Walk into almost any organization today and ask about its AI strategy. The answers are remarkably similar.
The organization has created an AI center, hired data scientists and AI engineers, purchased enterprise licenses, trained employees, launched pilot projects, and formed a governance committee. Some teams are using AI to draft documents, analyze data, write code, summarize meetings, prepare presentations, or respond to customers.
These are meaningful steps. They show that the organization recognizes the importance of AI and is beginning to build capability.
But they do not necessarily indicate transformation.
Most organizations measure AI maturity by counting what they have acquired or deployed: models, licenses, pilots, users, training programs, and productivity gains. The underlying assumption is that more AI means greater maturity.
I believe that assumption is wrong.
The most important question is not how much AI an organization has adopted. It is whether AI has changed how the institution learns, creates knowledge, makes decisions, organizes work, and exercises judgment.
That is the real measure of AI maturity.
AI maturity is not primarily a technological progression. It is a cognitive progression. An organization advances as AI moves from a specialized tool at the edge of work to part of the medium through which the institution thinks.
Figure 1. The AI Maturity Ascent: From Tool to Medium. AI maturity progresses from localized assistance to system-level transformation. The first two levels remain within the Tool Phase. The transition to Level Three requires crossing an architectural hurdle. Level Four marks the emergence of AI as a cognitive medium.
Two Phases Separated by One Difficult Transition
The framework shown in the figure divides AI maturity into two broad phases.
The first is the Tool Phase. AI is used to improve specific tasks, support employees, produce analyses, or increase efficiency. Its impact may be significant, but it remains localized. The organization continues to operate through its established workflows, roles, structures, and decision processes.
The second is the Medium Phase. AI is no longer simply applied to work. It becomes part of the environment in which work, learning, knowledge creation, and decision-making occur. AI is woven into the cognitive architecture of the institution.
The difficult transition sits between these two phases.
Moving from Level One to Level Two is largely additive. Organizations hire specialists, purchase tools, expand access, and train users.
Moving from Level Two to Level Three is architectural. It requires redesigning workflows, roles, data flows, decision rights, governance, incentives, and accountability.
This distinction explains why so many organizations plateau at Level Two. They have learned how to distribute AI tools, but they have not yet redesigned the institution around distributed human-AI cognition.
Level One: AI Specialization
At Level One, AI belongs primarily to specialists.
A centralized data science or machine learning team develops models, produces forecasts, and generates reports. Other parts of the organization receive the outputs but rarely participate directly in the underlying process.
Imagine a hospital in which a small analytics group builds a model that predicts patient readmission. The model produces a risk score, which appears in a dashboard or report. Clinicians may use the result, but they do not meaningfully interact with the model, examine alternative interpretations, or shape how the system reasons.
The AI capability may be technically sophisticated, but its institutional role remains limited. Specialists create the intelligence. Everyone else receives it.
At this level, AI improves prediction, classification, automation, and operational efficiency. It may reduce costs or identify patterns that humans would otherwise miss. But it does not fundamentally change the organization’s cognitive workflow.
The experts still think. Others consume the results.
The organization has AI, but AI remains peripheral to how the organization thinks.
Level Two: AI Competency
At Level Two, AI moves beyond the specialist group and becomes a widespread organizational capability.
Employees begin interacting directly with generative AI systems. They use them to draft, summarize, analyze, translate, brainstorm, search, and prepare recommendations. They learn how to frame prompts, question outputs, identify limitations, and recognize uncertainty.
AI literacy begins to function like a new form of organizational literacy.
AI-generated outputs also become a new kind of text. They can be read, interpreted, challenged, debated, revised, and incorporated into ongoing work.
This is an important transition. It distributes capability more broadly and gives people direct access to cognitive resources that were previously available only through specialists.
Yet something fundamental may remain unchanged.
A faculty member uses AI to prepare a lecture, but the curriculum remains the same. A researcher uses AI to summarize literature, but the research process remains linear. A clinician consults an AI assistant, but the care workflow remains fragmented. An executive uses AI to prepare a briefing, but the decision structure remains hierarchical and episodic.
Individuals are becoming more capable with AI, but the organization itself is not yet thinking through AI as an integrated system.
