We May Be Teaching People to Use AI for a Version of AI That’s Already Disappearing
by Claire L. Brady, EdD
For the last few years, a lot of AI training has focused on one skill: how to tell AI exactly what you want it to do.
Write a better prompt. Provide more context. Break a complicated task into smaller steps. Give the AI a role. Build a workflow. Refine the output. I have taught plenty of this myself. And a lot more.
But Ethan Mollick's latest One Useful Thing made me wonder whether we're preparing people for a version of AI that may already be giving way to something different? Mollick writes that he expected humans to need to become managers of AI agents—carefully assigning work, structuring teams, and deciding how agents should collaborate. Instead, he is seeing increasingly capable AI systems do much of that organizing themselves. They find information, develop plans, delegate tasks to other agents, and coordinate work with far less human direction than he expected.
That has some pretty significant implications for higher education.
Prompting Isn't Going Away. But It May Matter Less.
We absolutely should help faculty, staff, and students learn how to communicate effectively with AI. But I would be cautious about building an entire AI literacy strategy around prompt engineering. The durable skill isn't memorizing the perfect prompt formula. It is knowing what you are trying to accomplish.
As AI gets better at determining how to complete a task, humans may increasingly be responsible for something harder: defining the goal, establishing constraints, providing judgment, evaluating the result, and deciding whether the work should be done at all. That is a very different kind of AI literacy.
From Giving Instructions to Setting Direction
Mollick describes personal agents that can learn context from messages and connected accounts, develop their own plans, and even identify mistakes without being explicitly asked to look for them.
Think about what that shift means on a campus.
Today, we might teach someone how to prompt AI to draft an email, analyze a spreadsheet, or summarize a report. Tomorrow, that person may be supervising an agent that notices the report is late, finds the relevant data, analyzes it, drafts the summary, identifies a discrepancy, and asks whether it should contact someone about it.
The skill isn't simply prompting anymore. It is oversight. Can you establish appropriate boundaries? Recognize when an agent has exceeded them? Evaluate work you didn't personally produce step-by-step? Know when human review is required? Take responsibility when an automated process gets something wrong?
Higher Ed Should Prepare for That Shift Now
This doesn't require throwing out today's AI training. It means widening it. Teach prompting, but pair it with goal-setting and verification. When piloting agentic tools, test not only whether they complete the task but what decisions they make along the way. Build human checkpoints around consequential actions involving students, employees, money, institutional data, or external communication. And perhaps most importantly, stop measuring AI readiness by how many people know how to use today's tools.
Mollick ultimately argues that humans may spend less time organizing AI work than he expected, while remaining responsible for deciding where all that capability should be pointed. That may be the leadership skill we should be developing now. The future of working with AI may be less about giving better instructions and more about exercising better judgment. And unlike the perfect prompt formula, that skill isn't likely to become obsolete with the next model release.
Note: This image was created using ChatGPT
