Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) have moved rapidly from metabolic medicine into neurology, but the evidence to mid-2026 refuses to resolve into a single narrative of neuroprotection. Large cardiovascular outcome trials establish these drugs for ischaemic stroke prevention, whereas rigorous phase 3 trials in Parkinson's disease and Alzheimer's disease found no effect on clinical progression despite favourable cerebrospinal fluid biomarker shifts. I argue that this pattern is not contradictory but structured by two axes. The first is tissue reversibility, which determines whether disease modification is possible at all: a remodellable vascular substrate responds, an autonomously degenerating parenchyma does not, and a functionally disordered but structurally intact network should. The second is molecular access, which determines which agent can work, and it inverts across indications: long-acting human analogues reduce stroke risk while short-acting exendins do not, yet exenatide lowers intracranial pressure in idiopathic intracranial hypertension because the choroid plexus lies outside the blood-brain barrier. Chronic migraine occupies the reversible, extra-barrier quadrant. Recent receptor-level work shows GLP-1-derived peptides inhibit TRPV1 noncompetitively at an extracellular site without hyperthermia, and that a fragment relieves pain independently of GLP-1 receptor signalling. This generates a falsifiable prediction - analgesia separable from weight loss - and a trial design capable of refuting it.