Departments still operate independently. Knowledge continues to move through legacy structures. Decision rights remain unchanged. Performance is still evaluated primarily at the level of individual work.
This is the central limitation of Level Two. AI competency expands individual capability while preserving institutional architecture.
The organization becomes AI-enabled, but not AI-native.
Why Level Two Can Look Like Transformation
Level Two is deceptive because it produces visible activity.
Employees are using AI every day. Productivity appears to increase. New use cases emerge rapidly. Leaders can point to adoption rates, training participation, pilot results, and employee enthusiasm.
The institution may look transformed from the outside.
But the underlying architecture often remains intact.
Work is still organized around the same roles. Decisions still move through the same hierarchy. Data remains fragmented. Governance remains separate from operations. Learning remains episodic rather than continuous.
AI is being added to the institution, but the institution is not being redesigned around AI.
This is why activity alone is a poor measure of maturity. An organization can have thousands of active AI users and still remain at Level Two.
The technology has spread.
The cognitive system has not changed.
The Architectural Hurdle
The wall between Levels Two and Three is not a shortage of more capable models. It is the difficulty of structural redesign.
Crossing this hurdle requires leaders to confront questions that technology adoption alone cannot answer:
Which work should remain primarily human, and which work should become a human-AI process?
Where should AI enter a workflow, and where must human judgment remain decisive?
How should information move across departments and systems?
Who is responsible when decisions emerge through human-AI interaction?
How should outcomes feed back into future learning?
What forms of validation, contestability, and oversight must be embedded in the process?
How should roles, incentives, and authority change when cognition becomes distributed?
These are not software questions. They are questions of organizational architecture.
They also touch institutional power. Redesigning a workflow can change who contributes, who decides, who receives credit, and who bears responsibility.
This is where resistance often emerges. Expanding access to AI is relatively easy because it leaves most institutional arrangements intact. Redesigning work around AI is much more difficult because it changes relationships, authority, and accountability.
Organizations remain at Level Two not because they lack ambition, but because they underestimate the depth of the transition.
They treat AI as a capability to distribute rather than as a new component of the institution’s cognitive architecture.
Level Three: AI-Native Architecture
At Level Three, genuine transformation begins.
The organization stops asking only how employees should use AI. It begins redesigning the institution around distributed cognition.
AI becomes embedded in core workflows rather than sitting beside them. Information flows continuously rather than moving through isolated reports. Feedback loops become shorter. Learning becomes part of daily operations. Human and AI roles are explicitly designed rather than left to emerge informally.
Consider the hospital example again.
At Level One, a model generates a readmission risk score.
At Level Two, clinicians can interact with an AI assistant to understand the score, summarize the patient record, or explore possible interventions.
At Level Three, the entire workflow is redesigned. The AI system continuously synthesizes clinical data, identifies changing risk patterns, generates possible interventions, communicates uncertainty, and supports coordination across the care team. Clinicians provide context, evaluate recommendations, make judgments, and retain responsibility. Outcomes are returned to the system so that performance can be monitored and improved.
The relevant unit of performance is no longer the clinician working alone or the model operating alone.
It is the human-AI cognitive system.
This distinction is critical.
AI-native architecture does not mean replacing humans with AI. AI remains one component of a larger distributed system that also includes people, data, representations, workflows, professional norms, institutional values, and governance.
The objective is not to maximize the intelligence of one component. It is to improve the performance of the system as a whole.
At this level, organizations stop optimizing individual workers in isolation. They begin optimizing the interactions through which intelligence emerges.
They ask different questions:
How well does the system frame problems?
How effectively does it combine human judgment with machine generation?
How quickly does it learn from outcomes?
How reliably can participants question or override AI-generated recommendations?
How well does the system maintain alignment with institutional goals and values?
At Level Three, distributed cognition becomes operational.
The organization is no longer simply using AI. It is beginning to organize cognition around human-AI collaboration.
Level Four: AI as a Cognitive Medium
Level Four represents a deeper paradigm shift.
AI is no longer simply embedded in workflows. It becomes part of the institution’s epistemic infrastructure, meaning the infrastructure through which the organization knows, questions, interprets, and learns.
