CAC Co-intelligence:
Learning with AI, Not from it
AI Augmentation Roles: A Practical Guide to Purposeful Human–AI Collaboration
Beyond Simply Using AI
Artificial intelligence is changing what we can accomplish, but access to increasingly capable AI does not automatically make us more knowledgeable, creative, or effective.
Two people can use the same AI system for the same task and have very different experiences. One might receive an immediate answer and accept it. Another might use AI to explore alternatives, challenge assumptions, develop new skills, or extend their creative capabilities. The difference is not necessarily the AI model being used. It is the role AI has been asked to perform and how the human engages with it.
AI augmentation refers to using artificial intelligence to support, extend, and enhance human capabilities rather than unnecessarily replacing human thinking, judgement, and decision-making. Within CAC Co-intelligence it sits in the second stage (the AI stage).
Understanding the different roles AI can perform allows us to move beyond routine prompting towards more purposeful human–AI collaboration. This is relevant to students developing knowledge and skills, educators designing learning experiences, and professionals seeking to work more effectively within AI-enabled workplaces.

Why AI Augmentation Roles Matter
Most people are familiar with asking AI questions or requesting completed outputs. However, these interactions represent only some of the ways AI can contribute. AI can help us learn, evaluate, create, investigate, organise, and make decisions. But these different activities require different forms of human engagement. For example, asking AI to explain a difficult concept may help someone access information. Asking it to question their understanding may require them to actively reason through the concept. Asking it to challenge a proposed solution may expose weaknesses they had not considered.
Each approach changes the cognitive work being undertaken by the human and AI. This matters for three reasons.
1. Different roles develop different capabilities
If the goal is to develop reasoning, a Socratic interaction may be valuable. If the goal is to evaluate an existing argument, Adversarial AI may be more appropriate. If the goal is to develop an unfamiliar skill, AI Tutor may offer better support. The role should follow the capability we want to develop, not simply what the technology makes convenient.
2. The same AI support can help or hinder
Delegating data organisation to AI might allow a professional to concentrate on a complex decision. But delegating all the problem-solving to AI could undermine a student's opportunity to develop that very skill. Similarly, detailed explanations can support understanding, yet excessive explanation and guidance may reduce opportunities for independent practice.
Effective augmentation requires deciding which human activities should be supported, which can be delegated, and which must remain independently practised.
3. Different roles create different responsibilities
Using AI as a tutor requires checking the accuracy and suitability of explanations. Using AI as a creative partner requires actively directing and evaluating the emerging work. Using embedded AI agents requires understanding their permissions, information sources, limitations, and ability to take action.
As AI capabilities expand, understanding these responsibilities becomes increasingly important.
Understanding the Six Augmentation Roles
The six roles below provide a practical way to explore different forms of human–AI interaction. They are not mutually exclusive, nor do they represent a rigid progression from simple to advanced AI use. Depending on the activity, several roles may be appropriate at different moments.
1. Centaur: Strategic Division of Labour
Primary purpose: Help determine and manage how work should be distributed between humans and AI.
Centaur is based on the deliberate division of responsibilities between human and AI contributions. Its defining feature is not simply delegating tasks. It is exercising judgement about what should be delegated, what should remain human-led, and how the contributions should be coordinated.
Why use Centaur? A complex task often contains activities with different requirements. Some are repetitive, computational, or organisational. Others depend heavily on contextual understanding, professional expertise, creativity, ethical considerations, or accountability. Centaur encourages the person to examine these differences rather than automatically assigning everything to AI.
How does it work? Consider a consultant investigating service delivery delays. Before involving AI, the consultant identifies the activities required and develops an initial allocation plan. AI can then help examine that proposed division, highlight possible risks, and suggest alternative allocations. The consultant decides which tasks to retain, delegate, or share. Once the allocation is established, AI completes appropriate components while the consultant manages, verifies, and integrates the contributions.
What capability does this develop? Strategic task allocation, informed delegation, coordination, and evaluation of human–AI work.
When might it be unsuitable? When delegation removes opportunities to practise essential knowledge or skills, or when AI cannot reliably perform a delegated activity. A student learning statistical analysis, for example, should not automatically delegate all statistical reasoning to AI simply because the technology can produce a result.
Key question: Am I making a deliberate division of labour, or simply handing over work because AI can complete it?
