The neural coding metaphor is so ubiquitous that we tend to forget its metaphorical nature. What do we mean when we assert that neurons encode and decode? What kind of causal and representational model of the brain does the metaphor entail? What lies beneath the neural coding metaphor, I argue, is a bureaucratic model of the brain. Neural coding is a popular metaphor in neuroscience, where objective properties of the world are communicated to the brain in the form of spikes. Most commentators have recognized that the neural coding metaphor is often misused, but they diverge on the extent to which these problems are constitutive of that metaphor. What is wrong with metaphors (R1)? Metaphors can in principle be useful, as they allow reusing concepts from a different domain. But they can also be misleading when applied to very different domains. Perhaps sensory transduction can be framed as a problem of communication. But are perception and cognition really cases of "world-brain communication" (Gallistel)? Unfortunately, this question is rarely explicitly formulated and addressed. Instead, the metaphor captures language and thought in disguise, preempting the meaning of words such as "representation" and "information", in a way that introduces confusion between the organism's and the observer's perspective (information for whom?). To understand what lies beneath "neural codes", one must then take a pragmatic approach: how does the neural coding metaphor unfold in reasonings about brain and cognition? The neural coding metaphor promotes a particular way of understanding causality in complex systems (R2), explanations of the type "A causes B" (e.g. the firing of neuron X causes behavior B). This is an inadequate way of understanding even moderately complex systems of coupled components, such as a system of gears or even a parking lot. In systems, explanations are to be articulated at the level of the organization of processes, not single or even pairs of components. What kind of model of organization features agents that pass formally encoded information along a chain of command with no dynamical constraint? Conceptually, what lies beneath the neural coding metaphor is more than the computer model (Reeke): it is a bureaucratic model of the brain. The neural coding metaphor is tightly linked with the concept of representation, as many commentators have noted (R3). Representation is an important concept, but all supporting arguments are articulated at the level of persons, not neurons-they are considered useful, or necessary to explain certain aspects of cognition. Therefore those arguments do not entail that representations are neural encodings, as the forms of the bureaucratic model. In fact they cannot be encodings, because encodings need a reader and then we need to explain how the activity of reading produces an experience with representational content. The way out of the infinite regress is to conceive representation pragmatically in terms of processes with certain properties. This is a challenge the neural coding metaphor covers.