Key takeaways:
- AI has changed what a completed student task can prove.
- The issue is not AI use itself, but whether the learner remains visible inside the work.
- Schools need better evidence of thinking, judgement, revision, and ownership.
- Better evidence does not mean more documentation; it means making the most important human work visible enough to teach and discuss.
- The first leadership move is to review one common task and ask what human capability it actually reveals.
Keeping the Learner Visible in Learning
Transcript
There is a particular kind of pause that happens in schools now. It is not always dramatic. It often comes after a teacher has read a piece of work that is better than expected, cleaner than expected, more fluent than expected.
The work is complete. The student has submitted it. The criteria appear to be met. On the surface, there may be nothing obviously wrong. In fact, the work may be quite good. And then, quietly, the question arrives: how much of this shows what the student actually did?
This week I want to sit with that question, because I think it is one of the first honest questions schools need to ask in an AI-present learning environment. Not because we should begin with suspicion. Not because every polished piece of work should be treated as a problem. But because the relationship between student work and student learning has changed.
For a long time, teachers have been able to infer quite a lot from a completed task. A written response usually carried traces of reading, interpretation, structure, and revision. A research project usually carried traces of searching, selecting, comparing, evaluating, and synthesising. A presentation usually carried traces of preparation, understanding, and communication.
Those inferences were never perfect. Teachers have always known that final products are partial evidence. A strong answer does not always mean deep understanding. A polished essay does not reveal every decision behind the argument. A completed project does not show every misconception that was resolved, avoided, or hidden. Still, the final product often gave teachers enough evidence to begin a useful learning conversation.
AI changes the reliability of that inference. A student can now receive support with planning, summarising, drafting, editing, translating, reorganising, generating examples, checking clarity, and improving fluency. Some of that support can be genuinely helpful. It can give students access, confidence, feedback, and momentum. It can help a student get started when they are stuck. It can help a student see options they might not have generated alone.
The problem is not that support exists. Support has always been part of learning. Teachers support. Peers support. Worked examples support. Graphic organisers support. Feedback supports. The more useful question is whether the support leaves the learner more capable.
That is the distinction I want leaders to hold carefully. This is not a simple argument about stopping AI. It is a deeper argument about learning evidence. If the visible product can be improved quickly, schools need a clearer view of what happened underneath the surface.
A polished paragraph may not show whether the student understood the concept before the language was improved. A strong set of slides may not show whether the student created the structure. A research summary may not show whether the student judged the quality of the source material. A reflection may not show whether much reflecting actually happened.
That last one matters. Reflection is often treated as a safeguard, but a reflection added at the end can become just another product. It may sound thoughtful without revealing much of the thinking that shaped the work. If the real learning happened earlier, then the evidence needs to appear closer to the moments where decisions were made.
Assessment-for-learning research has always rested on this idea: teachers and students need evidence they can act on. Black and Wiliam's work made this clear long before generative AI entered the classroom. Evidence matters because it helps the learner and the teacher decide what should happen next. But if the evidence we receive is only the final artefact, and that artefact has been shaped by invisible support, the next learning decision becomes harder to make.
This is where I think many school leaders are feeling the tension. They are not necessarily worried because students are using technology. They are worried because the learner is becoming harder to see.
And that is an important distinction. If the problem is framed only as technology use, then the response becomes rules, restrictions, and detection. Those things may have a place, but they do not answer the deeper learning question. The deeper question is whether the learning experience still requires students to notice, question, compare, judge, revise, explain, and take responsibility.
The OECD Learning Compass is useful here because it reminds us that future-ready learning is not only about knowledge acquisition. It connects knowledge, skills, attitudes, values, and agency. In other words, the question is not only what learners know. It is how they use what they know with judgement, responsibility, and intention.
That broader view makes the challenge clearer. A student can submit a finished product that shows something, but not everything. It may show completion. It may show fluency. It may show accuracy. But does it show judgement? Does it show ownership? Does it show the learner becoming more capable?
