Brando Miranda — November 2026 · ~16 min read

TL;DR. Animals learned from consequences long before humans could argue with their thoughts, and repeated behavior can become habitual rather than something we decide afresh. My hypothesis is that changing those consequences and practicing new responses will often do more for entrenched anxiety and avoidance than verbal cognitive restructuring—the part of cognitive behavioral therapy (CBT) that examines and revises thoughts—especially when executive control is strained. That is not proof that CBT fails: CBT includes behavioral learning, and cognition can help; my objection is to treating a better explanation as if it were already a changed response.


I can recognize that a thought is exaggerated and still feel exactly the same. The explanation changes. The reaction doesn’t.

That gap is why the cognitive part of CBT has never made intuitive sense to me. I’ve been thinking about it since discovering Judson Brewer’s work during my time at the Massachusetts Institute of Technology, as I mention on my performance psychology page. A conversation with a friend brought the question back into focus: what actually changes a response that understanding hasn’t changed?

My answer starts with evolution, then learning, then the practical difficulty of reasoning while anxious. I want to distinguish what the evidence supports from the further bet I’m making.

Learning came first

Evolution had a very long time to produce organisms that could adjust their behavior. It did not wait for language, introspection, or a therapist’s worksheet.

Even Aplysia, a sea slug, can learn from the consequences of its actions. Brembs and colleagues demonstrated operant reward learning and associated cellular changes in this animal. Within vertebrates, Grillner and Robertson describe striking conservation of basal ganglia circuits—networks involved in selecting actions and learning—from lampreys to mammals.

The elaborate prefrontal machinery humans use for deliberation came much later. The anatomical history is more complicated than “old emotional brain, new rational brain”: Preuss and Wise distinguish older prefrontal regions shared by mammals from later additions and expansions in primates. The sea-slug result also does not mean a slug has a human striatum. These are two pieces of evidence: learning without human-style reasoning, and conserved vertebrate circuitry.

But the order matters to my argument. Learning from consequences is foundational; verbal reasoning is not where learning began. My suspicion is that therapy should work with those established learning capacities rather than assume that an explicit conclusion will overrule them.

Here is the inferential jump: foundational does not automatically mean more effective as a treatment target. Evolution also produced flexible planning, and a newer capacity can be useful precisely because it changes how an older one operates. Evolution gives me a reason to investigate an experience-first approach, not a theorem that ranks therapies.

Consequences change behavior. Repetition can make it stick.

There are two related parts to what I mean. I am not claiming two independent brain systems.

First, operant conditioning: learning from what an action does. If leaving an uncomfortable situation brings relief, that relief can make leaving more likely next time. This is negative reinforcement: a behavior is strengthened because it removes something unpleasant. “Negative” does not mean punishment, and “reinforcing” does not mean healthy in the long run.

Imagine someone avoiding a presentation. They already know that avoidance is shrinking their life. But canceling brings immediate relief. A correct sentence about the long-term cost does not necessarily compete well with that immediate consequence.

Reward-learning research gives this intuition a mechanism. A reward prediction error is a mismatch between expected and received reward; Schultz, Dayan, and Montague connected dopamine activity with such learning signals. That is evidence about learning, not proof that every feeling is a dopamine signal or that all therapy works through reward prediction errors.

Second, repeated behavior can become habitual. What I call “pure” habit change is the repetition side: a cue increasingly brings out a response without a fresh argument about whether to perform it. The emphasis shifts from changing the attractiveness of an outcome to changing what happens when the cue arrives.

The distinction matters. A goal-directed action remains sensitive to what its outcome is currently worth. A habit can persist even after that outcome becomes less valuable. Dolan and Dayan’s review describes interacting circuits involving the dorsomedial striatum for goal-directed control and the dorsolateral striatum for habitual control, with much of the causal evidence coming from animal studies. In Yin and colleagues’ rat experiment, dorsolateral striatal lesions disrupted habit formation while preserving sensitivity to the outcome’s value.

So “dorsal striatum” is too broad, and “the habit lives in this one spot” is too simple. Nor does repeating something prove it has become a habit in the experimental sense. But the important possibility survives: a response can become less dependent on deciding to do it each time.

That is what I want from change. Not a better argument I must reconstruct whenever I’m distressed, but a different response I have practiced enough to use.

There is a catch: the same persistence makes unwanted habits hard to change. “Stronger” can mean more stubborn, not healthier. Changing a habit’s perceived payoff and rehearsing an alternative are related levers, not interchangeable descriptions of one guaranteed cure.

