The Five Categories of Bullshit

In the AI Story

The taxonomy emerged from Graeber's anthropological discipline of taking categories seriously. Most discussions of pointless work treat it as a single phenomenon. Graeber recognized that distinct mechanisms produce distinct species, each requiring distinct analysis. A flunkey is not a duct-taper, and the institutional reform that would eliminate one will not touch the other. The categories function as a clinical taxonomy — the diagnostic precision a doctor needs to distinguish between superficially similar conditions.

Applied to AI, the taxonomy generates predictions that have begun to be tested empirically. Duct-taping faces direct elimination as AI integrates the systems that previously required human bridges. Flunkies persist with new titles — the 'Chief AI Strategy Officer' performing the same status-signaling function as the executive assistant of an earlier era. Box-tickers multiply as new AI governance frameworks generate compliance requirements that did not previously exist. The taxonomy's value is showing why the simple narrative of 'AI eliminates jobs' obscures more than it reveals.

Graeber documented overlap among the categories. A real position frequently combines elements: the corporate compliance officer is partly a box-ticker (generating documentation), partly a goon (existing because regulators require it), partly a flunkey (making the legal department appear robust). The taxonomy's purpose is not to sort positions into mutually exclusive boxes but to identify the multiple mechanisms operating in any given case.

The empirical method Graeber used — reading thousands of self-reports from workers across industries — produced a taxonomy grounded in the lived experience of those performing the work. This methodological choice distinguished his approach from the productivity literature that infers function from organizational charts. The workers performing bullshit jobs know exactly what category they occupy. They simply cannot say so without professional consequence.

Origin

Graeber developed the taxonomy through analysis of testimonies submitted in response to his 2013 essay. The categories crystallized as he sorted hundreds of self-descriptions into recurring patterns. The five-category structure was not imposed in advance but emerged from the data — a fact Graeber emphasized when defending the framework against critics who suggested he had constructed the categories to reach predetermined conclusions.

Key Ideas

Mechanism specificity. Each category names a distinct institutional pathology, requiring distinct diagnostic and remedial approaches.

Flunkies signal status. Their function is to make a superior visible as superior — a function that survives technological change because it serves psychology, not productivity.

Goons are mutually self-justifying. Each side employs them because the other does, producing arms races that AI accelerates rather than resolves.

Duct-tapers face direct elimination. AI fixes the underlying systems they patched — but their displaced labor must be absorbed somewhere.

Box-tickers multiply. Every new AI capability generates new compliance requirements, expanding rather than contracting the documentation apparatus.

Debates & Critiques

The taxonomy has been criticized as too parsimonious by some scholars who identify additional categories (such as 'reputational firewalls' or 'symbolic specialists') and as too generous by others who argue it should distinguish degrees of pointlessness more rigorously. Graeber treated the five categories as analytically useful rather than ontologically fixed — a working framework subject to revision as evidence accumulated.