Those are leadership questions as much as classroom questions.
The product still matters. We should not pretend otherwise. Students need to produce work of quality. They need to communicate clearly, reason carefully, build accurate explanations, and meet high standards. The issue is not whether products matter. The issue is whether products can carry the whole story alone.
The human work is often in the questions students ask, the choices they make, the options they reject, the revisions they justify, and the ownership they can demonstrate when asked to stand behind the work. These are not decorative extras. They are often where the learning is most visible.
So what does this look like in practice?
A student writing an explanation might begin with an initial claim, use AI to test for gaps, compare the AI feedback with class learning, revise the response, and annotate one change that improved accuracy. The final explanation still matters. But now the teacher can see something about the student's judgement.
A student preparing a presentation might generate two possible structures, choose one, and explain why that structure serves the audience better. The slides still matter. But now the teacher can see something about the student's decision-making.
A student completing a design task might use AI to generate possible improvements, reject two suggestions because they do not meet the design constraint, and defend the final choice. The product still matters. But now the teacher can see something about the student's ownership.
None of this needs to become heavy. It does not require documenting every prompt or every step. It requires identifying the moments where human capability matters most and making those moments visible enough to improve the learning conversation.
The practical move this week is deliberately small. Choose one common task. Not the whole assessment system. Not every unit. One task teachers already use. Ask what the final product reveals about the learner, and what it leaves hidden.
Then ask the AI-present version of the question: if a capable AI tool supported this work somewhere in the process, what human capability would still be visible?
If the answer is unclear, the task may still have value. But it may be carrying less evidence than the school needs.
The response might be a short explanation of why a student chose one idea over another. It might be an annotation showing what changed between draft and final version. It might be a comparison between an AI suggestion and the student's own reasoning. It might be an oral check where the student explains what they understand and what they are prepared to defend.
The point is not more paperwork. It is better evidence.
Visible Agency begins with this recognition: the learner cannot disappear inside the support. AI may help students produce better work, but the school still needs to see the thinking, judgement, revision, and responsibility that make the work evidence of learning.
That is why the first question is not, "Did the student use AI?" The stronger question is, "What did the learner still have to think through?"
If that question becomes part of the professional conversation, schools can respond to AI with more than fear or detection. They can begin designing learning where the human work becomes visible again.
I want to add one more layer before we close, because this is where the leadership work becomes practical.
The question, how do we know the student did the thinking, should not become a slogan. It should become a design question. If it stays as a slogan, it can easily become suspicion. Teachers may begin to look at polished work and wonder whether something is wrong. That is understandable, but it is not enough. The stronger move is to ask what the task was designed to reveal.
A task that only asks for a final answer will usually leave the learner harder to see. A task that asks for one decision, one comparison, one revision, or one defence gives the teacher a better view. It also gives the student a clearer role. The student learns that using support does not remove the need to think. It increases the need to understand and explain the thinking.
This is an important culture message. Students should not hear that AI makes them suspicious. They should hear that learning still requires responsibility. They should hear that quality work matters, and that the thinking behind quality work matters as well.
For leaders, this is also a workload message. Teachers do not need to become investigators of every hidden process. They need help designing better evidence into the work. That evidence can be small. A comparison. A revision note. A source check. A one-minute oral explanation. The test is whether the evidence helps the teacher and student understand the learning more clearly.
So if you are leading this conversation in a school, I would begin with one task and one team. Ask what the final product currently shows. Then ask what it leaves hidden. Then choose one human capability the task should reveal. Not five. One. Judgement. Revision. Ownership. Questioning. Explanation.
Once that capability is named, the design question becomes much easier: where will students make that capability visible?
That is the first practical step toward Visible Agency. It is not a policy rollout. It is not a detection strategy. It is a clearer relationship between the work students produce and the learning the school needs to see.
For the fuller framework, download the Visible Agency Leadership Paper at https://leecrockett.net/visible-agency.
