How to Classify AI Work: The Cognitive Use-Stack Approach

By Dr. Staffan Canback, Tellusant, Inc. | 2026-09-25

Tellusant operates at different cognitive levels with AI, from the mundane to the inventive. Here we classify the uses by cognitive order.

It strikes me that most people discuss AI in terms of tasks. I have to think that this is a bit reductive. It is straightforward to think about tasks, but what are the underlying cognitive complexities?

The graph below serves as a classification of AI use for me and my colleagues. On any given day we work at all four levels. In the long term, levels 3 and 4 are what distinguish us, while levels 1 and 2 are critical for our productivity.

AI CLASSIFICATION SCHEMA

flowchart LR %%{init: {'themeVariables': { 'fontFamily': 'Arial'}}}%% AI["`**AI Cognitive Use**`"]:::none L4["`**Level 4 — Invent** *Theory, method & concept development* Problem framing and/or method is being invented`"]:::strong L3["`**Level 3 — Synthesize** *Problem solving & synthesis* Problem known Method must be developed`"]:::strong L2["`**Level 2 — Analyze** *Analytical support* Problem known Method substantially known Interpretation needed`"]:::strong L1["`**Level 1 — Execute** *Task execution* Problem known Method known`"]:::strong AI --> L4 AI --> L3 AI --> L2 AI --> L1 %% ========= STYLES ========= classDef strong fill:#E8F5E9,stroke:#1B5E20,stroke-width:2px,color:#111; classDef none fill:transparent,stroke:transparent,color:#111;

Our Uses of AI

AI use can be classified according to the cognitive role it plays in completing a task. The classification is not based on the technical complexity or size of the output. Instead, it reflects how well the problem and method are specified and how much analysis, synthesis, and conceptual development are required.

My AI use can range from figuring out how to:

Create a graph in Mermaid like the one above in 10 minutes (Level 1)

Analyze productivity trends for all countries and a few thousand companies. Countries in a few hours; entire task over weeks (Level 2)

Apply Möbius transforms to inflation or relative market share in order to achieve symmetry. On and off over 6 months (Level 3)

Invent The Strategy Grid framework over the past year (Level 4)

The Cognitive AI Use-Stack

Here is a bit more detail on the graph. From bottom to top:

Level 1 — Execute: Task Execution

The problem and the method are known. AI performs a well-specified, mundane task with limited need for interpretation or problem solving.

Typical uses include formatting and restructuring information, translation, data conversion, routine coding, syntax correction, straightforward information retrieval, and executing established calculations or procedures.

A task can be technically extensive and still belong at this level if the method is predetermined.

Level 2 — Analyze: Analytical Support

The problem is known and the general method is substantially understood, but interpretation and judgment are required. AI processes information to identify patterns, diagnose problems, make comparisons, or derive conclusions.

Typical uses include summarizing and comparing evidence, debugging code, analyzing data, interpreting statistical results, identifying anomalies, assessing alternatives against specified criteria, and explaining why a result occurred.

Level 3 — Synthesize: Problem Solving and Synthesis

The problem is known, but the method for solving it is not fully specified. AI helps formulate an approach, integrate information from different sources or disciplines, evaluate alternative methods, and construct a solution.

Typical uses include designing an analytical approach, developing a model architecture, combining research findings into a coherent framework, exploring trade-offs, designing an algorithm, or finding a solution to a novel and incompletely structured problem.

Level 4 — Invent: Theory, Method and Concept Development

AI participates in the development of new intellectual constructs rather than primarily applying existing ones. Both the framing of the problem and the method may be open to development.

Typical uses include creating new methodologies, mathematical transformations, metrics, conceptual frameworks, theories, algorithms, or decision approaches.

The defining characteristic is not simply that AI produces something new, but that it contributes to the development of the underlying method or concept.

— — —

The four levels therefore represent a progression in cognitive contribution:

\[\mathbf{Execute \rightarrow Analyze \rightarrow Synthesize \rightarrow Invent}\]

A practical way to distinguish the levels is to ask how much of the problem and solution method is known in advance:

Level 1: problem known, method known.
Level 2: problem known, method substantially known, interpretation required.
Level 3: problem known, method must be developed.
Level 4: problem framing and/or method itself is being developed.

The levels describe the nature of AI use rather than the inherent difficulty of the task. For example, generating hundreds of lines of routine code may constitute Level 1 execution, while developing a short new mathematical formulation could constitute Level 4 invention.

Sources

Anderson, Lorin W., and David R. Krathwohl, eds. 2001. A Taxonomy for Learning, Teaching, and Assessing: A Revision of Bloom’s Taxonomy of Educational Objectives. New York: Longman.

Parasuraman, Raja, Thomas B. Sheridan, and Christopher D. Wickens. 2000. “A Model for Types and Levels of Human Interaction with Automation.” IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans 30 (3): 286–297. https://doi.org/10.1109/3468.844354.

Doshi, Anil R., and Alastair P. Moore. 2026. “Toward a Human–AI Task Tensor: A Taxonomy for Organizing Work in the Age of Generative AI.” In Handbook of Artificial Intelligence and Strategy, 124–142. Edward Elgar Publishing. https://doi.org/10.4337/9781035345885.00014.

Scott, Inara. 2026. “The AI Cognitive Pyramid: A Conceptual Framework for Generative AI Use, Critical Thinking, and Learning Design in Higher Education.” SSRN, March 1. https://doi.org/10.2139/ssrn.6353318.