Visible Agency: How to Design AI-Supported Learning Without Outsourcing Student Thinking

AI has changed what student work can tell us.

For many years, teachers have been able to look at a finished piece of work and make a reasonable judgement about the thinking behind it. A written response, presentation, project, research task or design solution usually carried traces of the student’s understanding. The quality of the product did not tell the whole story, but it often gave teachers enough evidence to begin a useful learning conversation.

That evidence is now less reliable.

In an AI-rich learning environment, students can produce more polished work with less visible struggle. They can generate fluent explanations, cleaner summaries, stronger introductions, better images, improved code, more organised slides and more sophisticated language than they may be ready to produce independently. In many cases, this support can be valuable. AI can help students explore possibilities, clarify confusion, test ideas, improve expression and receive feedback that would otherwise be unavailable in the moment.

The useful question is not simply whether AI helped. The useful question is what the student still had to think through.

Visible Agency is a learning design principle that makes student thinking, judgement and ownership observable when AI is part of the learning process. It helps teachers design AI-supported tasks where students must show what they questioned, compared, revised, accepted, rejected and ultimately took responsibility for.

Learner agency remains the broader educational goal. We still want students to become more capable, intentional, reflective and responsible learners. We still want them to make choices, set directions, monitor progress, seek feedback, revise their thinking and take ownership of their learning. What has changed is the kind of evidence teachers need in order to trust that agency is actually present.

In an AI-rich classroom, learner agency can no longer be assumed from the final product alone. It has to become visible through the decisions, judgements, questions, revisions, comparisons and ownership students demonstrate throughout the learning process.

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

The future of learner agency is not decided by whether students use AI. It is decided by whether their thinking remains visible, responsible and judgeable when AI is part of the process.

This concern is also emerging in research on generative AI and education. A scoping review by Roe and Perkins (2024) found that generative AI may enhance learner agency through personalisation and support, while also raising risks around learner autonomy, equitable access and the changing nature of agency itself.

Why does AI make learner agency harder to judge?

AI can make student work look better very quickly. That is one of its strengths. It can help students organise ideas, improve phrasing, generate examples, compare options, identify gaps and refine a final product. Used well, this can make learning more accessible, more ambitious and more responsive.

However, better-looking work is not always deeper learning. A student may submit a clear explanation without being able to explain the reasoning behind it. They may produce a polished essay without having made the key decisions. They may present a sophisticated argument without having weighed the alternatives. They may use accurate vocabulary without being able to apply the concept with confidence.

This is one reason AI-supported learning should not be judged only by the quality or efficiency of the final output. Ding and Magerko (2025) argue that many educational AI evaluations focus heavily on technical performance, accuracy or task efficiency while overlooking learner agency, contextual learning processes and ethical considerations.

This does not mean AI support is the issue. Support has always been part of learning. Teachers support. Peers support. Worked examples support. Graphic organisers support. Feedback supports. Scaffolds support. The more useful distinction is whether the support leaves the learner more capable.

If AI completes the cognitive work for the student, the final product may improve while the learner’s capacity remains unchanged. If AI strengthens the student’s ability to question, compare, decide, revise and explain, then the tool has supported agency rather than replaced it.

That difference is now one of the central design challenges for teachers.

The question is no longer, “Did AI help?” The better question is, “What did the student still have to notice, decide, test, revise and own?”

What is Visible Agency?

Visible Agency is the observable evidence of student thinking, judgement and ownership during AI-supported learning.

It appears when students can show the decisions they made, the questions they asked, the prompts they refined, the options they compared, the feedback they considered, the revisions they made and the responsibility they took for the final result.

It is not a compliance process. It is not a decorative reflection added at the end. It is not a worksheet asking students to prove they did the work after the real learning has already finished.

Visible Agency is a learning design principle. It asks teachers to design tasks where the student’s thinking remains present throughout the process.

A task with Visible Agency may ask students to explain why they chose one AI suggestion over another. It may ask them to compare an AI-generated explanation with their own understanding. It may ask them to identify where AI was helpful, where it was limited and what they had to change. It may ask them to document the shift between their first idea, the AI-supported feedback and their final decision.

The purpose is not to make AI use harder. The purpose is to make learning clearer.

AI should not make student thinking disappear. It should give students more opportunities to examine, strengthen and explain their thinking.

This is where the design challenge becomes powerful. AI can be used to bypass thinking, but it can also be used to reveal thinking. It depends on the task.

