AI fluency programs are teaching the wrong skill.

FREE ISSUE

AI fluency programs are producing a new kind of illiteracy.

Not ignorance of the technology. Something more specific: training graduates to operate an interface while leaving untouched the judgment layer that determines whether the output is any good.

The failure is precise. And it matters because the programs are scaling while the underlying problem stays unfixed.

What institutions built

Most leaders recognized the gap early and built a response.

Workshops on prompting. Required modules in first-year experience courses. Certificates in AI literacy. The frame was familiar: a new technology arrives, institutions build a competency around it, graduates leave better prepared.

That frame worked for spreadsheets. It worked for statistical software. The assumption has been that it works here too.

It does not. And the reason matters more than the failure itself.

Why prompting is not a durable skill

The models improve on a curve that makes the interface progressively easier to use.

The distance between "ask the AI a question" and "get a plausible answer" is shrinking by design. The major labs are spending extraordinary resources making that distance disappear entirely. This is not a side effect of AI development. It is the goal.

A graduate whose competitive edge is closing the interface gap faster than the model closes it automatically is building on a foundation with a known expiration date. The prompt that required sophistication in 2023 is handled by default model behavior two years later. The interface is being automated. The judgment behind what to do with the output is not.

Most AI fluency programs were designed for the gap as it existed when the programs were funded. That gap is smaller now. It will be smaller again when this year's graduates are three years into their careers. Institutions are training for a capability that the technology itself is actively commoditizing. The courses exist. The certificates are being issued. The capability they certify is shrinking in value by design.

What does not expire

The requirement to know what a good answer looks like.

To diagnose a bad one before it causes damage. To design conditions under which AI output creates real value rather than credible-sounding filler. To know what question was worth asking in the first place.

These are not prompting skills. They are judgment skills. The difference sounds abstract until you put it in a specific context.

A student who can prompt an AI to write a market analysis is useful. A student who can tell you what the analysis got wrong, why the assumptions are fragile, and what the decision-maker actually needed to know is valuable. The gap between those two outcomes is not a prompting problem. It is a curriculum design problem dressed up as a technology adoption problem.

Three failures worth naming

The label is doing harm.

"AI literacy" frames the capability as a literacy problem, which implies it can be solved by exposure. Literacy is acquired through contact with the material. Judgment is not. It is a disposition built through repeated evaluation under real stakes, with feedback from someone who can tell the difference between a sharp output and a plausible one.

The label shapes the intervention. This label is shaping the wrong one.

Faculty development cannot be an afterthought.

The most common implementation is a single required module owned by a center that has no authority over curriculum. That structure produces checkbox compliance, not capability.

The judgment layer has to live inside the disciplines, taught by faculty who know what good judgment looks like in their domain. A business school professor can teach a student to evaluate a weak financial model. A general AI literacy module cannot teach that same student to know what the model missed. The discipline specificity is not optional. It is the whole mechanism.

Assessment is the leverage point.

If an assignment can be completed by an AI with no human judgment applied, the assignment is not teaching judgment.

That is not a reason to ban AI. It is a reason to redesign the assignment. The question shifts from "can this student use the tool" to "can this student create value with it." Those two questions produce different course designs, different assessment criteria, and a different answer to what faculty development needs to accomplish. Universities know how to teach content. They are considerably less practiced at teaching judgment as a transferable, assessable capability.

What the reframe requires

Judgment requires the student to work with ambiguous information under real constraints, evaluate outputs against a standard they have internalized, and produce work that holds up to scrutiny.

Most AI fluency modules require none of this. They require the student to produce output. The standard for that output is often "it exists and looks reasonable." That standard cannot distinguish a capable graduate from a capable relay.

The institutions beginning to get this right are not replacing AI fluency programs. They are redesigning the underlying question. Not whether students can use the tool. Whether they can create value with it. That reframe produces different course structures, different assessments, and a different relationship between AI program ownership and the departments where judgment actually lives.

The reframe also clarifies who the problem belongs to. A center with no curriculum authority cannot fix this. The fix requires faculty development within disciplines, assessment redesign with real stakes, and institutional authority vested in someone who can push that work across departments and hold the line when the easier option is another module.

This is a reframe problem, not a technology problem. The technology is not broken. The frame around it is.

The decision

What would it mean for your institution to answer the harder question: not whether your students can use AI, but whether they can create value with it?

Who owns that distinction inside your organization? Do they have the authority to act on it?