Visible Agency Design Test for AI Learning Tasks

AI is no longer something schools can treat as outside the learning environment: It is not waiting outside the school gate for permission.

Schools can set expectations for when AI should be used, when it should be limited and when it should be set aside. Those boundaries are important. But the larger reality remains: students now have access to tools that can generate, organise, improve and polish work in ways that were not previously available.

This means learning tasks need to be designed with AI in mind, even when AI is not the point of the task.

A teacher may ask students to write an explanation, develop an argument, interpret a data set, design a product, prepare a presentation or investigate a local issue. None of these has to be an “AI task.” But in an AI-present environment, students may still use AI somewhere in the process.

The work now is to design tasks clearly enough that the learner remains visible.

Visible Agency should be designed before it is expected.

The Visible Agency Design Test is a practical diagnostic for reviewing learning tasks in an AI-present environment. It helps teachers and leaders ask whether the task itself is likely to keep student thinking, judgement, ownership and responsibility visible if AI becomes part of the process.

Research on generative AI in education increasingly recognises both the potential of AI to support learning and the need to protect cognition, agency, ethics and meaningful student participation as these tools become more available in educational contexts (Roe & Perkins, 2024; Favero et al., 2026).

This is not an assessment rubric for completed student work. It is not an AI detection tool. It is not a compliance checklist.

It is a task design test.

The purpose is to look at a proposed learning task and ask a simple question:

If AI becomes part of this task, will the learner still be visible?

That question matters because AI changes the relationship between output and ownership. When students can generate, improve, reorganise or polish work with AI support, the final product can no longer carry the whole burden of evidence. Teachers need to see more than the work students produce. They need to see something of the thinking, judgement, revision and responsibility that shaped it. This is consistent with assessment-for-learning research, which emphasises the importance of eliciting evidence of student understanding and using that evidence to guide learning, feedback and next steps (Black & Wiliam, 1998; Nicol & Macfarlane-Dick, 2006).

Learner agency is the goal. Visible Agency is the evidence.

The Design Test gives schools a way to make that evidence part of the task from the beginning.

What the Visible Agency Design Test does

The Visible Agency Design Test helps teachers and leaders review whether a learning task has been designed so student agency remains visible in an AI-present environment.

It examines the design itself: the instructions, the learning purpose, the final product, the role AI may play, the required process evidence, the decisions students must make and the way students will explain or justify their work.

It does not judge whether a student has completed the work well. It asks whether the proposed task is likely to make student thinking, judgement, ownership, metacognition and responsibility visible.

A task can invite AI and still require deep learning. A task can restrict AI and still leave student thinking unclear. A task can make no mention of AI and still be completed with significant AI support. The deeper question is whether the learner remains present if AI becomes part of the process.

The Design Test makes that question practical.

AI-invited, AI-permitted and AI-possible tasks

In an AI-present environment, teachers are designing for three situations.

In an AI-invited task, AI is deliberately included as part of the learning process. Students may be asked to use AI to generate alternative arguments, compare explanations, test a claim, receive feedback or explore possible designs. The design challenge is to make sure AI supports the learning demand while students still judge, adapt, verify, revise and explain the role AI played.

In an AI-permitted task, AI may be used within clear boundaries, but it is not the centre of the task. Students might be allowed to use AI for brainstorming, feedback, vocabulary support, drafting questions or checking clarity. The design challenge is to make the boundaries clear and ensure students can explain how AI supported their thinking.

In an AI-possible task, AI has not been invited or permitted, but students could still use it to generate, organise, improve or polish the work. This is the situation many teachers are most concerned about. A task may simply ask students to write an explanation, prepare a presentation, analyse a source, solve a problem or create a design. The task may not mention AI at all. But if AI could complete much of the visible work, the teacher needs to ask where the learner’s thinking, judgement and ownership will become visible.

The Visible Agency Design Test applies to all three situations. It does not begin by asking whether AI should be used. It begins by asking whether the learner will remain visible if AI becomes part of the process.