If the task only asks for a final answer, AI may do too much of the work invisibly. If the task asks students to compare, justify, revise and reflect, AI can become part of a richer learning process. Visible Agency is therefore not an anti-AI position. It is a pro-learning position.

This aligns with a broader shift towards human-centred AI in education. Viberg et al. (2026) identify human oversight, AI-human complementarity, AI competencies and relational emergence as key aspects of human agency in AI education, while also highlighting transparency and cognitive offloading as practical dilemmas.

For a deeper look at this first design challenge, see How to Make Student Thinking Visible When AI Is Part of the Process.

How does Visible Agency protect learner agency?

Learner agency is often described through language such as voice, choice, ownership, self-direction and responsibility. These ideas remain important, but they are sometimes easier to name than to see.

A student choosing a topic may show agency, but choice alone is not agency. A student selecting a format may show agency, but format choice alone is not agency. A student using AI to produce a final response may show agency, but only if the learning design requires the student to remain intellectually active.

Agency is not simply freedom. It is the capacity to act with intention.

In learning, that means students are not only doing the task. They are making sense of the task. They are interpreting success. They are choosing strategies. They are monitoring the quality of their thinking. They are making decisions when there is more than one possible path. They are responding to feedback. They are taking responsibility for the relationship between their process and their outcome.

AI makes this both more difficult and more important. It becomes more difficult because AI can hide the student’s thinking behind a fluent product. It becomes more important because students now need to learn how to work with powerful tools without surrendering their judgement to them.

This is why Visible Agency is needed.

Visible Agency does not replace learner agency. It makes learner agency observable. It asks teachers to design learning experiences where students cannot simply submit what AI helped them produce. They must show how they thought, what they questioned, what they changed, what they accepted, what they rejected and why the final work represents their own learning.

What did the student still have to think through?

When AI is part of the process, teachers need a sharper way to evaluate the learning demand of a task. The most useful question is: what does this task still require the student to think through?

This question changes the conversation. Instead of asking only whether AI was allowed, we ask what kind of thinking remained necessary. Instead of focusing only on whether the final work is original, we ask whether the learning process required judgement, comparison, revision and ownership. Instead of treating AI as either a shortcut or a threat, we examine how it functions inside the task design.

A student might use AI to generate three possible explanations of a scientific concept. That could become thin learning if they simply copy the best one. It could become strong learning if they compare the explanations, identify misconceptions, connect the strongest version to evidence, rewrite it in their own words and explain why their final explanation is more accurate.

The difference is not the tool. The difference is the thinking the task requires.

A student might use AI to help draft a persuasive paragraph. That could reduce learning if the paragraph is accepted without interrogation. It could deepen learning if the student has to identify the claim, evaluate the evidence, strengthen the reasoning, revise the tone and explain the effect of each change.

Again, the difference is not whether AI appears. The difference is whether the student’s agency remains visible.

The strongest AI-supported tasks do not ask students to pretend they worked alone. They ask students to show how they worked with support while remaining responsible for the thinking. For teachers designing tasks where judgement is the central demand, How to Design AI Tasks That Require Student Judgement develops this part of the framework more fully.

What is the Visible Agency Design Test?

Visible Agency becomes practical through a simple design question: does this AI-supported task still require the student to think, judge, revise, reflect and take responsibility?

The Visible Agency Design Test translates that question into a practical review tool for teachers and leaders. It helps teams examine whether an AI-rich task preserves student agency or quietly allows AI to do too much of the learning work.

The test looks at seven areas.

First, it asks whether student thinking is visible. Can the teacher see what the student questioned, changed, noticed or decided?

Second, it asks whether the task requires judgement. Does the student have to evaluate AI output rather than simply accept it?

Third, it asks whether ownership remains with the student. Can the student explain and stand behind the final meaning of the work?

Fourth, it asks whether the task requires metacognition. Does the student reflect on how AI affected their thinking, confidence, strategy or understanding?

Fifth, it asks whether the task protects effort integrity. Does AI support meaningful effort rather than replace it?

Sixth, it asks whether AI use is transparent. Can the student explain where, how and why AI was used?

Finally, it asks whether the task develops appropriate use. Does the student learn when AI is helpful, when it is limited and when it is not the right tool?

