Technology companies like numbers that move upward. More users, more sessions, more clicks, more time in the product. If an AI coach is used more often this month than last month, it is tempting to call that success. From a teacher's point of view, however, the picture is not so simple.
If a student needs the coach every time a new problem appears, opens deeper and deeper hints, and becomes less willing to begin without help, increased use is not evidence that the student is learning. It may be evidence that the student is becoming dependent on the support. The number went up, but the thing we actually care about may have moved in the wrong direction.
I do not want students to use less help simply for the sake of using less help. There is nothing noble about withholding support from a student who needs it, and independence does not mean isolation. The goal is to help students become more capable of making decisions, recognizing problems, using strategies, and explaining their own thinking. Over time, the support should be able to fade because the student has grown.
Help should have a direction
Think about the temporary structures used to build an arch. They hold the stones in place while the arch is being assembled, but they are not the arch. If removing the structure causes everything to collapse, the building is not finished.
Educational support works in much the same way. A hint, model, conference, checklist, sentence starter, or AI question can make difficult work possible. The support is doing its job when the student begins to take over some of what the support was providing. The student notices the missing step, asks the question, selects the strategy, or checks the work without being prompted.
That transfer does not happen automatically. Students may perform well with a model in front of them and then struggle when the surface of the task changes. They may follow a checklist perfectly without understanding why the steps matter. This is why we have to vary the practice, ask for explanations, and return to important ideas after some time has passed.
The strongest evidence arrives later
Immediate success feels good. A student fixes the answer, improves the draft, and moves the score in the right direction. Those are worthwhile outcomes, but they are still early evidence.
The more interesting questions come later. Can the student retrieve the idea without being told where to look? Can they detect a similar mistake in their own work? Can they use the strategy on a different task? Can they succeed with the same amount of support, or perhaps a little less?
This changes what we should want from educational AI. The purpose is not to keep the student engaged with the coach forever. The purpose is to help the student develop knowledge and habits of mind that continue to work when the coach is absent, quiet, or no longer needed.
Design for the day the help is gone
Every support should contain some idea about how it will eventually fade. A first hint might simply ask the student to look again at the evidence. A deeper hint might name the kind of reasoning that is missing. An example might use different content so that the student still has to transfer the idea. Later, the student should meet a related task and have a chance to begin without the same prompt.
This is the direction I want Coach to move. It should help when help is useful, but it should also pay attention to whether students become better able to think, revise, and explain on their own. That is harder to measure than clicks or usage, and it takes longer to see. It is also much closer to what teachers mean when we talk about learning.
The best outcome is not a student who becomes excellent at using Coach. It is a student who becomes more capable, more confident, and less dependent on Coach over time.