Feelings matter, but reward is not a synonym for feeling

My personal shorthand is that feeling something changes me more than explaining it. I stand by that as a description of my experience. It is not an adequate definition of reward learning.

Berridge, Robinson, and Aldridge distinguish pleasure, motivation, and learning—“liking,” “wanting,” and learned associations. Those can come apart. Something can pull us toward it without being especially pleasurable, and learning need not involve an intense conscious feeling.

This is also why repetition deserves its own place in the argument. The valuable practice might be quiet and uneventful. I do not need every useful repetition to produce a breakthrough.

Brewer’s approach interests me because it asks a different question from “Can you defeat this thought?”: What is this behavior actually giving you? Ludwig, Brown, and Brewer propose that attending to an unhealthy behavior’s experienced consequences can change its learned value, reducing the need for effortful self-control. That is a mechanistic proposal, not a demonstration that mindfulness beats CBT.

The orbitofrontal cortex, a prefrontal region involved in representing expected outcomes, features in this account. But it is not a single reward scoreboard insulated from thought. Schoenbaum and colleagues describe a more specific role in using representations of outcomes. The involvement of prefrontal cortex itself should warn us against equating “experience-first” with “subcortical only.”

Anxiety makes the demand for deliberation harder

The evolutionary argument is about what learning can do. The executive-function argument is about when a technique is usable.

Cognitive restructuring can ask me to hold a thought in mind, consider evidence, generate an alternative, and use it while distressed. Those demands involve executive functions: processes such as working memory and flexible control of attention. If those resources are already strained, adding a demanding mental task may be a poor fit.

Stress can impair working memory and cognitive flexibility, although its effects on inhibition are more mixed, as Shields and colleagues’ meta-analysis explains. “The prefrontal cortex switches off” is too literal. And stress research does not establish that reappraisal always fails: Shermohammed and colleagues found preserved reappraisal success under their laboratory stress manipulation.

I also have attention-deficit/hyperactivity disorder (ADHD), so the effort involved matters personally. But neither ADHD nor anxiety licenses a blanket inference that cognitive therapy cannot work. Willcutt and colleagues found executive-function weaknesses associated with ADHD while emphasizing that they are not universal. And Safren and colleagues found benefits from an ADHD-focused CBT program in adults who remained symptomatic while taking medication. That is evidence against dismissing the whole approach, not a test of which component helps anxiety in people with ADHD.

The strongest reply is that CBT trains skills while people are calm, so they can use them later. Fair. That answers the caricature of a therapist demanding elaborate reasoning in the middle of panic. It also brings practice back into the explanation—but practice alone does not prove that habit formation, rather than the content being practiced, caused improvement.

My narrower question is: how much deliberation must a person supply at the moment the skill is needed? I suspect that matters a great deal. Mindfulness and concentration practice appeal to me partly as ways to rehearse a response, but they also take attention and practice. I cannot assume they have a smaller mental-effort bill for everyone.

What “experience-first” would mean in therapy

By experience-first, I mean arranging opportunities to act differently and learn from the consequences, then using reflection to understand and retain what happened. I do not mean that thoughts are outside experience, or that reflection must always come last.

Consider the presentation example. “People probably won’t reject you” is a proposition. Speaking, making a mistake, and discovering that the interaction remains manageable is a test. Reflection can help identify the prediction, interpret the outcome, and plan another attempt. My objection is to treating agreement with the proposition as if that test had already happened.

This is where my criticism meets CBT rather than simply opposing it. Exposure and behavioral experiments already arrange experiences like these. In Craske and colleagues’ inhibitory-learning account, exposure can build new learning that competes with an existing threat association. Violating a feared prediction is one strategy for strengthening that learning. The goal is not merely to feel calmer before the session ends, and the old fear memory need not be erased.

That is not identical to operant conditioning or dorsolateral-striatal habit learning. Threat learning, consequences of avoidance, and practiced responses can all matter without being one mechanism. Calling all of them “bottom-up” would hide the distinctions my argument needs.

I also have to drop the stronger claim that thinking cannot influence learned value. Doll and colleagues showed that instructions could bias choices despite contradictory feedback. Li, Delgado, and Phelps found that instructed reward probabilities improved performance and altered reward-learning-related brain responses. Those are laboratory tasks, not therapy trials, but they undermine the idea of a learning system that verbal knowledge cannot reach.

