How does students' prior computational thinking skill change what they gain from AI coding assistants?

Why prior skill could change what a student gets out of the tool

An AI coding assistant can hand a student working code in seconds. Whether that becomes learning depends on what the student does with the answer. A student with stronger computational thinking — the habit of breaking a problem into parts, tracing an algorithm, and checking whether a result makes sense — has more internal machinery to compare the AI's suggestion against. A student with weaker skills has less to compare against, so accepting the output is the path of least resistance.

The clearest signal comes from a study comparing two ways of using a generative AI assistant in programming learning: one that supplies answers, and one that uses Socratic questioning to push students to reason [4]. The comparison measured computational thinking components such as decomposition, algorithmic thinking, evaluation, and generalization after the learning activity, and it also tracked "epistemic laziness" — essentially, a willingness to let the tool do the thinking [4]. That design matters because it isolates the style of AI support rather than the mere presence of AI.

The gains show up in reasoning, not just in finished code

When AI support is built around questioning rather than answering, students in K-12 programming courses showed effects on learning achievement, computational thinking, and metacognitive awareness — their awareness of how they were thinking and learning [2]. The same study frames a problem that matters for the prior-skill question: giving identical assistance to every student is not the same as giving each student what they need, which is why adaptive support is raised as a design goal [2].

This is consistent with the idea that the tool's value is partly in the dialogue it forces. A student who is prompted to explain a step has to retrieve and use their existing computational thinking; a student who is simply given the step does not. The studies support this as a plausible mechanism, not a proven cause — the comparisons are about support styles, not about measuring each student's starting skill and then predicting their gain.

The students with the least prior skill are the heaviest users

Among undergraduate computer science students, AI coding assistants were most frequently used by learners with beginner and intermediate programming skills [3]. That usage pattern is the practical reason prior skill matters so much: the students most likely to lean on the assistant are also the ones with the least independent basis for judging its output. The same study examined how AI-assisted tools affect coding competency, but it did not measure computational thinking, so it cannot tell us whether heavy use by beginners narrows or widens the gap [3].

Perception data point in a similar direction. In an introductory programming course, students described AI tools as helpful for grasping code, and a subset of 20 students was interviewed about the challenges they faced [5]. That is a report of how students felt, not a measure of what they learned, and the study did not assess computational thinking [5]. Still, it fits the picture of beginners using the tool as an explainer — a use that could build understanding or could substitute for it, depending on how the student engages.

What we still cannot say about prior skill

None of these studies directly measured students' computational thinking before the AI intervention and then compared gains across skill levels. The productivity-focused work on AI coding assistants looks at outcomes such as task completion time and code quality among undergraduate computing students, but it does not measure computational thinking at all [1]. So the honest answer to the question is that the mechanism is plausible and partially supported, while the direct test — does a high-CT student gain more, or differently, than a low-CT student from the same assistant? — is still missing from this evidence.

What the evidence does support is a practical distinction. The style of AI support appears to shape whether students reason or coast [2][4], and the students most exposed to these tools are those with the least prior skill [3]. Until studies measure prior computational thinking and track gains across levels, the safest reading is that prior skill likely changes the character of the benefit — more reasoning practice versus more answer-getting — rather than simply scaling it up or down.

Sources used in this answer

Related research questions