2. Socratic AI: Deepening Understanding Through Questioning
Primary purpose: Develop reasoning and understanding through purposeful, responsive questioning.
Socratic AI differs from conventional answer-focused interactions. Rather than immediately supplying an explanation or solution, AI asks questions designed to encourage a person to articulate their understanding, examine assumptions, justify reasoning, and recognise knowledge gaps.
Why use Socratic AI? Reading an explanation can produce a sense of familiarity without necessarily creating deeper understanding. Socratic questioning encourages the learner to actively construct and examine their own reasoning. It may also reveal misconceptions that are not immediately apparent when someone simply reads a polished explanation.
How does it work? Consider a student who has studied correlation and causation but remains uncertain about the distinction. After independently reviewing the concepts, the student asks AI to question their understanding one step at a time. AI might ask whether two variables moving together necessarily indicates a causal relationship. The student explains their reasoning. AI then responds with a further question that targets a potential assumption or gap. The interaction continues until the learner can provide a more defensible explanation.
What capability does this develop? Reasoning, metacognition, self-explanation, and awareness of one's understanding.
When might it be unsuitable? When a learner has insufficient foundational knowledge to engage meaningfully with the questions. In such cases, explanation or direct instruction may be necessary first. Socratic AI should not become endless questioning without progress. Well-designed questioning must be purposeful, responsive, and appropriately challenging.
Key question: Am I developing and explaining my understanding, or merely responding to questions without meaningful reflection?
3. Adversarial AI: Strengthening Judgement Through Challenge
Primary purpose: Test arguments, assumptions, evidence, and proposed decisions through constructive opposition.
Adversarial AI acts as a critical challenger rather than an agreeable assistant. Its purpose is not to disagree automatically, but to identify credible weaknesses, competing explanations, and potential consequences that warrant consideration.
Why use Adversarial AI? When we develop an idea or recommendation, we can become invested in our original position. We may overlook alternative explanations, underestimate risks, or interpret evidence in ways that reinforce our assumptions. Adversarial AI can help expose these vulnerabilities before a decision is finalised.
How does it work? Imagine a professional who has recommended redistributing staff responsibilities to address service delays. After independently developing the recommendation, they ask AI to act as a sceptical senior manager. AI challenges the available evidence, questions whether task allocation is actually responsible for the delays, and identifies possible consequences of redistributing staff. The professional must respond by defending their reasoning, gathering additional evidence, or revising their position. Importantly, AI should acknowledge a convincing defence rather than manufacture disagreement.
What capability does this develop? Critical thinking, evaluative judgement, argumentation, and evidence-based decision-making.
When might it be unsuitable? When learners have not yet established sufficient understanding to defend a position, or when poorly calibrated criticism creates confusion rather than productive challenge. AI-generated counterarguments also require scrutiny. An AI system can produce a persuasive criticism that is nevertheless inaccurate.
Key question: Am I testing my position against credible challenges, or treating AI's objections as automatically valid?
4. AI Tutor: Developing Knowledge and Skills
Primary purpose: Provide responsive instruction, explanation, practice, scaffolding, and feedback that support progressive learning.
An AI Tutor can combine several instructional approaches depending on what the learner needs. This makes it different from a narrowly defined questioning or adversarial role.
Why use an AI Tutor? People do not always need the same type or amount of support. One learner may benefit from a worked example. Another may need feedback on a misconception. A more advanced learner may need increasingly difficult practice activities. An AI Tutor can adapt its explanations, questioning, and guidance in response to the learner's demonstrated understanding, although the accuracy of that adaptation must not be assumed.
How does it work? Consider a university student struggling to interpret p-values in statistical hypothesis testing. The student has already studied the topic and attempted some problems independently. They bring their difficulties to AI, asking it to establish their current understanding before providing targeted explanations and guided practice. AI may demonstrate a concept, ask questions, provide hints, correct misconceptions, and gradually reduce assistance as the learner progresses. Ultimately, the learner attempts a new problem without AI support.
What capability does this develop? Knowledge acquisition, skill development, applied understanding, and increasing learning independence.
When might it be unsuitable? When AI provides excessive scaffolding, when explanations cannot be reliably verified, or when assistance prevents students from demonstrating independent competence. A tutoring interaction should gradually reduce dependence rather than create a permanent need for AI guidance.
