should metabolism be operationalized in future cognitive models? Should one model glucose uptake in prefrontal cortex during executive tasks, astrocytic lactate support during memory formation, mitochondrial constraints in hippocampal plasticity, or age-dependent shifts in cortical energetic allocation? The answer is likely pluralistic. Different cognitive domains may depend on different metabolic signatures, and no single scalar measure may suffice. Addressing these questions will be instrumental to apply the guidelines proposed by Haueis and Colaço on both evaluative and generative metabolic considerations.

For this reason, the future envisioned by Haueis and Colaço may require tools capable of integrating multiscale complexity. Artificial intelligence, machine learning, and systems biology approaches may be particularly useful for combining metabolomics, magnetic resonance methodologies, electrophysiology, developmental data, and behavioral phenotypes into unified predictive frameworks. Such approaches could identify hidden metabolic states linked to attention, learning, fatigue, resilience, or disease vulnerability. This would move the field from broad energetic function toward mechanistically testable, multilevel models.

In summary, this important article opens an interdisciplinary agenda that deserves broad attention. It convincingly argues that cognition cannot be fully understood while neglecting metabolism. Metabolic constraints enable researchers to rule out energetically infeasible models while also guiding the construction of biologically plausible ones. The next step may be to embrace a richer view of metabolism itself: distributed across neurons and glia, dynamic across time, heterogeneous across regions and compartments, and extending far beyond ATP production alone. Integrating this broader metabolic biology into cognitive theory models may substantially reshape our understanding of how minds emerge from living brains.