These questions do not turn AI use into a compliance exercise. They help teachers design learning where AI can support students without making their thinking disappear. This is consistent with broader calls for educational AI evaluation to include explainability, ethical responsibility, accessibility, workflow and refinement, not only technical performance (Ding & Magerko, 2025).

The full Visible Agency Design Test provides a practical checklist for reviewing AI-supported tasks before they are used with students.

What changes in AI-supported learning design?

Visible Agency changes the way we design tasks. It moves the focus from product completion to process evidence. It asks teachers to design learning experiences where the pathway matters, not only the outcome. It encourages students to make their reasoning visible before, during and after AI support.

This does not mean every task needs to become longer. It means the right parts of the task need to become more visible.

Instead of asking students to “write an explanation of climate change,” a teacher might ask students to draft their initial explanation, use AI to identify possible gaps or misconceptions, compare the AI feedback with class learning, revise the explanation and then annotate three changes they made. The final explanation still matters, but now the teacher can see more of the learning.

Instead of asking students to “create a persuasive speech,” a teacher might ask students to generate three possible openings, evaluate which one best suits the intended audience, explain the choice, draft the speech, use AI to test clarity and then revise the argument. Again, the product still matters, but the student’s judgement becomes visible.

Instead of asking students to “research a historical event,” a teacher might ask students to compare an AI-generated summary with approved sources, identify what was missing or oversimplified, create a more accurate explanation and reflect on how their understanding changed.

The design shift is subtle but significant. The teacher is no longer only asking, “What did you produce?” The teacher is also asking, “What did you notice? What did you decide? What changed? What do you now understand? What are you responsible for?”

These are agency questions.

How can AI strengthen student agency?

AI can weaken agency when it removes the need for students to think. It can strengthen agency when it gives students better conditions for thinking.

AI can help students see options they may not have generated alone. It can provide immediate feedback when the teacher is working with other students. It can help students rehearse explanations, test assumptions, improve clarity, access vocabulary, explore perspectives or identify gaps.

These uses can support agency because they give students more material to think with. The important design move is to keep the student in the position of judgement.

AI may suggest, but the student decides. AI may generate, but the student evaluates. AI may explain, but the student connects. AI may challenge, but the student revises. AI may support, but the student owns.

AI becomes educationally useful when it gives the learner something better to think with, not when it removes the need to think.

When students use AI well, they do not simply get to the answer faster. They become more aware of the choices involved in producing a strong answer. They learn to question the first version. They learn to compare alternatives. They learn that fluency is not the same as accuracy. They learn that feedback is useful only when acted on thoughtfully.

This is where AI can become a powerful agency tool. It can make thinking more visible by giving students something to respond to: a suggestion, misconception, draft, counterargument, different explanation, set of criteria or possible improvement.

The student’s agency appears in the response.

This is also where metacognition becomes essential. Students need to notice not only what AI produced, but how AI affected their own thinking. The Education Endowment Foundation (2025) identifies metacognition and self-regulation as important to effective pupil learning, especially when students are taught to plan, monitor and evaluate their learning within curriculum and subject learning.

How to Use AI to Strengthen Metacognition explores this reflective side of the framework in more detail.

Why can AI make student agency harder to fake?

At first, it may seem that AI makes agency easier to fake. A student can submit a polished product that hides how little thinking they did. That is true when the task only values the product.

Well-designed AI-rich tasks can make agency harder to fake because they require students to show the thinking behind the work. They must explain decisions, compare alternatives, revise intentionally and account for their use of support.

A student can copy an answer. It is harder to fake a thoughtful comparison between versions. A student can submit an AI-generated paragraph. It is harder to fake an explanation of why specific revisions improved the argument. A student can use AI to summarise a topic. It is harder to fake a conversation about what the summary missed, where it was misleading and how their understanding changed.

Visible Agency gives teachers better evidence. It also gives students a clearer message. The purpose of the task is not to hide AI use or perform independence. The purpose is to learn, think, decide and take responsibility.

This is a healthier direction for classrooms. Instead of turning AI into a policing issue, teachers can turn it into a design issue. The central concern becomes the quality of the learning process, not only the detection of unauthorised assistance.

Where the practical concern is whether AI is replacing too much effort, How to Stop AI From Replacing Student Thinking provides a more focused exploration of task design, effort integrity and cognitive demand.

What should teachers look for?

When teachers design for Visible Agency, they begin to look for different kinds of evidence.

