At 2:13 in the morning, the university’s learning platform was still busy.

A lecture was playing. A quiz was being completed. A discussion post had just been submitted. Another assignment was being uploaded.

Everything looked normal.

Except for one small detail:

The student was asleep.

He wasn’t sitting at his desk. He wasn’t watching the lecture. He wasn’t struggling with the quiz or searching for the right words for his assignment.

His computer was doing the work.

More precisely, an AI agent was doing it.

And this is where the story becomes uncomfortable.

Because this is no longer simply about a student asking ChatGPT to write an essay. The AI isn’t helping the student complete the course.

The AI is becoming the student.

That is the uncomfortable possibility created by the rise of AI agents in education — and universities are only beginning to understand what it could mean for learning, assessment, academic integrity, and the value of a degree.

The new kind of cheating

For years, educators have worried about students using artificial intelligence to write essays, solve homework problems, or generate answers.

The response has been relatively familiar: improve plagiarism detection, introduce AI-detection systems, redesign assignments, and remind students that inappropriate AI use may constitute academic misconduct.

But agentic AI changes the nature of the problem.

An AI agent is not limited to generating a paragraph in response to a prompt. Depending on the system and the permissions it receives, an agent can interact with websites and software, navigate interfaces, follow instructions, and perform sequences of tasks.

In an online course, that could potentially mean opening the learning platform, navigating course material, answering questions, completing quizzes, writing discussion responses, and submitting assignments.

The student may not need to sit there and prompt the AI one question at a time.

They could potentially hand over much of the workflow.

And that creates a strange new scenario.

A university believes it has enrolled a student. But behind the screen, a machine may be attending the classes, completing the work, and generating the evidence that the student has learned.

We’ve already explored the hidden risks of artificial intelligence in education, but agentic AI represents a fundamentally different category of concern.

This is no longer simply about AI assisting learning.

It is about the possibility of AI replacing the learner’s participation in the learning process.

At 2:13 AM, the student is asleep.

But their coursework isn’t.

Who earned the degree?

Now imagine that this continues for four years.

The AI watches the lectures.

The AI completes the assignments.

The AI writes the discussion posts.

The AI answers the quizzes.

The AI helps prepare the projects.

Eventually, the student receives a degree.

Now imagine that person walks into a job interview.

The employer looks at the résumé and sees a university qualification. But what does that qualification actually tell us?

Does it tell us that the person understands the subject?

Does it tell us that they can analyse a difficult problem?

Does it tell us that they can communicate an argument?

Does it tell us that they can apply their knowledge to a situation they have never encountered before?

Or does it primarily tell us that they successfully completed a sequence of assessments — assessments that may have been performed partly or substantially by a machine?

That is the deeper problem AI is forcing universities to confront.

And it goes beyond the familiar debate about whether AI is genuinely improving education.

AI may very well improve education.

It may provide personalised tutoring, instant feedback, accessibility support, language assistance, and new ways of exploring difficult concepts.

But there is a more fundamental question that comes first:

How do we know that the human being actually learned?

The problem was hiding in plain sight

It would be easy to turn this into a story about dishonest students.

But there is a more interesting problem underneath.

Perhaps AI isn’t destroying online education.

Perhaps it is exposing weaknesses that were already there.

Consider how many online courses are structured:

Watch the lecture.

Read the chapter.

Answer the quiz.

Write the discussion post.

Submit the assignment.

Receive a grade.

Repeat.

The system is built around a simple assumption:

The person behind the account is the person doing the work.

For most of the history of online education, that assumption was reasonable.

Agentic AI challenges it.

If an AI system can navigate a learning platform, interpret instructions, generate responses, and complete a sequence of tasks, then an assignment that once appeared to measure learning may actually be measuring something else:

the ability to delegate the work to a machine.

That is a much bigger problem than plagiarism.

Traditional plagiarism asks:

“Did this student copy someone else’s work?”

Agentic cheating asks:

“Did this student do the work at all?”

Those are fundamentally different questions.

From AI Assistance to AI Substitution

The distinction becomes clearer when we compare traditional AI-assisted cheating with agentic AI cheating.

Traditional AI Cheating	Agentic AI Cheating
Student asks AI for an answer	AI may navigate the platform autonomously
AI helps with individual tasks	AI can potentially coordinate multiple tasks
Student remains involved in the process	Student involvement may be minimal
Usually focused on producing an output	Focused on completing an entire workflow
Detection often focuses on the submitted work	Detection may require understanding the process
Student uses AI as a tool	Student potentially delegates the learning activity itself

This is why the arrival of agentic systems matters.

Tools such as Claude Cowork and OpenAI’s browser automation were designed primarily around productivity and computer interaction.

But the underlying capability is broader.

Once an AI system can understand a goal, interact with a browser, navigate websites, make decisions, and perform multiple steps without constant human intervention, the technology doesn’t inherently know whether the task is:

“Book me a meeting.” or “Complete my online university coursework.”

The technology is agnostic about the purpose.

Education, however, cannot be.

