Imagine two people in the same course. One is ready to move on and needs a harder problem. The other has years of experience doing the work but hasn't yet found the language the course uses to explain it. They're given the same lesson, the same pace, and the same test.
One gets bored. The other starts wondering whether they belong.
We often treat that as a difference in ability. Sometimes it's a difference between the person and the way we've organized the learning. AI gives us a chance to pay much closer attention to that difference. It also gives us a reason to ask a bigger question: what are we preparing people to do once answers are easy to get?
The pace is not the person
When I say intelligence is becoming a commodity, I mean that increasingly capable tools can explain, analyze, draft, and recommend at a cost that changes what's practical. I don't mean those tools are always right, equally accessible, or free to everyone. The infrastructure, language, support, and trust required to use them still matter.
But the direction opens a possibility. An explanation can be tried again. An example can be connected to work someone already understands. A learner can get another question, another kind of feedback, or more time on the part that hasn't clicked.
Meeting someone where they are means noticing what they can do today and helping them take a meaningful next step. It means treating a wrong answer as information about what to teach next. An AI system should keep checking its assumptions about the learner, with ways for the learner and teacher to correct them.
That changes the experience of education. The person gets a path they can enter, rather than a timetable they have to catch.
Keep the standard. Open up the route.
We still need shared expectations. People need to read carefully, reason with numbers, understand their field, and recognize when they're out of their depth. A nurse, an engineer, or a project manager can't replace competence with confidence.
The opportunity is to separate the standard from the route used to reach it. Two people can demonstrate the same understanding after different amounts of practice, with different examples and different support. A common benchmark can tell us what someone needs to be able to do. It doesn't have to prescribe every step or become the ceiling on their potential.
That also changes assessment. Ask someone to explain their reasoning, apply it to a new situation, and notice what changes when a constraint moves. Include work they do independently, so we can see what they've learned rather than only what their tools can produce.
We should be careful here. A randomized study of high-school mathematics found that access to a general-purpose AI interface could improve assisted practice while harming later unassisted performance. Its more structured tutoring approach reduced that harm. That finding belongs to a particular setting. It gives us a reason to test whether the same problem appears elsewhere: completing the work and learning to do the work can come apart.1
Different paths. Shared capability.
Add complexity, weigh the tradeoffs, then apply the idea in a new situation.
Start with experience, connect the idea, then carry it into another setting.
Work through the difficult step, use feedback, then try independently.
Conceptual illustration
Education needs a bigger ambition
For a long time, employability has been one of education's central promises. Learn the method. Demonstrate the skill. Become someone an organization can rely on to do the work.
That promise still matters. But when AI can produce more of the analysis, plans, and first drafts, it leaves a harder responsibility in human hands. Someone has to decide what the work is for, which recommendation deserves trust, and what consequence they're prepared to accept.
We need people who see themselves as decision makers.
That doesn't mean everyone becomes an executive. A decision maker might be a technician who knows when to stop a process, an analyst who challenges a convenient assumption, or a new project manager who asks whether an urgent deadline is worth the risk it creates for customers. They know where their authority ends. They also know that raising the right question is part of their contribution.
Skill gives a person the means to act. Judgment helps them decide when, why, and with what responsibility. Education can develop both. The OECD's Learning Compass already places student agency at the center of a broader educational vision: helping learners navigate unfamiliar situations with purpose and responsibility.2
The next step is to make that ambition visible in the learning itself.
Give the learner a decision to carry
Picture a project-management lesson. A conventional exercise asks the learner to identify the correct process. A richer exercise gives them a troubled project, an anxious sponsor, a limited budget, and a team that's stopped speaking honestly.
What would you do next? What are you assuming? Whose perspective is missing? What would make you change your mind?
The learner chooses, explains the choice, and sees a plausible consequence. Then a constraint changes. They have to reconsider. AI could help make that practice more responsive, while an educator checks the quality of the scenarios and the reasoning they reward.
The useful cycle is simple: frame a problem, weigh the options, make a choice, examine what follows. Repeat it with enough variation that the learner has to think.
The feedback should reach beyond whether they selected the preferred option. Did they notice the tradeoff? Did they distinguish evidence from a confident claim? Did they ask for help when the stakes exceeded their knowledge? Could they explain a choice to someone affected by it?