So my bet is about relative contribution and reliability, not an impermeable wall between thought and learning.

The goal, the patient, and the timing

A practicing clinician’s perspective sharpens the goal: change need not mean making a feeling disappear. I might still feel anxious and nevertheless give the presentation rather than escape it. What changes is my relationship to the feeling and what I do next—not necessarily the feeling’s immediate intensity.

That is more than a personal perspective. Hayes and colleagues describe psychological flexibility: being able to act in service of what matters while remaining open to present experience. It is central to acceptance and commitment therapy (ACT), itself part of the cognitive-behavioral family. Stockton and colleagues’ review found mediation results generally consistent with this model, but also important methodological limitations. That is promising process evidence, not proof that acceptance is the sole cause of improvement—or an argument against every form of CBT.

The patient matters, too. Does someone need a concrete instruction? Do they want to act but feel blocked by fear? Does being directed provoke pushback? Are they not yet participating in a shared therapeutic task? I find these useful clinical questions, not a validated four-way taxonomy or a basis for diagnosing someone from their cooperation. There is research on one related distinction: Beutler, Edwards, and Someah found that patients high in reactance—pushback against being directed—did better with a reflective, less directive therapist stance than an authoritative one. That does not establish four fixed patient types, or how common each is.

And the task is not delivered in a vacuum. The communication and the relationship matter: Flückiger and colleagues found a robust association between the therapeutic alliance—the collaborative relationship—and outcomes. An association is not an experimental demonstration of causation, and this review does not establish when to introduce a particular cognitive technique.

My takeaway is to offer an intervention, notice how the person responds, and adapt the next step. Timing and delivery are part of that judgment. This qualifies my argument: an unsuccessful reframe might reflect a mismatch with the person or the moment, not a useless cognitive process. It does not establish that early restructuring is generally harmful. Experience-first should mean learning to respond differently, not demanding a new feeling on command—or ignoring the person doing the learning.

The clinical evidence has to constrain the story

A plausible neuroscience story is not enough. Three findings keep me honest.

CBT works better than placebo conditions on average. Carpenter and colleagues’ review included 41 randomized trials for anxiety and related disorders. The pooled target-symptom effect was Hedges’ g = 0.56, with a 95% confidence interval of 0.44–0.69—a standardized benefit favoring CBT. The estimate has limitations, including larger effects in analyses restricted to treatment completers. Still, an argument that CBT simply cannot work has to confront this evidence. A treatment package’s benefit does not, by itself, identify its active component.

Explicit cognitive restructuring is not necessary for every effective treatment. In Richards and colleagues’ trial of 440 adults with depression, behavioral activation—helping people re-engage in activities—met the prespecified criterion for being no worse than CBT by a clinically important amount. At 12 months, the reported difference was 0.1 points on the nine-item Patient Health Questionnaire, with a 95% confidence interval of −1.3 to 1.5; the trial’s non-inferiority margin was 1.9 points. This supports a behavioral treatment option for depression. It does not establish superiority, generalize automatically to anxiety, or show that cognitive change played no role.

Cognitive work can contribute, including before exposure. Johnco and colleagues randomized 249 adults with elevated public-speaking anxiety to an intensive exposure session delivered in three ways. Behavioral experiments and cognitive restructuring before exposure both outperformed exposure without explicit expectancy processing. The anxiety advantage of behavioral experiments over restructuring plus exposure was not statistically significant, and follow-up lasted one week. Still, the behavioral-experiment condition produced greater expectancy change in a shorter session, which the authors read as a possible modest advantage for challenging a feared prediction after exposure rather than before it. Nor can I claim restructuring alone is inert: Ciharova and colleagues’ network meta-analysis—a synthesis combining direct and indirect treatment comparisons—found benefits for restructuring, behavioral activation, and their combination in adult depression, without evidence that their effectiveness differed.

Those findings leave room for my hypothesis, but not for announcing that the cognitive component is useless or that “experience first, cognition second” is a universal sequence. They also do not establish that mindfulness is superior. Goldberg and colleagues’ meta-analysis found no detectable average difference between mindfulness-based interventions and established evidence-based treatments. That is not proof of equivalence for every condition or person.

Why “so much evidence” does not settle it

The obvious objection is that CBT has a huge research literature. Why should my evolutionary intuition outweigh it? It shouldn’t. But three questions keep getting collapsed: does the treatment package help, is it better than credible alternatives, and which components cause the improvement? A large, well-conducted evidence base can increase confidence in the first answer without settling the other two.