They look for the student’s starting point. They look for the questions the student asked. They look for the choices the student made. They look for the feedback the student considered. They look for the revisions the student made. They look for the reasoning behind those revisions. They look for the student’s ability to explain and defend the final work.

These forms of evidence do not need to be complicated. In many classrooms, they can be captured through short annotations, process notes, conferencing, version comparison, oral explanation, reflection prompts or learning journals.

The key is to make the evidence useful. Visible Agency should not become another layer of paperwork. It should help teachers see the learning and help students become more conscious of their own thinking. A useful test is whether the evidence improves the learning conversation.

Can the teacher give better feedback because of it? Can the student explain their learning more clearly because of it? Can both teacher and student see what changed?

If the answer is yes, the evidence is probably worthwhile.

What should school leaders notice?

Visible Agency is not only a classroom design issue. It is also a leadership issue.

As schools respond to AI, there is a temptation to focus first on rules, tools and detection. These have a place, but they do not answer the deeper learning question. The deeper question is whether the school has a shared understanding of what quality learning looks like when AI is available.

Teachers need more than permission statements. They need design principles. They need examples. They need a common language for talking about student thinking, judgement, revision and responsibility. They need time to examine tasks together and ask whether those tasks still require meaningful learning.

Visible Agency gives leaders a constructive frame. It helps schools move from “Are students allowed to use AI?” to “How do we design learning so students remain active, responsible and thoughtful when they use AI?”

That is a more useful professional conversation.

It also protects teachers from having to solve the AI challenge as isolated individuals. If Visible Agency becomes a shared design principle, teams can review tasks together, build common expectations, examine student evidence and refine practice over time.

AI-rich learning should not depend on every teacher inventing their own private rules. It should be guided by shared learning principles.

The Visible Agency shift

The shift is simple, but it changes the work.

• Do not only ask students to produce. Ask them to show how they thought.

• Do not only ask whether AI was used. Ask what the student still had to decide.

• Do not only assess the final product. Look for evidence of judgement, revision and ownership.

• Do not treat AI as separate from learning design. Build the task so AI use becomes part of the thinking process.

The goal is not to keep AI out of learning. The goal is to keep the learner in it.

This is the future-facing challenge for learner agency. AI is not going away. Students will continue to have access to tools that can generate, summarise, translate, organise, explain, design and revise. The question is whether schools will respond by narrowing learning or by designing it more deliberately.

Visible Agency offers a way forward. It allows teachers to embrace AI support without lowering the expectation for student thinking. It allows students to use powerful tools without disappearing behind them. It allows leaders to focus on learning design rather than only compliance.

Most importantly, it keeps the learner visible.

Where to go next

Visible Agency is a broad design principle. The following articles develop the key parts of the framework in more practical detail.

• A practical next step is The Visible Agency Design Test. This tool helps teachers and teams examine whether an AI-supported task requires visible thinking, human judgement, ownership, metacognition, effort integrity, transparency and appropriate use.

• For the evidence side of the work, read How to Make Student Thinking Visible When AI Is Part of the Process. This article explores how students can show their questions, reasoning, comparisons, revisions and decisions without turning the learning process into unnecessary documentation.

• For the cognitive demand side of the work, read How to Stop AI From Replacing Student Thinking. This article examines how task design can preserve the learning demand of a task while still allowing AI to support the process.

• For the judgement side of the work, read How to Design AI Tasks That Require Student Judgement. This article explores how students can evaluate AI responses, compare alternatives, identify limitations and make informed decisions.

• For the reflective side of the work, read How to Use AI to Strengthen Metacognition. This article focuses on how AI can help students think about their own thinking, including how they use feedback, monitor confidence and recognise changes in understanding.

• For the ownership side of the work, read How to Design AI-Rich Tasks That Still Require Student Ownership. This article examines what responsibility, authorship and explanation mean when students work with AI.

Frequently asked questions

Final thought

Learner agency has always been about more than choice. It is about the student’s capacity to act with intention, awareness and responsibility.

AI does not make that less important. It makes it more visible as a design challenge.

If students can use AI to produce work without thinking deeply, the task needs redesigning. If students use AI to question, compare, decide, revise and take responsibility, the task may be doing something far more powerful.

Visible Agency gives teachers a way to hold that line with clarity and confidence.

The work ahead is not simply to decide whether students can use AI. The work is to design learning where their thinking remains visible, their judgement remains active and their ownership remains intact.

That is the promise of Visible Agency.