Knowledge is no longer produced as a fixed report and updated periodically. It becomes dynamic, interactive, and continuously revised.
AI participates in what I call representational work at scale. It can generate hypotheses, reframe problems, simulate scenarios, compare alternatives, synthesize evidence, and refine explanations. These representations become part of the material through which people and institutions reason.
This is the difference between a tool and a medium.
A tool supports a task.
A medium shapes the conditions under which cognition occurs.
The Internet made information widely accessible. AI makes the transformation of knowledge widely accessible. It does not merely retrieve representations. It generates, translates, reorganizes, tests, and recombines them.
At Level Four, a research institution may continuously connect new findings with existing evidence, generate competing hypotheses, identify contradictions, and update research priorities.
A university may create learning environments in which students interact with adaptive systems that generate explanations, challenge assumptions, and adjust learning pathways while faculty design the conditions for intellectual development and evaluate judgment.
A health system may continuously integrate clinical outcomes, research evidence, operational data, and patient experience into a learning architecture that supports both care and discovery.
A company may treat strategy not as an annual planning exercise, but as a continuous process of scenario generation, evaluation, action, and feedback.
In each case, people do not complete their thinking first and then turn to AI for assistance. Thinking unfolds through interaction with AI as one component of a larger distributed cognitive system.
The institution begins to think within the medium.
AI Does Not Become the Whole System
It is important to be precise about what Level Four means.
AI does not become the institution’s intelligence. It remains one part of a distributed cognitive system.
Humans, AI, data, artifacts, workflows, culture, and governance remain interdependent components. No single component contains the intelligence of the whole.
AI may generate possibilities, identify patterns, simulate outcomes, and expand the cognitive reach of the system. But humans continue to provide purpose, values, context, judgment, accountability, and institutional direction.
The danger is not that AI becomes the whole system.
The danger is that its influence grows while humans fail to develop the forms of intelligence needed to guide and govern the system. People may become increasingly willing to accept AI-generated outputs without sufficient reflection, contestation, or responsibility.
Greater AI capability therefore increases the importance of human judgment.
Organizations must cultivate the ability to frame problems, evaluate alternatives, recognize uncertainty, challenge outputs, make ethical distinctions, and assume responsibility for decisions.
The more powerful the cognitive medium becomes, the more important it is for humans to remain active participants in shaping it.
The Paradigm Shift: From Tool Use to Representational Work at Scale
The ascent from Level One to Level Four is not simply a progression from less AI to more AI.
It is a transition from using AI to perform tasks to organizing cognition within an AI-mediated representational environment.
At the lower levels, AI improves work that the organization already knows how to do.
At the higher levels, AI changes how the organization defines problems, constructs knowledge, evaluates possibilities, and coordinates action.
That is why transformation cannot be measured by counting licenses, prompts, models, or pilots. Those indicators may show activity, but they do not reveal whether the cognitive architecture has changed.
A highly active organization can remain at Level Two.
A smaller organization with fewer models may be more mature if it has redesigned core workflows, embedded feedback, clarified human and AI roles, and established governance for distributed cognition.
The difference is not the volume of AI.
It is the organization of intelligence.
Five Questions That Reveal Your Organization’s Level
Leaders can locate their organization by asking five practical questions.
1. Who works directly with AI?
If AI capability remains concentrated in a technical group, the organization is likely at Level One.
If broad groups of employees can use, interpret, and critique AI, the organization may be at Level Two.
If AI participation is designed into institutional workflows, the organization may be approaching Level Three.
2. Where does AI sit in the workflow?
If AI is opened in a separate application after the real work has already been defined, it remains a tool.
If the workflow itself has been redesigned around human-AI interaction, the organization is moving toward AI-native architecture.
3. What is the unit of performance?
If performance is evaluated primarily as individual production, the cognitive architecture remains traditional.
If the organization measures the quality of human-AI system performance, including coordination, judgment, learning, and outcomes, it is becoming AI-native.
4. What form does knowledge take?
If knowledge remains static, document-based, and periodically updated, AI has not yet become a medium.
If knowledge is continuously generated, challenged, revised, and integrated into decisions, the institution is approaching Level Four.
5. What would happen if AI disappeared tomorrow?
This may be the clearest diagnostic of all.