The task does not need to be about AI. It needs to be designed for a world where AI is present.

How to use the Visible Agency Design Test

Use the Visible Agency Design Test before teaching a task.

Start with the task as it currently exists. Review the learning outcome, the instructions students will receive, the product students will produce, and any requirements for process evidence, explanation, reflection or AI disclosure.

Then score the task against the seven dimensions of Visible Agency:

1. thinking visibility;
2. judgement requirement;
3. ownership;
4. metacognition;
5. effort integrity;
6. transparency;
7. appropriate use.

Each dimension is scored from 0 to 2.

0 — Not yet designed in
The task does not currently require this form of Visible Agency if AI is used.

1 — Partly designed in
The task includes some opportunity for this form of Visible Agency, but the evidence may be optional, unclear or weakly connected to the learning purpose.

2 — Clearly designed in
The task deliberately requires evidence of this form of Visible Agency, so the learner remains visible even if AI is used.

The purpose of the score is not to label the task as good or bad. The purpose is to identify where the design is already strong and where it can be strengthened. A low score is useful. It shows where the task needs more deliberate design.

How to interpret the score

The full Design Test has a possible score of 14.

0–5: Strengthen the evidence design
The task may allow AI to carry too much of the visible work. Review the learning demand and identify where student thinking, judgement, ownership or metacognition should become more visible.

6–10: Build from the strongest dimensions
The task includes some opportunities for students to show agency, but several dimensions could be strengthened. Begin with what is already working, then add one or two deliberate design moves in the lowest-scoring areas.

11–14: Visible Agency is clearly designed in
The task gives students clear opportunities to show thinking, judgement, ownership, metacognition and responsible AI use. It is likely to provide stronger evidence of the learner inside the work.

The score should always lead to redesign. A task does not need to score 14 to be useful. Some tasks will emphasise certain dimensions more than others depending on the learning purpose. The aim is to make Visible Agency a conscious design decision rather than an assumption.

A stronger pattern for Visible Agency evidence

A stronger task does more than ask students to reflect at the end.

Reflection has value, but a final reflection can be too easy to produce after the real thinking has disappeared. Visible Agency is stronger when the task creates evidence at several points in the process. This reflects a broader assessment principle: learning evidence is more useful when it is elicited during the learning process and connected to feedback, self-regulation and student action rather than relying only on a completed final product (Black & Wiliam, 1998; Nicol & Macfarlane-Dick, 2006).

A useful pattern is:

1. Before support: students show an initial idea, question, interpretation, plan or position.
2. During support: students compare, verify, annotate, reject, adapt or revise using criteria or evidence.
3. After support: students explain or discuss the decision they now stand behind and connect it to the evidence or reasoning that shaped it.

This pattern does not make the task more complicated. It gives the learner more places to appear.

Students should not only submit the work. They should be able to stand behind it.

That does not require an interrogation or a lengthy oral examination. It may be a short written statement, a peer discussion, a teacher conference, a presentation defence, an annotated decision trail or a brief ownership conversation. The form can vary. The important point is that students can connect the final work to the thinking, evidence, choices and revisions that shaped it.

An ownership conversation is a brief written, spoken or peer-supported moment where students explain the decisions behind their work and connect those decisions to evidence, criteria, feedback or purpose.

This is one of the strongest ways to protect Visible Agency in an AI-present environment.

The Seven Dimensions of the Visible Agency Design Test

The seven dimensions translate the Visible Agency principle into a practical task design check. Each dimension asks whether the task has been designed so student agency remains visible if AI becomes part of the learning process.

Dimension 1: Thinking visibility

Design question:
Does the task require students to make part of their thinking process visible?

Thinking visibility means the task creates evidence of what students think, change, question or decide. It helps teachers see more than the final product.

A task with strong thinking visibility might require students to record an initial idea, refine a question, compare two options, annotate a revision, explain a decision or identify a moment where their understanding changed. The evidence does not need to be long. It needs to reveal something meaningful about the learner’s thinking.

