Adapting Homework for an Agentic World

✨Myrddin✨ — Mon, 25 May 2026 15:42:20 GMT

Of all the concerns over AI, the one that gets the least attention relative to its importance is the impact on education. The goal of our education system is to give students the skills and knowledge to succeed in life. How do we determine whether students have learnt what we have decided they need to learn? We assign them tasks and then judge their work. Now we have tools that can solve any task that a student might reasonably be set in school. How do we modify our system to account for this?

This situation is not entirely new; calculators, translators, and online answers have all made it more difficult to assess student abilities. Schools have found ways to deal with this by modifying the tasks set. AI agents are orders of magnitude more disruptive to the traditional forms of assessment. Examinations may still be useful, but many students will avoid work until it is too late, so we cannot rely on them making the logical decision to invest in their own education (especially at young ages). Similarly, trying to outright ban AI usage is not only unlikely to work but might mean we miss out on an amazing way of improving our learning efficiency. AI tools have an incredible ability to address one of the most difficult problems in education, which is the differing needs of every student. It can target the specific areas a student doesn’t understand as well as provide infinite examples in whichever format the student prefers. Motivated students will learn more than ever before, but unmotivated ones will end up relying on it. We must push students towards the former.

The Shift from Output to Understanding

Which subject will be least impacted by AI tools? My guess is PE. If you rock up at the pitch without having trained, it’s going to be very obvious to everyone. In-person assessment is key. Great, so we should just have students sit exams every class and then job done? That would be a massive waste of teachers’ time and not very engaging. Students already put more effort into subjects they enjoy, so making it enjoyable should be a priority, as it means they are less likely to check out and just rely on AI tools.

Instead of searching for a new and innovative solution, why not adapt one from the past? Oral assessment. This (in the UK at least) is a form of assessment that is rarely used anymore but is the most effective way of determining whether someone has actually understood something. I believe in a two-minute conversation I could get a better idea of whether someone understands something than in a three-hour exam.

One example of this could be assigning a topic for students to research as homework, and then in class there is a discussion within the class led by the teacher who is responsible for assessing whether the students have engaged with the topic appropriately. This is, of course, heavily dependent on the school and the teacher to assess fairly and correctly (I have some ideas of how this could be standardised somewhat).

Why should we do this?

Many things that we currently test for will no longer be relevant in the future. However, core fundamentals will always be relevant, most critically metalearning. The ability to learn is the most important thing to impart to students; the ability to solve problems and utilise information effectively are also key. Even if many tasks will ultimately be delegated to AI, we must be able to ask the right questions. I believe that by focusing on understanding and not memorisation or repetitive work we can hone in on whether students are really being equipped for the future. AI, like any technology, is capable of providing amazing benefits and incurring terrible costs; it is up to us to decide how we use it.

This article was 100% written by me, a human.

Subscribe now

Test Article

✨Myrddin✨ — Tue, 19 May 2026 20:11:29 GMT

Ignore this