AI Task Disruption Matrix
Rank tasks by definition, documentation, complexity, and available examples.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 97%
The AI Task Disruption Matrix evaluates work along dimensions that are more predictive than whether a task appears easy. One dimension asks whether the work is clearly defined and supported by documented procedures, standards, or examples. Another distinguishes systematic application from open-ended original thought, while still recognizing that intellectually demanding work can be routine and well documented. Tasks with a clear definition of good and abundant training material are prime candidates for near-term AI performance, even when humans require years of professional education to perform them. Tasks involving genuinely novel thought, sparse precedent, or opaque judgment are less directly exposed. Regulation, liability, and deployment constraints are evaluated afterward because they may delay adoption even when technical capability already exists.
Origin
Kip Bodnar proposed a two-by-two during Marketing Against The Grain, and Nathan Labenz refined it by separating intellectual difficulty from routine, documented execution.
Core principles
- 01Clear standards make tasks easier to automate.
- 02Intellectual difficulty does not protect routine work.
- 03Training data and a visible definition of good are major enablers.
- 04Original thought remains less exposed than documented practice.
How to run it
- 1
Define the Unit of Work
Select one concrete task and describe its inputs, expected output, and user of that output.
Pro tip Use task-level units such as taking a medical history rather than broad labels such as healthcare.
- 2
Score Definition and Documentation
Assess whether practitioners share explicit procedures, standards of care, templates, or professional rules.
Pro tip Look for manuals, training materials, rubrics, and exam questions.
Watch out A task can be documented without being simple.
- 3
Assess Precedent
Determine whether the task has enough examples for a model to learn what strong performance looks like.
Pro tip Distinguish accessible training material from private institutional knowledge.
- 4
Separate Routine from Original Thought
Classify whether the task applies established patterns or creates an analysis for which little precedent exists.
Pro tip Consider whether a competent answer could be derived from known standards.
Watch out Do not assume all creative-looking output requires original thought.
- 5
Prioritize and Apply Constraints
Prioritize well-defined, precedent-rich tasks for testing, then account for regulation, liability, and integration effort.
Pro tip Keep technical readiness and permission to deploy as separate scores.
In the wild
Taking a patient history and generating a differential diagnosis is intellectually demanding, but medicine supplies standards of care, extensive educational material, and clear performance benchmarks. The matrix therefore ranks the task as technically exposed even though regulation and clinical accountability may delay deployment.
→ The organization distinguishes AI capability from the legal and operational readiness to use it.
An analyst develops a thesis about a technology released after the model's training cutoff, with little published precedent and no settled rubric for success. The task scores low on documentation and precedent, so AI is used for support rather than primary judgment.
→ Human original thought remains central while AI handles bounded transformations and research assistance.
Common mistakes
Equating Complexity with Safety
Highly trained professional tasks can still be routine, documented, and learnable by AI.
Ignoring Deployment Constraints
Technical capability does not automatically overcome regulation, liability, or organizational resistance.
Scoring Entire Professions
Occupations mix tasks from several quadrants, making role-level classification misleading.
Is it for you?
Best for
Executives, operators, and professionals prioritizing AI experiments across a portfolio of tasks.
Not ideal for
Predicting exact employment totals or long-term breakthroughs in general-purpose AI.
From the transcript
“There's a lot of uncertainty. There's a lot of judgment calls being made in it.”
“I think things that are intellectually demanding can still be like fairly routine and well documented.”
“You know, as long as there is a clear sense of what good looks like and, you know, lots of training data out there, then…”
From the episode
GPT-4 Beta User Reveals What Jobs It Will Destroy In 2023 with Nathan Labenz (#103)
Nathan Labenz