Score 0 — Not yet designed in
The task asks students to submit a final product, but it does not require evidence of thinking, questioning, decision-making or revision.

Score 1 — Partly designed in
The task includes some opportunity for students to show thinking, but the evidence is optional, informal or not clearly connected to the learning purpose.

Score 2 — Clearly designed in
The task requires students to show specific evidence of thinking, such as questions asked, decisions made, comparisons considered, changes made or reasoning used.

Strengthening move:
Add one point in the task where students must show a meaningful decision: what they first thought, what changed, what they questioned, or why they chose one path over another.

Dimension 2: Judgement requirement

Design question:
If AI or another source of support is used, does the task require students to evaluate, compare or verify what they receive?

Judgement requirement means the task asks students to assess the quality, accuracy, relevance, usefulness or appropriateness of the support they use. If AI becomes part of the process, students should have to think about the value of what it provides rather than moving directly from AI output to final product.

AI literacy frameworks commonly emphasise that learners need to understand, use, evaluate and critically reflect on AI systems and their outputs. Generative AI literacy also includes responsible use, awareness of limitations and the ability to interact with AI systems in informed and ethical ways (Long & Magerko, 2020; Zhang & Magerko, 2025).

A task with strong judgement requirement might ask students to compare AI responses against class criteria, verify claims against reliable sources, identify weak reasoning, question assumptions, reject unsuitable suggestions or explain why one response is stronger than another.

Score 0 — Not yet designed in
The task could allow students to use AI output or other support without evaluating it.

Score 1 — Partly designed in
The task suggests that students should check or improve the support they use, but the judgement process is vague or optional.

Score 2 — Clearly designed in
The task requires students to evaluate AI output or other support using criteria, evidence, purpose or audience before deciding how to use it.

Strengthening move:
Ask students to identify one suggestion they accept, one they adapt and one they reject, and explain the reason for each decision.

Dimension 3: Ownership

Design question:
Does the task require students to explain, discuss and stand behind the decisions that shaped the final work?

Ownership means students remain responsible for the purpose, decisions and final meaning of the work. In an AI-present environment, ownership is broader than whether students produced every word or idea themselves. It is shown when students can explain what they accepted, adapted, rejected, revised and finally stood behind.

A task with strong ownership might require students to explain their final position, defend a design choice, justify a revision, describe how evidence shaped the outcome or identify which parts of the final work reflect their own understanding. This connects with research on learner agency and autonomy, where agency involves intentional, informed participation rather than passive receipt of support or output (Ryan & Deci, 2000; Roe & Perkins, 2024; Favero et al., 2026).

Ownership may be shown through a written statement, annotated decision trail, peer discussion, teacher conference, presentation defence or short explanation conversation. The form can vary, but the student should be able to connect the final work to the decisions, evidence and revisions that shaped it.

Score 0 — Not yet designed in
The task does not require students to explain the decisions behind the final work.

Score 1 — Partly designed in
The task asks students to reflect generally on their work, but it does not clearly require them to take responsibility for specific decisions or final meaning.

Score 2 — Clearly designed in
The task requires students to explain, justify or defend the decisions that shaped the final product.

Strengthening move:
Add an ownership moment: a short written statement, peer discussion, teacher conference or presentation defence where students explain one decision they made, the evidence that shaped it and why they stand behind the final work.

Dimension 4: Metacognition

Design question:
Does the task require students to notice and explain how their thinking changed during the process?

Metacognition means students become aware of how their thinking changed. In an AI-present environment, this includes noticing how AI or other support influenced their understanding, questions, decisions or revisions.

A task with strong metacognition might ask students to explain what became clearer, what assumption changed, what question deepened, what they understood differently after feedback, or how their first idea developed into the final version. This aligns with research on self-regulated learning, which emphasises planning, monitoring, evaluating and adjusting one’s learning strategies over time (Zimmerman, 2002; Education Endowment Foundation, 2025).