We may be asking the wrong question

Universities are understandably asking:

“How do we stop students from using AI?”

But perhaps that isn’t the most useful question.

A better question might be:

“How do we know that learning has actually happened?”

There is an important difference.

A student can submit an excellent essay without understanding it.

A student can pass a multiple-choice test through memorisation without being able to apply the knowledge.

A student can receive help from a tutor without becoming dependent on the tutor.

And now, potentially, a student can delegate significant portions of the academic workflow to an AI agent.

So perhaps education needs to move away from simply checking whether something was submitted.

Instead, it needs better ways of determining what exists inside the student’s head.

That means assessment may have to become more personal, more interactive, and more difficult to outsource.

Examine the Thinking, Not Just the Product

Ask a student to defend an argument.

Give them a new problem they have never seen before.

Ask why they selected one solution instead of another.

Ask them to explain a concept in their own words.

Challenge one of their assumptions.

Change the conditions of the problem and see whether they can adapt.

Ask them to respond to an unexpected question.

In other words:

Don’t just examine the product. Examine the thinking behind it.

This doesn’t mean universities should abandon technology or return to handwritten examinations for everything.

It means assessment needs to evolve alongside the technology.

UNESCO has advocated a human-centred approach to AI in education, emphasising that AI should support human development and learning rather than simply replace human agency. Its guidance also highlights the need to rethink educational practices as generative AI becomes increasingly capable.

The challenge, therefore, isn’t simply to make AI impossible to use.

It is to design assessments where using AI does not eliminate the need to think.

AI may force education to become more human

There is an interesting irony here.

We often talk about AI making education less human.

But AI may ultimately force education to become more human.

If machines can generate essays, answer routine questions, summarise books, solve standard problems, and perform increasingly complex digital tasks, then universities may have to place greater value on the things that cannot simply be outsourced:

Judgement.

Curiosity.

Reasoning.

Creativity.

Communication.

Collaboration.

Problem-solving.

Adaptability.

And most importantly:

Understanding.

The classroom of the future may therefore look very different from the classroom of the past.

Not because teachers disappear.

Not because universities disappear.

And not because AI is inherently bad for education.

But because the old evidence of learning may no longer be reliable.

As BBC Education reports, institutions around the world are increasingly grappling with what authentic assessment should look like in an age of generative AI.

The question is becoming impossible to ignore:

What does it mean to prove that someone knows something when machines can produce convincing evidence of knowledge?

The degree may have to prove more than completion

Universities have traditionally relied heavily on outputs.

A paper was submitted.

A test was completed.

A project was presented.

A grade was awarded.

Credits accumulated.

A degree was issued.

But the rise of increasingly capable AI systems creates a gap between completion and competence.

A completed assignment does not necessarily prove understanding.

A high grade does not necessarily prove independent reasoning.

And a degree may not always provide enough information about how the underlying knowledge was acquired.

This doesn’t make university degrees meaningless.

It makes the evidence behind them more important.

Future assessment may therefore involve more oral examinations, live problem-solving, project demonstrations, personalised questioning, practical work, collaborative activities, supervised assessments, and continuous evaluation.

The goal would not be to prove that students never used AI.

The goal would be to determine whether the student can think, explain, adapt, and perform without outsourcing the entire intellectual process.

Some countries are already rethinking what students need to learn

The challenge isn’t limited to academic integrity.

AI is also changing the skills universities need to teach.

China is replacing traditional university degrees with AI-focused programmes, reflecting a broader recognition that the skills graduates need are changing as artificial intelligence becomes embedded in professional work.

But changing the curriculum is only part of the problem.

Universities can teach students how to use AI.

They can teach prompt engineering, AI literacy, automation, data analysis, and machine learning.

They can encourage students to work alongside intelligent systems.

But one fundamental question remains:

How do we verify that the human being acquired those skills?

If a machine earned the credits, what exactly does the degree certificate represent?

The real risk isn’t that students will use AI

Perhaps this is the most important distinction.

The future of education is unlikely to be completely AI-free.

Students will use AI.

Teachers will use AI.

Universities will use AI.

Employers will expect graduates to know how to work with AI.

Trying to eliminate the technology entirely may therefore be unrealistic.

The real danger is different.

The danger is allowing AI to become a substitute for learning while continuing to mistake the resulting output for evidence of learning.

That is a much harder problem.

And it cannot be solved simply by installing another AI detector.

Final thoughts: The Sleeping Student

And that brings us back to the sleeping student.

At 2:13 in the morning, the university’s system records another completed task.

From the university’s perspective, everything appears normal.

A student has attended.

A student has completed the assignment.

A student has passed the assessment.

But the student is asleep.

The unsettling possibility isn’t simply that students may cheat with AI.

It is that our educational systems could continue to certify learning without actually knowing whether anyone learned anything.

And that forces universities to reconsider what their assessments are really measuring.

Perhaps the most important question about a degree in the age of agentic AI will no longer be:

“Did you pass?”

It may be:

“Who or what actually learned?”