If removing AI would inconvenience employees but leave the institution’s core operating model intact, AI remains a layer.
If removing AI would fundamentally disrupt how the organization learns, discovers, decides, and governs, AI has become cognitive infrastructure.
Greater Maturity Requires Stronger Governance
The movement toward AI as a cognitive medium creates significant advantages, but it also increases systemic risk.
When AI is used by one person for one task, an error may remain local.
When AI becomes embedded across workflows, a flawed assumption can propagate throughout the institution. Biases can become embedded in shared systems. Repeated reliance on the same models can homogenize thinking. Generated explanations can acquire authority without adequate scrutiny.
At Level Four, governance is no longer limited to approving models or writing policies.
Governance becomes the design of how the institution knows.
That requires:
Traceability, so participants can understand how outputs were produced.
Continuous validation, so performance is monitored as models, data, and contexts change.
Contestability, so people can question recommendations and pursue alternatives.
Human accountability, so responsibility does not disappear into the system.
Diversity of models and perspectives, so institutional cognition does not become artificially uniform.
Mechanisms for intervention and override, so humans retain meaningful control.
Governance must operate inside the cognitive system rather than outside it.
A committee that reviews AI periodically cannot govern a system that generates and influences decisions continuously. Oversight must become embedded, adaptive, and connected to operational reality.
How Leaders Can Cross the Architectural Hurdle
Organizations will not move beyond Level Two through more training alone. They need a deliberate program of cognitive redesign.
First, select a core institutional workflow rather than another isolated use case. Choose a process that matters to learning, discovery, decision-making, or governance.
Second, map where cognition currently occurs. Identify who frames the problem, who generates options, which representations carry knowledge, where judgment is exercised, and how outcomes are incorporated.
Third, redesign human and AI roles explicitly. Do not allow role allocation to emerge through convenience or enthusiasm. Define where AI generates, where humans evaluate, where joint iteration occurs, and who remains accountable.
Fourth, build feedback into the workflow. A system cannot become cognitively mature if it does not learn from outcomes.
Fifth, measure system performance rather than tool usage. Adoption metrics tell leaders whether people are using AI. They do not tell leaders whether the institution is thinking better.
The goal is not to move every activity to Level Four. Different tasks may appropriately remain at different levels.
The goal is to understand the differences and make deliberate architectural choices.
The Question Leaders Should Be Asking
Most organizations continue to ask:
Do we have AI?
More advanced organizations ask:
How widely is AI being used?
The more consequential question is:
What kind of cognition are we building?
Organizations that remain at Levels One and Two may achieve meaningful efficiency gains. They may improve individual productivity, expand access to expertise, and automate valuable work.
But they preserve the architecture of the existing institution.
Organizations that cross into Levels Three and Four begin to redesign how thinking happens. They create new forms of coordination, learning, knowledge production, and decision-making.
That is where structural advantage begins.
The future will not belong to the organizations with the largest number of AI tools.
It will belong to those with the best-designed distributed cognitive systems.
Continue Reading
This framework is developed in Chapter 6, “From Tool to Medium,” of my book, The Cognitive Revolution: How AI Is Reorganizing Intelligence, Expertise, and Institutions.
Book: https://www.amazon.com/dp/B0GWMHQSNG
Website: https://www.dcognition.ai
Jiajie Zhang, PhD
Author of The Cognitive Revolution: How AI Is Reorganizing Intelligence, Expertise, and Institutions
Dean and Glassell Family Foundation Distinguished Chair
D. Bradley McWilliams School of Biomedical Informatics
UTHealth Houston
How This Essay Was Created
I originate the ideas, arguments, conceptual framework, conclusions, and the first draft of every essay. I then use AI as a cognitive partner to challenge my thinking, improve organization, strengthen clarity, and refine the writing. The intellectual contributions are my own. This collaborative process reflects the central thesis of The Cognitive Revolution: intelligence increasingly emerges through distributed cognitive systems.
© 2026 Jiajie Zhang. All rights reserved.
Brief quotations with attribution and a link to the original publication are welcome. No part of this article may be reproduced, republished, or distributed in whole or in substantial part without prior written permission from the author.
No approved comments yet. You can begin the discussion.