Score 0 — Not yet designed in
The task does not ask students to notice or explain how their thinking changed.

Score 1 — Partly designed in
The task includes a reflection question, but it is general or disconnected from the learning process.

Score 2 — Clearly designed in
The task requires students to explain how their thinking, understanding, questions, decisions or revisions changed during the task.

Strengthening move:
Ask students to complete one metacognitive prompt: “I first thought…” “My thinking changed when…” or “I now understand…”

Dimension 5: Effort integrity

Design question:
Does the task preserve the intended learning demand if AI is used?

Effort integrity means the task still requires students to do the thinking, reasoning, decision-making or meaning-making that the learning outcome depends on. AI may support the process, but the important cognitive work should remain with the learner.

A task with strong effort integrity makes the learning demand clear. If the goal is argument, students must still reason from evidence. If the goal is design, students must still understand the user need and justify trade-offs. If the goal is mathematical reasoning, students must still interpret, explain and defend the solution. If the goal is scientific understanding, students must still connect evidence to explanation.

This dimension is important because research on AI in education warns that poorly aligned AI use can encourage cognitive offloading, passive acceptance of output and reduced independent reasoning, while human-centred and pedagogically aligned AI use can support agency, reflection and effortful learning (Favero et al., 2026; Roe & Perkins, 2024).

Score 0 — Not yet designed in
AI could complete the main learning demand of the task with little meaningful student thinking required.

Score 1 — Partly designed in
The task includes some student effort, but AI could still carry a substantial part of the intended learning.

Score 2 — Clearly designed in
The task is designed so students must still complete the essential thinking, reasoning or decision-making required by the learning outcome.

Strengthening move:
Identify the core learning demand, then add a requirement that AI cannot complete on behalf of the student: a decision, explanation, justification, comparison, interpretation or defence.

Dimension 6: Transparency

Design question:
Does the task make the use of AI or other significant support visible enough to discuss, assess and improve?

Transparency means students can explain where and how AI or other significant support influenced the task. This should support learning, integrity and conversation, not create unnecessary documentation.

A task with strong transparency might ask students to identify the role AI played, describe the kind of support they requested, explain which output they used or adapted, or briefly note how AI influenced a decision. The goal is thoughtful visibility, not exhaustive logging. Transparency is also consistent with responsible AI literacy, where students need to understand and communicate how they interact with AI systems, what the systems can and cannot do, and how AI support shaped the work (Long & Magerko, 2020; Zhang & Magerko, 2025).

Score 0 — Not yet designed in
The task gives no guidance on how students should make AI use or other significant support visible or explainable.

Score 1 — Partly designed in
The task asks students to disclose support, but the disclosure is vague, technical or disconnected from learning.

Score 2 — Clearly designed in
The task requires students to explain AI use or other support in a way that supports learning, integrity and teacher understanding.

Strengthening move:
Ask students to include a brief support note: “I used AI or another source of support to…” “I checked this by…” “I changed my work because…”

Dimension 7: Appropriate use

Design question:
Does the task help students decide when AI is useful, limited or unnecessary?

Appropriate use means students learn that AI is not automatically the right tool for every part of learning. Sometimes AI can help generate possibilities, test an explanation or provide feedback. At other times, observation, discussion, practice, direct experience, human feedback or independent thinking may be more appropriate.

A task with strong appropriate use might ask students to decide where AI should and should not be used, compare AI support with another source of support, identify a point where human judgement is required, or explain why they chose not to use AI for part of the task. This dimension reflects the broader goal of AI literacy: helping learners use AI critically, ethically and purposefully rather than treating it as a default source of answers (Long & Magerko, 2020; Zhang & Magerko, 2025).

Score 0 — Not yet designed in
The task gives students no opportunity to consider whether AI is appropriate for the learning purpose.

Score 1 — Partly designed in
The task allows students to choose whether to use AI, but does not require them to explain when or why it is useful.