More studies do not automatically mean a better treatment. Cuijpers and colleagues’ large review of CBT for depression found substantial benefits against control conditions, but only a small advantage over other psychotherapies that became statistically nonsignificant in most sensitivity analyses. That does not prove all therapies are equivalent, and depression is not every disorder. It does show why “most studied” and “best” are different claims.

Who conducts the research can matter. Researcher allegiance means investigators’ preference for a treatment or theoretical approach. Munder and colleagues found an association between allegiance and treatment outcomes across 30 meta-analyses. That is a reason to scrutinize comparisons, not proof of fraud or a problem unique to CBT. Nor does it make favorable CBT findings vanish: Tolin’s review found an advantage over psychodynamic therapy, though not interpersonal or supportive therapies, and a CBT advantage remained after statistically controlling for allegiance ratings. An allegiance association alone cannot tell us how much reflects bias rather than expertise or genuine treatment differences.

The published literature is not necessarily the complete literature. Driessen and colleagues traced government-funded depression-psychotherapy trials and obtained unpublished results. Adding those results reduced the estimated benefit against control conditions, but did not eliminate it. This was evidence about depression psychotherapies, not CBT alone. It also did not establish that reviewers caused the missing publications: of the thirteen unpublished trials, only two had been submitted and rejected, six were never submitted, and for the remaining five it was unclear whether submission was ever attempted.

My institutional worry goes further: could a field become self-reinforcing when the same therapeutic tradition produces the evidence and judges what deserves recognition? Leichsenring and colleagues argue that many review, funding, and guideline committees are dominated by CBT researchers. But their critical review is not a systematic panel-membership audit, and they disclose psychodynamic commitments of their own. I would not turn that argument into the claim that an American Psychological Association board contains only CBT practitioners, or that CBT’s standing has been explained away. Researcher allegiance, publication bias, and bias in peer review or guideline decisions are distinct claims requiring distinct evidence.

So I am not saying “the studies are biased, therefore CBT is wrong.” I am saying that a large literature for a package containing exposure and behavioral practice does not establish that verbal restructuring is its indispensable ingredient. I want independent replication, balanced expertise in head-to-head comparisons, transparent evaluation criteria, and tests that separate components. Take the benefits seriously; do not let the treatment’s reputation answer a mechanism question that the studies did not test.

The bet I would actually test

For entrenched avoidance and cue-triggered responses, I predict that changing consequences and repeatedly practicing alternative responses will contribute more to durable behavioral improvement than verbal restructuring alone, particularly when executive demands are high.

That prediction needs a fair comparison: a specified problem and population; matched therapist time, expertise, and credible treatment rationales; clearly measured restructuring and behavioral practice; and outcomes beyond whether someone can repeat a more balanced thought. Preregistered outcomes and independent investigators would help test my preferred explanation rather than merely reproduce my own allegiance. I would want everyday functioning, behavior when the cue returns, persistence after treatment, and the effort required to use the skill. Baseline executive function could test the additional prediction about who benefits—not be assumed to explain the result afterward.

I would weaken my view if restructuring alone produced equally durable behavioral gains with comparable or lower effort. A reliable added benefit from restructuring when behavioral practice is held constant would also count against the strongest version of my claim. Conversely, simply failing to detect an added benefit in a small study would not settle it in my favor. And if benefits do not vary with executive function, that challenges my executive-demand explanation even if the broader learning argument survives.

Where I land is still a strong preference, not neutrality. I want therapy to help me learn a different response, not merely become more articulate about the response I already have. Sometimes thought will guide that learning. Sometimes it will consolidate it. Sometimes a practiced action may matter more than another explanation.

Insight can tell me what to change. I want practice and experience to make the change usable.

This is an argument about treatment mechanisms, informed by personal experience, not medical advice. It is not a reason to stop a treatment that is helping you.


If you’d like to cite this post:

@misc{miranda2026cbtexecutivefunction,
  author = {Miranda, Brando},
  title  = {Why CBT Doesn't Make Sense to Me: Evolution, Habits, and the Limits of Insight},
  year   = {2026},
  month  = {November},
  howpublished = {\url{https://cs.stanford.edu/people/brando9/2026/11/08/why-cbt-doesnt-make-sense-to-me.html}},
  note   = {Blog post}
}

Acknowledgments

Thanks to Andrea for the conversations and clinical perspective that helped sharpen this argument.