Imagine your kid refuses to do his homework. “I can’t, Mom, I’m too exhausted.” You just read Haueis and Colaço’s thought-provoking piece on metabolic considerations; you know your kid had a long day at school, and so you are sympathetic to their plea: maybe this one time, skipping homework is okay because it’s genuinely impossible for your child to muster the resources for homework. Or so you think, until your partner walks into the scene and aptly motivates your son by promising them a chocolate bar. Your child, feeling a surge of motivation, finishes his homework.

This little anecdote illustrates a key problem in the literature on mental effort. It is incredibly hard to distinguish between metabolic depletion and motivational degradation: absent strong motivational variation, both predict decreases in task performance. Subjects in research settings typically have milquetoast motivation: most researchers can neither offer them large rewards nor threaten them with genuinely painful consequences if they do a half-hearted job.

This difficulty in differentiating metabolic and motivational factors might be one reason for why ego depletion was so successful (Baumeister et al., Reference Baumeister, Bratslavsky, Muraven and Tice1998): task performance decreasing as motivation wanes might have been wrongly attributed to local resource depletion, rather than to motivational factors (cf. Vohs et al., Reference Vohs, Schmeichel, Lohmann, Gronau, Finley, Ainsworth, Alquist, Baker, Brizi, Bunyi and Butschek2021). If that’s true, there is a risk in invoking metabolic factors in the explanation of effort: namely, that it serves as a tempting way to account for behavioral variations that are better explained in terms of the computational model itself.

Haueis and Colaço rightly reject local resource models of cognitive effort. Yet they argue that their main alternative, computational motivational models such as Kurzban’s opportunity cost model (Kurzban et al., Reference Kurzban, Duckworth, Kable and Myers2013) or Shenhav’s Expected Value of Control model (Shenhav, Botvinick, & Cohen, Reference Shenhav, Botvinick and Cohen2013; Shenhav et al., Reference Shenhav, Musslick, Lieder, Kool, Griffiths, Cohen and Botvinick2017), should take on board metabolic factors without much explanatory cost. For example, one could leave the EVC model intact but invoke metabolic factors as an external enabling condition that affects outcomes only insofar as it modulates parameters within the model.

Haueis and Colaço point to two recent theoretical proposals as successful examples of integrating metabolic and computational elements (Holroyd, Reference Holroyd2025; Zenon, Solopchuk, & Pezzulo, Reference Zenon, Solopchuk and Pezzulo2019). These are the strongest cases for hybridization in the effort literature. My concern is narrower. In both models, the metabolic component explains the same behavioral change that motivational/computational models already predict: gradual disengagement and performance decrements at costly tasks. This makes it difficult to determine, from behavioral data alone, whether the metabolic addition is doing independent explanatory work or tracking motivational dynamics under a different description. Metabolic data are not completely absent but scarce (though see Wiehler et al., Reference Wiehler, Branzoli, Adanyeguh, Mochel and Pessiglione2022). My point is simply that until much stronger convergent data are in hand, the explanatory benefit of hybridization in this domain remains uncertain.

Hybridization does not come without cost. In both experimental and real-world settings, motivation is found to provide a better explanation of self-control behavior than metabolic depletion. Muraven and Slessareva (Reference Muraven and Slessareva2003) showed that “depleted” participants who were offered cash rewards or told their efforts would help others performed just as well as nondepleted participants. Of course, in extreme cases, metabolic factors will matter: no money in the world will make you ace the LSAT when you’ve been awake for 72 hours straight. But in normal circumstances, the burden of proof for showing more than niche relevance for metabolic considerations is not on the advocates of motivational/computational models.

So far, I have mainly argued on methodological grounds that the literature on mental effort should not consider the hybridization of motivational/computational models and metabolic models cost-free. But there is also independent conceptual evidence that favors motivational/computational models over metabolic models.

Specifically, there is a nascent literature on the philosophy of effort (Bermúdez, Reference Bermúdez2025; Bradford, Reference Bradford2015; Massin, Reference Massin2017; Nelkin, Reference Nelkin2016; von Kriegstein, Reference Von Kriegstein2019). Two points from this literature matter in explaining model choice. The first is that effort strikes many as a unified phenomenon; the second is that effort has a normatively rich profile.

Philosophers observe that the functional role effort and difficulty play is quite independent of whether the effort is mental or physical. For example, we think both Magnus Carlsen and Usain Bolt have performed difficult achievements (Bradford, Reference Bradford2015), and we attribute agency to both mental and bodily efforts (Massin, Reference Massin2017). The evolutionary function of the feelings of effort as a way of prioritizing tasks, too, is motivational and cuts across domains (Bermúdez, Reference Bermúdez2025; Kurzban et al., Reference Kurzban, Duckworth, Kable and Myers2013). On a motivation-first picture, this datum is explained because motivation, like effort, cuts across domains (Hendrickx, Reference Hendrickx2026), whereas metabolic accounts usually must differentiate between mental and bodily effort as they draw on different resources and markers (say, glucose and lactate).

Another worry concerns the normative profile of effort. We care about the efforts people make. If my life is in danger and you make a lukewarm effort, I’ll blame you even if you had a long day at work! Grounding this normative profile requires tying effort to something more than metabolites: glucose and ATP are poor candidates for blaming a person, whereas an inappropriately low state of motivation is a direct indicator of the agent’s cares and values (Hendrickx, Reference Hendrickxforthcoming; cf. Nelkin, Reference Nelkin2016). At least prima facie, this normative profile is data in the sense that we usually take our social practices to be sensitive to descriptive phenomena.

Of course, these philosophical considerations might be trumped by future empirical evidence indicating a hitherto unknown strong role of metabolic considerations in explaining the psychology of effort. And Haueis and Colaço make a strong case that, among a variety of cognitive explanations, theorists should be more mindful to explicitly include metabolic factors in their models. But motivational/computational models have seen spectacular progress in predicting behavior and shedding genuine light on deep philosophical questions about effort and difficulty (Bermudez, Reference Bermúdez2025; Hendrickx, Reference Hendrickx2026, Reference Hendrickxforthcoming). Introducing metabolic considerations as a free parameter in otherwise functional models does not come at no cost and should be done only with utmost caution.