Score 2 — Clearly designed in
The task requires students to make or explain decisions about when AI is useful, limited or unnecessary.

Strengthening move:
Add a decision point: “Where might AI help your thinking, and where should you rely on another source of evidence, feedback or judgement?”

Examples of the Visible Agency Design Test in use

The examples below are not model units. They are design illustrations. The same principle can be adapted for different year levels, subjects and levels of complexity.

Each example begins with an initial task design, identifies the Visible Agency concern in an AI-present environment, and then shows how the task could be strengthened.

Example 1: English / Humanities — AI-possible task

Initial task design
Write a persuasive essay on whether cities should reduce private car use.

Visible Agency concern in an AI-present environment
If students use AI to generate arguments, structure the essay or polish the writing, the final piece may sound fluent while revealing little about how the student developed the position, evaluated evidence or took responsibility for the final claim.

Strengthened task design
Students begin by writing their own provisional claim and selecting two pieces of evidence they believe support it. They then compare their position with two alternative arguments, which may be generated through AI if permitted or drawn from supplied sources.

Students evaluate the arguments against agreed criteria for evidence, relevance and persuasiveness. They annotate one idea they accept, one they adapt and one they reject, then revise their claim or structure in response. Before submitting the final essay, students complete a short written or spoken ownership conversation explaining which decision most improved the argument, what evidence shaped that decision and why they stand behind the final position.

What changed
The strengthened task makes argument development visible across the process. Students show an initial position, evaluate alternatives, revise using criteria and defend a specific decision. AI can support comparison if permitted, but the student’s judgement remains visible.

Example 2: Science — AI-possible task

Initial task design
Create an explanation of how human activity affects a local ecosystem.

Visible Agency concern in an AI-present environment
If students use AI to generate or improve the explanation, the final response may be clear while revealing little about how the student understood the ecosystem, interpreted evidence or revised their explanation.

Strengthened task design
Students begin by drawing or writing their own cause-and-effect explanation of how one human activity affects a local ecosystem, using class notes, observation data or supplied source material. Before using AI or another source of support, they identify the relationship they are least certain about.

Students then compare their explanation with one additional explanation from another source, which may include AI if permitted. They annotate three points of comparison: one point of agreement, one point that extends their thinking and one point they need to verify. They check the uncertain point against class materials or a supplied source, revise their explanation and mark the exact change they made.

Their final submission includes the revised explanation, the annotated comparison and a short ownership moment where they explain what they changed, what evidence supported the change and why their final explanation is stronger.

What changed
The strengthened task creates evidence before, during and after support is used. Students first show their own understanding, then evaluate another explanation, verify a specific point and revise a visible part of their work. If AI is used, it becomes part of a comparison and verification process rather than a substitute for the explanation.

Example 3: Design / Technology — AI-permitted task

Initial task design
Design a product that helps students organise their schoolwork.

Visible Agency concern in an AI-present environment
If students use AI to generate product ideas, the task may produce creative possibilities quickly while revealing little about how the student understood user needs, evaluated trade-offs or chose which idea was worth developing.

Strengthened task design
Students begin by gathering evidence from users through a short interview, observation or survey. They identify one specific organisation challenge and write a design criterion based on that evidence.

Students then generate a range of possible solutions, using AI if permitted as one source of possibilities. They evaluate three ideas against user need, practicality and likely impact. They reject one idea, adapt one idea and select one to prototype. Their prototype submission includes the user evidence, the evaluation notes and a design decision statement explaining how the final idea responds to the need they identified.

Students then complete a short design defence in which they explain which user need shaped the final design, what trade-off they made and why this solution is worth developing.

What changed
The strengthened task anchors ownership in user evidence. AI can expand the range of possibilities, but students remain responsible for interpreting the need, weighing trade-offs and justifying the design decision.

Example 4: Mathematics / Data — AI-possible task

Initial task design
Explain the trend shown in a data set about household energy use.

