Credible AI courses? We need them.

AI education risks repeating the prompt engineering mistake.

Prompt engineering was briefly treated as a discipline, even a career path. Courses proliferated. Credentials followed. Then the interfaces improved, the models absorbed the complexity, and the entire category collapsed. Not because it was fake. Because it was brittle. It was anchored to tool mechanics rather than durable principles.

The good news: AI literacy is moving in the right direction. Across education systems, there is growing recognition that effective AI use depends on skills rooted in the humanities: Socratic questioning, critical reasoning, ethical judgment, collective problem solving. These capacities are essential. They are also notoriously difficult to grade.

That difficulty is not a flaw. It is the point.

The mistake would be to relocate these skills into a new "AI subject" or to repackage them as technical competencies. Philosophy, literature, civics, history, and rhetoric already house what AI education actually needs. The problem is not where these skills live. The problem is that they now sit at the backbone of any credible AI program, and we have not confronted what it means to assess them in an AI context.

The assessment challenge is deeper than most policy discussions admit. When outputs are substantially AI generated, output quality stops being a reliable signal of student capability. Strong products can emerge from weak inputs. Poor products can reflect good judgment. Two students can submit identical work with radically different levels of understanding.

If we continue to grade artifacts, we quietly make the case that the student is unnecessary.

That is the uncomfortable implication we have to face.

This does not mean assessment is impossible. It means assessment has to move upstream. We can no longer evaluate only what students produce. We have to evaluate how they reason, what they question, what they reject, and when they take responsibility.

Those outcomes are human. They include the ability to detect errors, recognize manipulation, constrain systems, refuse inappropriate use, and know when escalation is required. These are judgment skills.

Here is the hard part. Once we do this, traditional assessment models begin to break. Reliability drops. Standardization wobbles. Rubrics feel subjective. But this is not a failure of rigor. It is what happens when we finally assess something that matters.

We already accept this in law, medicine, ethics, and other judgment heavy fields. AI pushes education into the same territory whether we are ready or not.

This work should not be driven by vendors or platform advocates. The definition of assessable AI outcomes should come from the strongest minds in education, ethics, cognitive science, and assessment design. And those outcomes must be credited, or they will remain optional and sidelined.

The humanities do not need to be replaced. They need to be recognized as the core of AI competence.

If we avoid the hard work of defining and assessing principled AI understanding because it is difficult, we will default back to tool training. And if we do that, we already know how this ends.

Prompt engineering was the warning. We should listen to it.

Ryan James Purdy is an AI Governance and Compliance Author, Educator, and Advisor with almost thirty years of education experience across North America, Europe, and Asia. Contact him at jamespurdy624@gmail.com or connect on LinkedIn at linkedin.com/in/purdyhouse.