In short: The AI Substitution Framework maps work across three levels - high, moderate, and low substitution - so leaders can see where AI automates a task, where it co-produces, and where human judgment stays central.
A strategist who once needed a week and a team to model three scenarios can now generate twelve in an afternoon - each tailored to specific constraints, each technically competent. The bottleneck is no longer production. It is evaluation.
This shift is happening across industries, and it demands a new way of thinking about work. The AI Substitution Spectrum provides leaders with a practical framework to evaluate which components of work are most exposed to AI-driven change, and which require human judgment, contextual understanding, or legitimacy. It does not predict job loss. It clarifies how the locus of value shifts when the technical layer of work becomes inexpensive, fast, and widely accessible.
The question is which parts of the work AI does - which parts it substitutes for - and which parts become more central to human expertise.
The Three Levels of Substitution
Level 1: High Substitution
Tasks that are rule-governed, pattern-based, or technically demanding
These are tasks whose value historically derived from technical skill, tool mastery, or the ability to process large amounts of information. They follow rules, templates, or stable patterns that AI can learn and reproduce. Across structured settings, AI systems already operate at human-level or faster speeds, with consistent and reproducible performance where the cost of error is low.
Examples include data cleaning, categorization, and summarization; drafting structured text like reports and outlines; technical design tasks such as mockups and wireframes; editing existing video or audio assets; and running common statistical analyses.
Leadership implication: Organizations should plan for rapid productivity increases in this band and redesign workflows so that technical execution is not the anchor of a role. This is where substitution - of tasks, not necessarily people - will be most pronounced.
Level 2: Moderate Substitution
Tasks requiring curation, synthesis, and iterative assembly
At this level, AI acts as an engine of optionality, rapidly generating components, drafts, or varied approaches. The primary human value shifts from creating from scratch to curating and assembling. AI provides abundance; the human provides coherence.
The core task becomes sifting through technically competent AI outputs to identify what is relevant, filtering out hallucinations or subtle context errors, and synthesizing the best components into a usable whole.
Examples include taking five AI-generated strategic scenarios and blending the best elements into one coherent plan; generating fifty variations of marketing copy and selecting the final five that resonate with specific demographics; and using AI to generate individual UI components, then manually assembling and refining them into a functioning product.
Leadership implication: The skill requirement moves from production speed to editorial discernment. Managers must stop rewarding the mere generation of volume and start rewarding the ability to quickly distinguish between outputs that are technically plausible versus contextually accurate.
Level 3: Low Substitution
Tasks requiring judgment, sensemaking, legitimacy, and accountability
This is work in which the value lies in the ability to interpret ambiguity, weigh competing priorities, navigate relational dynamics, and take responsibility for decisions. AI can contribute information, but it cannot assume legitimacy or accountability for the outcome.
Consider a hospital system CMO who must decide whether to adopt an experimental treatment protocol. Early results are promising but the evidence base is thin. Patients are asking for it. Adopting too early risks harm; waiting too long means patients who might have benefited don’t. The decision will be scrutinized regardless of outcome.
Examples include defining what “impact” should mean for a program; assessing recommendations within political, cultural, or ethical constraints; shaping narrative and meaning for a team; making decisions that carry risk, loss, or system-wide consequences; establishing standards for quality; and teaching, mentoring, and developing capability in others.
Leadership implication: As AI expands optionality and production capacity, judgment becomes the central human contribution. Organizations must re-anchor roles around discernment, coherence, and values-based decision-making. These functions become more - not less - important.
The Collapse of the Technical Bottleneck
The most significant system-level change is the collapse of the technical production bottleneck.
Previously, ideas were cheap, but execution - coding, designing, writing - was expensive and slow. The difficulty of execution acted as a natural filter. Now, AI allows for the near-instantaneous generation of massive volumes of technically competent work: code, copy, designs, analysis.
The dam has broken.
Once AI can generate multiple tailored outputs quickly, organizations face new pressures: increased demand for personalization and rapid iteration; a desperate need for rigorous evaluation workflows upstream; and a higher volume of content requiring human discernment to ensure safety and brand alignment.
This is not AI replacing expertise. It is AI enabling individuals with expertise to operate at a level of specificity that previously required teams, time, or budget.
Using the Spectrum: Leadership Diagnostics
Conclusion
The AI Substitution Spectrum is not a prediction about job disappearance. It is a framework for understanding how the location of human expertise shifts when AI lowers the cost of production.
As the technical layer collapses, organizations must invest in evaluation capability, judgment, and coherence - and redesign roles to reflect a world where production is easy and discernment is scarce.
But this redesign carries its own risk. If the work that once built judgment disappears, how will the next generation of leaders develop the expertise to exercise it? That question may be the most consequential one this framework surfaces.
Frequently Asked Questions
What is the AI Substitution Spectrum?
The AI Substitution Spectrum is a diagnostic framework that maps which components of work are most exposed to AI-driven substitution and which require human judgment, contextual understanding, or legitimacy. It sorts work into three levels: High Substitution, Moderate Substitution, and Low Substitution. It clarifies how the locus of value shifts within a role when the technical layer of work becomes inexpensive, fast, and widely accessible.
What are the three levels of substitution?
Level 1, High Substitution, covers rule-governed, pattern-based tasks such as data cleaning, structured drafting, and first-pass analysis, which AI already performs reliably. Level 2, Moderate Substitution, covers curation, synthesis, and iterative assembly, where AI generates options and humans provide coherence. Level 3, Low Substitution, covers judgment, sensemaking, legitimacy, and accountability, where AI can contribute information but cannot assume responsibility for outcomes.
Does the framework predict which jobs AI will replace?
No. The Spectrum is a diagnostic for understanding where value is shifting within any given role, not a classification system for jobs. Most roles span all three levels, so the framework is for seeing how the location of human expertise moves rather than for labeling jobs as exposed or safe. It does not predict job loss.
How should leaders use the Spectrum?
Leaders can evaluate a workflow, function, or role by asking which components are rule-based candidates for Level 1, which require human synthesis or forensic review at Level 2, and which involve legitimacy, risk ownership, or defining value at Level 3. A fourth question asks where faster production will create new pressures, such as whether evaluation capacity can keep pace with ten times the content. The answers show where to re-anchor each role.
What is the collapse of the technical bottleneck?
Previously, ideas were cheap but execution such as coding, designing, and writing was expensive and slow, and that difficulty acted as a natural filter. AI now allows near-instantaneous generation of large volumes of technically competent work, so the bottleneck shifts from production to evaluation. This is not AI replacing expertise; it is AI enabling people with expertise to operate at a level of specificity that once required teams, time, or budget.