Key question: Am I becoming more capable of applying this knowledge independently?
5. Co-Creation: Expanding What We Can Develop Together
Primary purpose: Collaboratively generate, develop, combine, and refine ideas, designs, solutions, or other outcomes.
Co-Creation differs from simply asking AI to produce a finished result. The human and AI engage in an iterative process where human intention, expertise, and judgement continue to shape the emerging outcome.
Why use Co-Creation? AI can make certain technical or creative capabilities more accessible. Someone may have a clear vision for a design but lack graphic design skills. A professional may understand a complex problem but need assistance developing prototypes or exploring alternative solutions. Co-Creation allows human direction to combine with AI capabilities in ways that expand what can be produced or explored.
How does it work? Imagine wanting to create a personalised Zoom background without graphic design experience. Before using AI, you establish your initial creative vision: perhaps a modern academic workspace with bookshelves, subtle technology elements, and a welcoming atmosphere. You then bring that concept to AI and collaborate on variations. After selecting promising ideas, AI generates a preliminary design. You evaluate its appearance, request adjustments, refine colours and layout, and progressively guide the outcome. The final product reflects a continuing exchange between human decisions and AI contributions.
What capability does this develop? Creative direction, synthesis, iterative problem-solving, communication of ideas, and evaluation of emerging solutions. Co-Creation extends beyond visual design. It can also support product development, business innovation, educational resource design, process improvement, and collaborative problem-solving.
When might it be unsuitable? When independent creative performance is the explicit objective, when sensitive or proprietary content cannot be shared, or when premature AI suggestions constrain original thinking. An important distinction is that producing a sophisticated design with AI does not necessarily mean the person has acquired the technical skills to recreate it independently.
Key question: Am I actively shaping the evolving outcome, or simply choosing between things AI has produced?
6. AI Environment: Working Within AI-Enabled Systems
Primary purpose: Develop the capability to work effectively within tools and workflows where AI is integrated into ongoing activities.
AI Environment differs somewhat from the previous roles because it describes how AI capabilities are incorporated into the working or learning environment. These capabilities might provide contextual assistance or operate through agents that monitor information and perform defined activities.
Why use AI Environment? AI is increasingly becoming part of the systems people already use. Rather than accessing AI only through a separate chatbot, people may encounter AI assistance while managing projects, preparing documents, analysing information, or completing learning activities. Understanding these environments requires more than prompting skills. Users need to know what AI systems can access, what they are doing, what triggers their actions, and when intervention is necessary.
How does it work? Imagine a project manager using software with an embedded AI monitoring agent. The manager establishes which project information the agent can access, which activities it should monitor, and which actions require approval. The agent tracks project progress, identifies emerging delays, examines affected dependencies, and presents relevant findings. The manager independently considers the situation, evaluates the AI's evidence and recommendations, and decides on the appropriate response.
In education, a similar approach might involve learning platforms where AI agents monitor practice performance, identify recurring difficulties, and recommend relevant activities.
What capability does this develop? Adaptability, informed AI use, contextual oversight, critical evaluation, and the ability to work responsibly within AI-enabled workflows.
When might it be unsuitable? When system behaviour is opaque, information permissions are inappropriate, outputs cannot be meaningfully evaluated, or autonomous actions create unacceptable risks. Not every AI-enabled environment requires autonomous agents, and not every embedded AI feature should be granted permission to act independently.
Key question: Do I understand what this AI is doing, can I evaluate its contributions, and could I still make an informed decision without it?
Choosing an Augmentation Role: Start With the Capability You Need
There is no universally best augmentation role.
The most appropriate approach depends on the purpose of the task, the capabilities of the person using AI, the risks involved, and which responsibilities should remain human-led.
​
Consider: If your main purpose is to…
Centaur: Decide what humans and AI should each do
Socratic: Explore and deepen your understanding
Adversarial: Test an argument or recommendation
AI Tutor: Learn a concept or develop a practical skill
Co-Creation: Develop and refine an idea or solution collaboratively
AI Environment: Work effectively with AI embedded into tools and workflows
​
​This list is a starting point rather than a rigid classification.
A professional might initially use Socratic AI to clarify their understanding of a problem, then Adversarial AI to challenge a proposed solution, followed by Centaur to determine how implementation work should be distributed. Likewise, an AI Tutor might shift between explanation, questioning, practice, and constructive challenge as the learner progresses.
