AI is disrupting education, and it’s also opening real opportunities. I think both are true. How we respond depends on getting the order right: human competency and wisdom first, then AI skill on top of them.

Disruption

Students adopted AI faster than schools could respond. Stanford’s 2026 AI Index reports that 4 in 5 university students now use generative AI.

That use comes with misuse. A survey of more than 95,000 students at 20 public research universities estimated that about 9% of students who use AI have used it to cheat, and among daily users, the rate rose to 26%.

It also comes with lost learning, even when no one is cheating. High school math students with unrestricted access to GPT-4 scored 17% worse on exams once the AI was taken away. And in courses where AI can do much of the graded work, grades have risen while the skills behind them may not have (see Human Atrophy, Apathy, and AI Advancement).

Note what that means for teachers. Assignments that worked for decades (e.g., the take-home essay, the programming project) no longer tell us what they used to about what a student can do.

Opportunity

The opportunities are real, too.

In a randomized trial at Harvard, physics students learned significantly more, in less time, with an AI tutor than in an active-learning class. Note that the tutor was carefully designed around the course’s learning goals. It wasn’t a general-purpose chatbot. Even in the study where unrestricted GPT-4 hurt exam scores, a version designed to give hints instead of answers largely avoided that harm.

AI can help teachers, too. The Brookings report discussed below notes that AI can take on routine teaching tasks (e.g., parent emails, worksheets, quizzes), and it cites a U.S. study that found teachers who use AI save nearly six hours a week. The report also points to AI making classrooms more accessible (e.g., for students with dyslexia) and reaching students who are shut out of school entirely, like girls in Afghanistan who have been barred from formal education beyond primary school.

Why the risks currently outweigh the benefits

After consulting more than 500 people across 50 countries and reviewing more than 400 studies, the Brookings Center for Universal Education concluded in 2026 that the risks of AI in education currently outweigh its benefits. The main reason, in their view, is that the risks undermine children’s foundational development.

The authors called their report a “premortem,” a look at what could go wrong while there’s still time to change course. Among their recommendations, two stand out to me. One is making school less about completing tasks for a grade and more about curiosity. The other is AI literacy for teachers and students alike.

I agree with their conclusion, and I’d add that the kind of AI literacy matters a great deal.

The AI Literacy Framework

Most conversations about AI literacy focus on AI skill: knowing which chatbots exist, how to prompt them, and how to fit them into your work. That skill matters. But on its own, it’s the top of a pyramid with nothing underneath it.

The AI Literacy Framework describes three levels, each built on the one below it. The Faithful Automation framework is the middle level.

Foundational human competency

The base of the pyramid is everything a person knows and can do without AI (e.g., reading closely, writing clearly, solving a math problem, debugging a program, recognizing a bad diagnosis). It’s built the slow way, through study, practice, struggle, and experience.

This level comes first because every other level depends on it. You can’t anchor truth in a field you don’t understand. AI sounds just as confident when it’s wrong, and its tendency to agree with you is most dangerous when you’re out of your depth. Expertise is what lets you notice.

It’s also the level AI most easily erodes, as the studies above show. When the goal is knowledge and skill without dependence on AI, using AI can defeat the purpose of the task.

Note that students, by definition, are often still building this level. That’s a big part of why AI in education deserves so much care.

Wisdom and discernment

The middle of the pyramid is the ability to decide well whether and how to use AI. In practice, that means being able to:

• examine your intent in using AI for a task,
• anchor truth, knowing AI is neither transparent nor inherently trustworthy,
• steward your own thinking, time, and gifts, recognizing that the work forms you,
• and direct all of your work toward the service of others.

This is what the Faithful Automation framework is for. Its four questions (Intent, Truth, Stewardship, and Connection) are meant to build these habits of discernment one task at a time.

Wisdom isn’t something we produce on our own. As the framework puts it, we rely on Christ’s faithfulness and wisdom, and when we lack wisdom, we ask God, who gives generously (James 1:5). We also grow wise by walking with the wise (Proverbs 13:20). For teachers, that means our own wisdom with AI shapes our students far more than any policy does.

AI skill

The top of the pyramid is the ability to use AI technologies to solve problems within our vocations while minimizing the risks that come with them. That includes understanding what AI is and how it produces answers, knowing how it tends to behave and how we tend to behave when we use it, and having real experience using it.

This level is valuable. AI can help with a huge range of work, and pretending otherwise does students and colleagues no favors. But AI skill without the levels below it is risky. A person who’s fluent with AI but lacks expertise can’t tell when it’s wrong, and a person without discernment may hand off the very work that would have formed them.

Why the order matters

The pyramid is built from the bottom up. Human competency gives discernment something to stand on, and discernment is what keeps AI skill pointed in the right direction.

This matches what the Brookings researchers found. In the terms of this framework, AI tends to threaten the bottom of the pyramid.

It isn’t only true in school. Every vocation has its own foundation (e.g., a nurse’s clinical judgment, a developer’s understanding of their own code), and AI skill built on top of a weak foundation is fragile. Establishing human competency and wisdom as the foundation is what lets us use AI faithfully within our vocations.

None of this means waiting until the lower levels are “finished” before touching AI. They’re never finished. Rather, it means giving the most attention to whichever level is weakest.

Questions to consider

• Which level of the pyramid is strongest for you right now? Which is weakest?
• If you teach, which level are your learning objectives actually aimed at? Does your use of AI (and your students’ use) support that level or undercut it?
• Do you agree with this view of AI literacy? What would you add, remove, or reorder?