StewAI Blog · AI Workflow Design · 14 min read
Thinking Work Is Made of Fourteen Blocks (Whether You Like It or Not)
We decomposed every reasoning workflow we ship into reusable primitives. Fourteen blocks fell out. Here is the whole catalog, what each one does, the evidence behind it, and the honest part: fourteen is an engineering result, not a law of nature.
Frequently asked questions
What are the fourteen reasoning primitives?
Structured Intake, Isolated Panel, Checklist Sweep, Criterion Scorer, Mechanical Aggregator, Verdict Rollup, Consider The Opposite, Brief Renderer, Steelman Balance, Kill Criteria, Load Bearing Assumptions, Citation Check, Dedup Cluster, and Reference Class. Each is a standalone recipe with typed inputs and outputs that can run on its own or be called as a subrecipe inside a larger workflow.
Is fourteen the real number of thinking primitives?
No, and that is stated plainly in the article. Proposed counts across the literature range from zero (Piantadosi argues there are no cognitive primitives at all) to eleven (Schank's Conceptual Dependency primitive acts) to about sixty-five (Wierzbicka's semantic primes). The count is a consequence of how finely you slice, not a discovered constant. The narrower defensible claim is that a dozen or so well-chosen primitives cover most reasoning moves that knowledge work actually needs.
Why does the structure matter more than the prompt?
Most of these blocks work because of what the model is structurally prevented from seeing. Eight of the fourteen enforce isolation with a real parallel loop so one judgment cannot anchor another, and eleven compute their numbers in code rather than letting a model write them. Across the catalog there are more deterministic compute steps than model calls, so counts and composites cannot drift from the evidence behind them.
Read the full article on StewAI