Over time, the learner has a chance to build a different identity: I can understand a situation, contribute a judgment, and take responsibility for improving it. That identity has to be earned through practice and feedback. Encouragement helps when it's grounded in evidence of what someone can do independently.
A choice changes what comes next.
A smaller promise can protect the date, but some outcomes must wait.
Recheck the promise against the capacity already available.
No extra capacity is available.
Surface hidden constraints before choosing the next move.
Turn the team's constraints into choices the sponsor can own.
No extra capacity is available.
A bounded test trades some delivery capacity for evidence.
Use existing work to test the assumption that matters most.
No extra capacity is available.
Illustrative decision
Teachers and managers have a larger role
Teachers help learners interpret confusion, develop confidence without overconfidence, and learn to disagree with care. They choose what deserves attention and recognize when an apparently simple mistake points to something deeper. Those responsibilities matter even more when an automated explanation can sound convincing while being wrong.
AI can support that work when educators have time, training, and authority to shape its use. The OECD's Teaching Compass makes a related point: student agency depends on teachers having agency and professional support themselves.3 Buying a tool is only one part of changing the learning experience.
The workplace has a similar obligation. A graduate can arrive ready to exercise judgment and find an organization that rewards quiet compliance. A frontline employee can spot a problem and have no safe way to raise it.
If we want a workforce of decision makers, organizations need clear decision rights, room to question recommendations, and a way to learn from mistakes without hiding them. Accountability should make ownership possible. Managers will need to judge the quality of reasoning alongside the result, because a sound decision can still meet an unlucky outcome.
Education and work have to change together.
A different distribution of opportunity
The global possibility is substantial. More responsive learning could help someone study around a job, revisit a missed foundation, or work through an idea in language and examples closer to their life. People whose ability was obscured by a rigid pathway could have another way to demonstrate it.
Access needs deliberate design. Affordable connectivity, accessibility, local language quality, educator involvement, and protection of learner data all shape who benefits. UNESCO's guidance on generative AI in education puts human agency, equitable use, privacy, and age-appropriate design among the priorities.4 These are part of the educational work, not details to add after the product succeeds.
UniFrica gives us another way to explore this at Mintelan: connecting African scholars and universities with global classrooms through curated learning modules. The question is bigger than making content available. How does encountering expertise from different contexts change the assumptions a learner notices, the sources they trust, and the choices they consider?
If more people can develop informed judgment, the effects could reach far beyond productivity. More workers could challenge a weak assumption before it becomes an expensive failure. More communities could bring local knowledge into choices that affect them. More people could participate in decisions they previously felt unqualified to question.
That won't happen simply because the software gets better. Institutions still decide who gets heard, who holds authority, and whose consequences count. But education can help more people enter those conversations prepared to contribute.
A wider field of view.
Practice explaining a choice, questioning a recommendation, and knowing which decisions you can own.
Change the context to see whether understanding travels beyond a familiar exercise.
Local knowledge can reveal assumptions and consequences that a standard pathway overlooks.
Conceptual illustration
That's the possibility that interests me at Mintelan. Our research asks how AI can support human judgment. Tirtle gives us a place to explore the learning side of that question, starting with project-management education. Can practice and feedback help someone explain a choice, recognize uncertainty, and carry that ability into a new situation? Those are questions to test, not outcomes to assume.
The ambition should be larger than helping people keep up with machines. We can help people understand the choices in front of them and see themselves as capable of shaping what comes next.
When intelligence becomes abundant, education has a chance to make informed agency more abundant too.
Sources and notes
Bastani et al., Generative AI without guardrails can harm learning: Evidence from high school mathematics, PNAS, 2025. A randomized study in high-school mathematics; it does not establish effects across all subjects, ages, or products.
Return to articleOECD Learning Compass 2030. A conceptual framework for student agency and competencies, rather than proof that any particular AI product develops them.
Return to articleOECD Teaching Compass, 2025. A policy framework emphasizing teacher agency, competencies, and well-being.
Return to articleUNESCO, Guidance for generative AI in education and research, 2023. Guidance for human-centered, ethical, equitable, and age-appropriate use.
Return to article