Metabolic constraints are important for brain modeling. But to be useful, they have to be integrated with an understanding of the architecture of human-level cognition. Otherwise, they could even lead modeling in the wrong direction.

Haueis and Colaço (H&C) label all models as cognitive. However, just because cognitive science is so-called doesn’t mean that all models are. When type differences are ignored, an argument runs the risk of prescribing a monist modeling strategy. In their case: mechanistic. Such a prescription is empirically inadequate and overly restrictive, as modeling in cognitive science relies on many modeling types.

Haueis and Colaço distinguish evaluative and generative roles for metabolic considerations in cognitive modeling. I extend this framework to the default mode network, arguing that this network’s elevated aerobic glycolysis constitutes a metabolic constraint on spontaneous thought. This constraint shapes the richness of spontaneous thought across varying task demands and predicts load-dependent phenomenological shifts through inter-network resource competition.

Haueis and Colaço argue that metabolic considerations should constrain and generate cognitive models. Research in cognitive fatigue provides a test case. Passive fatigue induction reveals specific representational degradation that points to local rather than global metabolic mechanisms, pruning potential models. We argue that local aerobic glycolysis offers a coherent, testable generative account of these findings.

Haueis and Colaço argue that metabolic constraints should prune and generate cognitive models. Cognitive load theory offers an applied framework for testing these constraints. Working memory load can be viewed as the behavioral expression of local ATP limits. Primary–secondary knowledge, embodied offloading, and instructional efficiency map onto bioenergetic trade-offs. Extending CLT to human–AI symbiosis further shows how shared systems can support dynamic load balancing and testable metabolic predictions.

The commentators point out elaborations and challenges to incorporating metabolic considerations into cognitive modeling. We agree with many of them and highlight that these issues can only be addressed when we actively try to include metabolism while evaluating and generating models. The main message of our response is thus that metabolic considerations are epistemically valuable, regardless of the particular modeling approach, conceptualization, or specific ontological commitments regarding cognition. Yet, the commentator’s contributions also reveal that metabolism alone is never the decisive factor in cognitive modeling, and that its value for particular projects needs to be demonstrated in the modeling process, rather than assumed from the outset.