Haueis and Colaço argue that metabolic considerations should function both as evaluative constraints and as generative tools for cognitive modeling. I propose that their framework offers support for something more specific: a positive theory of metabolic rationality where cognitive strategy selection is itself a metabolic resource allocation problem. On this account, the brain dynamically allocates metabolic resources across a temporal horizon of anticipated cognitive demands, treating every act of deliberation as carrying a metabolic opportunity cost, i.e., the reduced capacity it creates for future cognition.

The empirical basis for this proposal is now substantial. Prefrontal and associative networks supporting deliberative reasoning have up to 67% higher energetic costs of signaling than sensorimotor regions (Castrillón et al., Reference Castrillón, Epp, Bose, Fraticelli, Hechler, Belenya, Lio, Sorg, Straulino, Yakushev and Riedl2023), and brain wiring reflects optimal tradeoffs between connectivity costs and adaptive advantage (Bullmore & Sporns, Reference Bullmore and Sporns2012). The critical insight goes beyond the well-known cost of deliberation: the brain’s energy budget is finite and temporally extended. An organism facing a continuous stream of decisions with unpredictable timing and importance cannot evaluate each in isolation. It must manage its metabolic reserves across the entire sequence. A physician in an emergency department, a forager in an uncertain landscape, and a parent navigating competing demands – all face continuous streams of decisions where exhausting metabolic resources on any single choice risks catastrophic failure on subsequent, potentially more critical ones. This transforms cognitive strategy selection from a local accuracy problem into a global resource allocation problem.

Allostatic regulation provides the physiological mechanism for this management. Sterling (Reference Sterling2012) argues that the brain proactively anticipates future demands and adjusts resource allocation accordingly, going beyond reaction to current needs. Applied to cognition, this implies that the brain continuously forecasts the metabolic demands of upcoming cognitive tasks and calibrates the precision of current processing accordingly. The neuromodulatory architecture supporting this calibration is revealing: the brain supplements precise synaptic computation with diffuse neuromodulatory systems that modulate the gain of neural populations at low metabolic cost (Attwell & Laughlin, Reference Attwell and Laughlin2001; see also Shine et al., Reference Shine, Müller, Bose, Breakspear and Moran2021). These systems set the level of computational investment that a given situation warrants, functioning as a metabolic resource allocator.

Predictive coding frameworks offer a natural computational substrate for this theory. Under predictive processing, the brain minimizes free energy through prediction error coding, a scheme that transmits only deviations from expectations, reserving metabolic expenditure for surprising information (Clark, Reference Clark2013; Friston, Reference Friston2010). This architecture is inherently metabolically conservative: stable environments generate small prediction errors, licensing low-precision processing and minimal metabolic expenditure, while novel or volatile environments generate large errors, triggering metabolically expensive precision-weighted updating. Crucially, predictive processing is hierarchically organized. Lower levels of the cortical hierarchy handle routine predictions at low metabolic cost; only when prediction errors cannot be resolved locally do they propagate upward to higher associative and prefrontal areas, that is, those regions that Castrillón et al. (Reference Castrillón, Epp, Bose, Fraticelli, Hechler, Belenya, Lio, Sorg, Straulino, Yakushev and Riedl2023) show to be most metabolically expensive. This means that the brain has an endogenous escalation mechanism: deliberative processing functions as a costly upper tier that the system recruits only when cheaper levels of the hierarchy fail to absorb prediction error. The apparent binary switch between System 1 and System 2 of dual-process models thus dissolves into a continuous metabolic gradient along the cortical hierarchy.

This hierarchical precision-weighting mechanism reframes several well-documented decision phenomena. Anchoring, for instance, becomes the metabolically rational reliance on existing priors when prediction errors are insufficiently large to justify the cost of updating higher-level models. Status quo preference reflects the brain’s default maintenance of established generative models whose predictions remain adequate, as updating is metabolically justified only when errors accumulate beyond a threshold set by the allostatic assessment of available resources. Loss aversion, in turn, can be understood through asymmetric precision weighting: negative prediction errors (worse-than-expected outcomes) receive higher precision than positive ones because they signal potential threats to metabolic homeostasis, warranting greater computational investment in updating (Palumbo et al., Reference Palumbo, Bortolotti and Sacco2006). All three phenomena are manifestations of a single underlying principle: the brain’s hierarchical allocation of metabolic resources tracks the expected informational value of further processing relative to its energetic cost.

This framework generates predictions that are distinct from both standard behavioral economics and ecological rationality accounts. First, cognitive strategy should vary with the organism’s current and anticipated metabolic state and with task structure (as ecological rationality predicts). Decision-makers should become more heuristic when subsequent demands are expected to be high, independently of task simplicity, a prediction about sequential task interaction that neither framework makes. Second, what looks like inconsistent risk attitudes across contexts may reflect metabolic state-dependence: risk-sensitive behavior shifts with the brain’s allostatic assessment of available resources relative to anticipated demands, so that risk preferences function as adaptive signals of the organism’s current energetic state (Palumbo et al., Reference Palumbo, Bortolotti and Sacco2006). Third, the framework predicts that institutional and cultural environments that reduce cognitive unpredictability should lower the metabolic cost of coordination, enabling more efficient collective decision-making, a social-level prediction that connects metabolic neuroscience to institutional design (North, Reference North1990; Veissière et al., Reference Veissière, Constant, Ramstead, Friston and Kirmayer2020).

Haueis and Colaço have provided cognitive science with further arguments to take metabolic constraints seriously. The natural next step is to extend such arguments to decision theory. The metabolic rationality framework sketched here suggests that human cognition is metabolically rational: adapted to manage a scarce energetic resource across the uncertain demands of an extended temporal horizon. This is a story about a different kind of intelligence: one that optimizes for sustained adaptive performance across a lifetime of decisions. Once we recognize this, the task for both science and policy shifts from correcting a supposedly flawed reasoner to understanding, and designing for, an exquisitely efficient one.