Visible Agency concern in an AI-present environment
If students use AI to interpret the data or generate an explanation, the final response may sound plausible while revealing little about how the student read the data, noticed patterns, tested an interpretation or justified the conclusion.

Strengthened task design
Students first annotate the data set by identifying the main trend, one anomaly and one possible explanation. They write a preliminary conclusion using specific values from the data.

They then compare their conclusion with an alternative interpretation from another source, which may include AI if permitted. Students identify where the alternative interpretation is supported by the data, where it overreaches and what evidence should carry the most weight. They revise their conclusion and include one rejected interpretation with an explanation of why the data does not support it.

Their final submission includes a brief ownership conversation, written or spoken, in which they explain which data point or pattern carries the most weight in their conclusion and which interpretation they rejected.

What changed
The strengthened task makes data reasoning visible. Students show how they read the data before considering an alternative interpretation, then use evidence to revise or defend their conclusion. AI may support comparison, but the mathematical reasoning remains anchored in the data.

Example 5: Inquiry / Research — AI-invited task

Initial task design
Investigate an issue affecting young people in your community and present your findings.

Visible Agency concern in an AI-present environment
If students use AI to generate research questions, summarise information or structure the presentation, the final product may appear coherent while revealing little about the student’s curiosity, source judgement, interpretation or responsibility for the findings.

Strengthened task design
Students begin by developing three possible inquiry questions and explaining which one they believe is most worth investigating. They may use AI to generate alternative questions or suggest possible lines of inquiry, but they must evaluate these against agreed criteria: relevance, researchability, significance and connection to the local context.

Students select one question, gather evidence from at least two non-AI sources and use AI only to help compare possible interpretations or improve clarity if permitted. Their final presentation includes the findings, a source judgement note, and a short explanation of how their question changed, which sources shaped their thinking and what they decided not to include.

Students finish with an ownership conversation in which they explain the inquiry decision they stand behind: the question they chose, the evidence they trusted most, or the interpretation they believe is strongest.

What changed
The strengthened task deliberately invites AI as one form of support, but the student remains responsible for the inquiry question, evidence, interpretation and final message. AI supports exploration while the learner’s judgement remains visible.

How to strengthen a task after using the test

The Visible Agency Design Test is most useful when it leads to a better task.

After scoring a task, look first at the lowest-scoring dimensions. A task does not need to be redesigned completely. Often, one or two deliberate design moves can make student agency much more visible.

If thinking visibility is weak, add a point where students show an initial idea, question, comparison or decision. If judgement is weak, ask students to evaluate AI output or another source of support against criteria or evidence. If ownership is weak, require students to explain what they accepted, adapted, rejected or stood behind. If metacognition is weak, ask students to describe how their thinking changed. If effort integrity is weak, identify the core learning demand and make sure students still do that work. If transparency is weak, ask students to briefly explain how AI or another source of support was used. If appropriate use is weak, ask students to decide where AI is useful and where another source of support is better.

The strongest redesign moves are usually small and precise.

Visible Agency does not require more complicated tasks. It requires clearer evidence of the learner inside the task.

Frequently asked questions

Where to go next

The Visible Agency Design Test is part of the Visible Agency series.

• For the broader framework, read Visible Agency: How to Design AI-Supported Learning Without Outsourcing Student Thinking.

• For the visible process side of the work, read How to Make Student Thinking Visible When AI Is Part of the Process.

• For the judgement side of the work, read How to Design AI Tasks That Require Student Judgement.

• For the ownership side of the work, read How to Design AI-Rich Tasks That Still Require Student Ownership.

• For the learning demand side of the work, read How to Stop AI From Replacing Student Thinking.

• For the metacognition side of the work, read How to Use AI to Strengthen Metacognition.

• For the broader ownership argument, read Visible Agency: What Student Ownership Means When AI Can Do the Work.

The purpose of the Design Test is simple: make it much harder for the learner to disappear from the work.