AI Literacy Matters More Than Coding in Modern Education

The Kid Who Couldn’t Code But Knew How To Talk To Machines

In a classroom in Switzerland, a student failed a programming assignment. He couldn’t write a single line of Python. But when the professor introduced a new exercise using a large language model, something strange happened. The student who had bombed the coding test became the best in the room at getting the AI to produce working code. He didn’t know how to program. He knew how to ask.

That moment, described in a recent paper by Yoshija Walter (2024), captures a shift that most education systems are not ready for. For the last decade, the rallying cry has been “everyone should learn to code.” Governments poured money into coding bootcamps. Parents enrolled six year olds in Scratch classes. The logic was simple: in a digital economy, programming literacy is the new basic skill.

But Walter’s research suggests we may have been aiming at the wrong target. The paper, published in the International Journal of Educational Technology in Higher Education and already cited over 750 times, argues that AI literacy matters more than coding proficiency in modern education. The finding is not just provocative. It is backed by a detailed case study from a Swiss university and a narrative review of the literature. And it forces a question that makes educators uncomfortable: if AI can write code for you, what is the point of learning to code?

What Is AI Literacy, And Why Is It Different From Coding?

Walter (2024) defines AI literacy as a cluster of competencies that go far beyond technical skill. It includes understanding what AI systems can and cannot do, recognizing their biases, knowing how to frame problems for machine intelligence, and evaluating the outputs these systems produce. Coding, by contrast, is about executing instructions. AI literacy is about directing intelligence.

The distinction matters because the two skills are not interchangeable. A person who can write a flawless for loop may have no idea how to craft a prompt that gets a language model to generate a coherent essay. Conversely, a student who cannot write a line of code may be highly skilled at breaking down a complex problem into the precise sequence of queries that an AI system needs to solve it.

Walter’s research involved a case study at a Swiss university where students were given tasks that required using AI tools. The students who performed best were not necessarily the strongest programmers. They were the ones who could articulate their needs clearly, iterate on prompts based on the AI’s responses, and critically evaluate the outputs. In other words, they had developed a kind of conversational fluency with machines.

This is not a niche skill. It is becoming the primary way humans interact with advanced technology. When you use ChatGPT, you are not programming. You are having a conversation. When you use Midjourney to generate an image, you are not writing code. You are describing what you want. The interface to computation is shifting from syntax to language.

The Prompt Engineering Paradox

One of the most counterintuitive findings in Walter’s (2024) paper concerns prompt engineering. You might assume that prompt engineering is just a fancy term for “asking good questions.” And it is. But the research shows that effective prompting is surprisingly difficult, and that it requires a specific set of cognitive skills that are not taught in standard curricula.

Walter (2024) found that students who excelled at prompt engineering shared several traits. They were good at breaking down complex tasks into smaller, more manageable queries. They understood the limitations of the AI system they were working with. They knew when to provide context and when to strip it away. They could detect when the AI was hallucinating or producing nonsense. And they were comfortable with iteration: they treated each prompt as an experiment, not a command.

These are not technical skills. They are metacognitive skills. They require self awareness, critical thinking, and a willingness to revise your own assumptions. Walter (2024) argues that these skills are more durable than coding knowledge because they transfer across different AI systems and evolve as the technology changes. The specific programming languages you learn today may be obsolete in a decade. But the ability to think clearly about what you want a machine to do, and to evaluate whether it did it correctly, will never go out of style.

This has direct implications for curriculum design. If prompt engineering is a core skill, then schools should be teaching students how to construct effective prompts, how to test them, and how to debug them when they fail. That is a very different activity from teaching Python or JavaScript.

Critical Thinking As The New Gatekeeper

The most alarming finding in Walter’s (2024) paper is not about AI literacy itself. It is about what happens when AI literacy is absent. Students who lack critical thinking skills are vulnerable to a new kind of error: they trust the machine too much.

Walter (2024) describes a phenomenon called “automation bias” in educational settings. When students use AI tools to generate answers, they often accept those answers without scrutiny. The AI produces fluent, confident text, and students assume it must be correct. This is especially dangerous because modern language models are designed to sound authoritative, even when they are wrong.

The research found that students who had been trained in critical thinking were significantly better at identifying AI errors. They were more likely to fact check claims, spot logical inconsistencies, and notice when the AI was making up sources. They treated the AI as a collaborator, not an oracle.

This is where the coding versus AI literacy debate gets sharpened. Learning to code might teach you some logical thinking, but it does not automatically make you a better evaluator of AI outputs. In fact, Walter (2024) suggests that coding education may create a false sense of competence. Students who can write code may assume they understand how the AI works, when in reality the AI’s internal operations are opaque even to its creators.

