AI Literacy Is Built, Not Announced

The students did not wait for permission. They opened ChatGPT, Claude, and Gemini, and they started using them in coursework while the institution was still drafting its first memo. That is the readiness gap, and most campuses have been trying to close it from the wrong end. They reach for a policy when what their people actually need is practice.

New research out of San Diego State University, published this week, shows what happens when a school closes the gap from the right end. It is one of the cleanest pieces of evidence we have seen for something Harvest Kernel has argued from day one: AI literacy is not a rule you announce. It is a skill you build in small, deliberate reps.

The study that quietly settles an argument

Researchers at SDSU tracked faculty who completed a self-paced generative AI micro-credential, then measured how their behavior and beliefs actually moved.¹ The peer-reviewed results landed in the Journal of Information Systems Education.² The numbers are not flashy. They are something better. They are believable.

Here is the part that matters. Faculty who finished became more likely to use generative AI and less skeptical of its place in education. At the very same time, they grew more insistent on verifying what AI produces and weighing the ethical risks. Confidence went up. Caution went up. Together. The lead researcher said the balance was the entire point, and that the goal was never to tell faculty they must use AI, but to give them the judgment to decide when it helps, when it does not, and how to talk with students about the difference.

Why confidence and caution rising together is the whole game

Now, you might be thinking that more confidence and more caution sound like opposites, and that a training program should pick a lane. That instinct is exactly backwards, and it is worth slowing down on.

Skepticism without skill is just avoidance. Confidence without judgment is just recklessness. The educators we worry about most are not the ones who refuse to touch AI, and they are not the ones who paste its output into a syllabus unread. They are the ones stuck oscillating between those two failure modes because nobody ever gave them reps. What the SDSU cohort developed was the thing that sits between fear and blind trust. Call it earned discernment, and understand that you cannot decree it. It accrues.

That is also why a one-time workshop or a downloaded acceptable-use policy never moves the needle the way leaders hope. Those things transfer information. They do not build capacity. The faculty in this study did not just read about prompting and ethics. They produced syllabus statements, ethical action plans, and annotated transcripts of their own AI interactions. They did the thing, examined what came back, and adjusted. That loop is where literacy actually lives.

Policy tells. Practice teaches.

We made this case a couple of weeks ago when we argued that your AI policy is not AI literacy. A policy is a fence. It marks where the boundaries are. It does not teach anyone how to move skillfully inside them. SDSU did not lead with a fence. It led with a path, and the faculty walked it.

One of the study authors put the demand plainly, noting that faculty want practical support, not just policy guidance, and that they are asking careful questions about bias, privacy, academic integrity, and the future of learning. Read that again, because it is a quiet indictment of how most AI rollouts are sequenced. Institutions write the rules first and offer the skill-building later, if at all. That order guarantees the readiness gap stays open, because rules describe a destination while practice provides the legs to get there.

And this is not a higher-education curiosity. Swap “faculty” for “your team,” “your staff,” or “you on a Tuesday morning,” and the finding holds. The professional who builds judgment through structured practice ends up both more capable and more careful than the colleague who attended a single mandatory training and called it literacy.

This is SeedStacking, validated in a faculty lounge

If the shape of this sounds familiar, it should. A self-paced credential, broken into small modules, each producing a real artifact, compounding into genuine fluency, is the SeedStacking approach wearing a university lanyard. Plant a small idea. Practice it until it sprouts. Grow it through application. Harvest a result you can actually use. The SDSU team would not call it that, but the mechanism is identical, and now there is peer-reviewed evidence that the mechanism works on the hardest audience of all: busy adults who were skeptical going in.

It also reframes what AI literacy even means. It is not a body of facts about transformers and tokens. It is a practiced stance: willing to use the tool, equipped to question it, and honest about its limits. You do not get that stance from a definition. You get it from reps, the same way you got fluent at anything else that once intimidated you.

Run your own micro-credential this week

Here is the part most coverage of a study like this leaves out. You do not need an institution, a budget, or a badge to do what those 145 faculty did. The structure is small enough to run on yourself, or your team, in a single week. Four reps, mapped to the four phases of SeedStacking.

Seed. Pick one real task you already do and dread, a recurring email, a lesson you rewrite every term, a report nobody reads, and run it through one AI tool. No stakes. Just see what comes back.

Sprout. Now do the harder thing the faculty did. Verify it. Hunt for the one claim that is wrong, the tone that is off, the citation that does not actually exist. This is the rep that builds caution, and it is the one most people skip on their way to either fear or blind trust.

Grow. Rewrite your prompt based on what you caught, then run the same task again. Notice how much of the quality lived in your framing, not the model. That noticing is the skill compounding.

Harvest. Write down, in one plain sentence, the rule you just learned about when this tool earns your trust and when it does not. That sentence is your literacy. It is durable because you earned it, not because it arrived in a policy memo you skimmed once.

Four reps. Maybe forty minutes. Do that across five different tasks and you will have built sturdier AI judgment than any one-and-done mandatory training delivers, for the exact reason the SDSU cohort did. You practiced the thing instead of being lectured about it. If you want the full four-phase version with prompts and a place to compare notes, that is what we build together every week inside the SeedStacking method.

Dean Le Blanc

Founder, Harvest Kernel

Dean helps educators and professionals build practical AI fluency through SeedStacking, small daily reps that compound into real skill. Join the free community.