The important skill is recognising which type of support is appropriate, why it is appropriate, and when the role should change.
Moving Beyond Basic Prompts
The role descriptions provide a useful starting point, but selecting a role is not sufficient by itself. A prompt such as "Act as my tutor" gives AI a broad instruction, but it does not necessarily establish the learner's current understanding, the intended outcome, the desired level of support, or how progress should be assessed.
More effective AI engagement usually involves specifying six elements:
Context: What is the task, situation, or subject?
Existing human contribution: What have you already considered, developed, or attempted?
Purpose: What do you want to improve, learn, investigate, or achieve?
AI role: What specific form of assistance should AI provide?
Boundaries: What should AI avoid doing, and what decisions remain yours?
Interaction and evaluation: How should AI respond, adapt, and help you assess progress?
A useful general prompting structure is:
"I am working on [task or problem]. I have already [my initial thinking or work]. My objective is [intended capability or outcome].
Act in the role of [augmentation role] and support me by [specific activity].
Please [interaction expectations]. Do not [boundaries or prohibited assistance].
Help me identify what I need to investigate, practise, or evaluate independently."
This structure should be adapted to the specific activity rather than applied mechanically.
For AI environments and agents, prompts may also need to establish authorised information sources, task triggers, available tools, approval requirements, and stopping conditions.
A more detailed prompt is valuable when it creates a more purposeful interaction, not simply because it contains more words.
How Do We Know Augmentation Is Actually Working?
This is perhaps the most important question.
A successful AI-generated output does not necessarily demonstrate successful augmentation. Someone may complete a difficult task using AI while developing little additional understanding or capability. Alternatively, an AI interaction may produce an imperfect immediate result while still contributing meaningfully to learning or improved judgement.
It is therefore useful to distinguish between three forms of value.
Task performance: Did AI help improve the efficiency, quality, scope, or feasibility of the work?
Human engagement: Did the person remain actively involved in the relevant thinking, evaluation, and decision-making?
Capability development: Did the interaction contribute to the person's knowledge, skills, judgement, creativity, or ability to undertake similar activities in the future?
The relative importance of these three dimensions depends on the situation. In education, capability development is often the primary objective. In professional practice, immediate task performance may be especially important, but it should not come at the expense of essential expertise, safe decision-making, or accountability.
A practical way to evaluate augmentation is to ask:
-
What did I contribute before AI became involved?
-
What did AI contribute that was genuinely useful?
-
What did I question, verify, reject, or change?
-
What can I now understand, explain, decide, or achieve?
-
Which parts of the task could I complete without AI?
-
Would I use the same augmentation approach next time?
These questions encourage reflection on the quality of human–AI interaction rather than merely celebrating task completion.
The Risks of Choosing the Wrong Role
Even when AI produces apparently useful results, an unsuitable augmentation approach can undermine the intended purpose. For example, choosing Centaur may be inappropriate if it encourages students to delegate essential practice. Socratic questioning can frustrate learners when explanations are required. Adversarial challenges can amplify confusion when counterarguments are unsupported. AI tutoring can create dependency through excessive assistance. Co-Creation may lead users to accept AI-generated ideas too early, while embedded AI agents can encourage passive reliance if users do not understand their operation.
These are not reasons to reject the approaches. They are reasons to use them deliberately. The objective is not to maximise AI involvement. It is to establish the most appropriate relationship between AI assistance and human responsibility. In some situations, the best choice may be to use less AI or no AI at all.
From AI Users to Capable Human–AI Collaborators
As AI becomes more capable and increasingly embedded in education and professional practice, people will need more than the ability to generate effective prompts. They will need to understand when AI can add value, which forms of assistance are appropriate, how AI's limitations influence its use, and which responsibilities should remain human-led.
The six augmentation roles offer a practical vocabulary for making these distinctions. CAC Co-intelligence provides an overarching cognitive process that reinforces independent thinking before, during, and after AI involvement.
Together, they support a more deliberate approach to human–AI interaction.
The real goal of AI augmentation is not simply to achieve outcomes that would otherwise be difficult. It is to use AI in ways that extend what humans can accomplish while developing and preserving the knowledge, skills, creativity, and judgement that make those accomplishments meaningful.