The real gatekeeper, Walter (2024) argues, is not technical skill. It is the willingness to doubt the machine.

How The Swiss University Case Study Was Done

The methodology of Walter’s (2024) study is worth explaining because it gives weight to the conclusions. The paper combines two approaches. The first is a narrative literature review, which synthesizes existing research on AI in education. The second is a case study conducted at a Swiss university, where the author observed students using AI tools in classroom settings.

The case study involved undergraduate and graduate students in courses that integrated AI tools into assignments. Students were given tasks that required them to use large language models to generate text, solve problems, and create content. Their performance was assessed not just on the final output, but on their process: how they formulated prompts, how they iterated, and how they evaluated the AI’s responses.

Walter (2024) does not report a specific sample size or effect size in the abstract, but the paper describes qualitative observations that were consistent across multiple cohorts. The key finding was that students who scored highest on measures of AI literacy were not necessarily the most technically skilled. They were the most metacognitively skilled.

This is a qualitative case study, which means it does not prove causation. But it provides a rich, detailed picture of what AI literacy looks like in practice. And it is grounded in a broader literature review that supports the same conclusion from multiple angles.

What The Research Does Not Prove

It would be easy to read Walter’s (2024) paper and conclude that coding education is a waste of time. That is not what the research says. The paper does not argue that coding is irrelevant. It argues that AI literacy is more foundational in the current moment, and that the two skills are not substitutes.

The study also does not prove that AI literacy can be taught effectively in all educational contexts. The case study was conducted at a university in Switzerland, which has a well funded education system and small class sizes. It is unclear whether the same approach would work in under resourced schools, or with younger students, or across different cultural contexts.

Another limitation is the rapid pace of AI development. The paper was published in 2024, and the specific tools and techniques described may already be outdated. What remains constant is the underlying principle: humans need to learn how to direct, evaluate, and collaborate with intelligent machines. But the specific methods for doing that will keep changing.

Walter (2024) also does not address the question of whether AI literacy can be meaningfully separated from domain knowledge. A student who knows nothing about biology will struggle to evaluate whether an AI’s biology answer is correct, no matter how good their critical thinking skills are. AI literacy is not a magic bullet. It works best when combined with substantive knowledge in a field.

The Practical Architecture Of An AI Literate Curriculum

Walter (2024) offers concrete suggestions for how to build AI literacy into education. These are not abstract recommendations. They are based on what worked in the Swiss university case study.

The first suggestion is to embed AI literacy across subjects rather than teaching it as a standalone course. Walter (2024) argues that AI literacy is not a technical skill like typing. It is a way of thinking that applies to history, science, art, and literature. A history student should learn how to use AI to analyze primary sources. A biology student should learn how to use AI to generate hypotheses. A literature student should learn how to use AI to explore different interpretations of a text.

The second suggestion is to teach prompt engineering explicitly. Walter (2024) found that students improved dramatically when they were given structured exercises in prompt construction. These exercises included writing prompts for specific tasks, testing them, analyzing why they failed, and rewriting them. This is a skill that can be practiced and improved, just like any other.

The third suggestion is to integrate critical thinking exercises that specifically target AI outputs. Walter (2024) recommends giving students AI generated text that contains deliberate errors and asking them to find and correct those errors. This trains the evaluative muscle. It also teaches students that AI is fallible, which is a lesson they need to learn early.

The fourth suggestion is to train educators. Walter (2024) notes that many teachers are unprepared to teach AI literacy because they lack the skills themselves. This is a systemic problem. If teachers cannot model effective AI use, students will not learn it.

What This Actually Means

• ▸Stop treating coding as the only path to technical competence. AI literacy is a separate, more transferable skill that deserves dedicated curriculum time. Students who cannot code can still become highly effective users of AI tools.

• ▸Teach prompt engineering as a core academic skill, like writing or argumentation. It requires practice, feedback, and iteration. It is not something students will pick up intuitively.

• ▸Build critical thinking exercises that specifically target AI errors. Students need to learn that AI is a tool, not an authority. They need to develop the habit of checking, questioning, and verifying.

• ▸Invest in teacher training before investing in new hardware. The bottleneck in AI education is not technology. It is the human capacity to use it wisely. Teachers need to be fluent in the same skills they are expected to teach.

• ▸Treat AI literacy as a cross curricular priority, not a computer science elective. It belongs in every classroom, because every discipline will be transformed by AI. The students who thrive will be the ones who learn to think with machines, not just code for them.