Will AI replace teachers? No. But it will change the argument.

No. Teaching is a relationship carrying a curriculum, and the relationship is not the removable part. What AI can take is a specific and substantial slice of the work around the lessons: the drafting, the retyping, the paperwork. That is worth having, and it is not the same thing as replacement.

The reason the question keeps being asked is not that teachers have misunderstood the technology. It is that they have correctly noticed that the question of what happens next is not being decided by them.

What is actually at risk?

Not the profession. Something more specific: the scope of professional judgement, and who gets to keep it. A tool that drafts a report has taken a task. A system that produces something that looks like a grade, which a tired school then uses as one, has taken a decision. The first is help. The second is the erosion people are worried about, and it happens by drift rather than by announcement.

There is also a workforce question that is genuinely open, and it is not about whether teachers exist. It is about whether the time released is returned to teaching, or absorbed as an expectation that fewer adults can now cover more. That is a decision made by schools and by policy, not by a model.

What does the guidance say about AI and professional judgement?

It holds the line, at least on paper. Department for Education guidance places responsibility for professional judgement, safeguarding, data protection and curriculum quality with the school, rules out AI making summative judgements about a pupil's attainment, and forbids it writing a statutory EHCP or deciding whether a child has a need.

Those are useful sentences to know by heart, because they are the ones to quote when a product is being demonstrated to your senior leadership team and nobody in the room is sure where the line sits.

Why is the scepticism reasonable?

Because teachers have been here before. Interactive whiteboards, a decade of learning platforms, several waves of adaptive software, each arriving with outcome claims that did not survive independent evaluation, each absorbing training time that was never returned. A profession that has been promised transformation five times and delivered a login screen four times is not being irrational.

The correct response to that is not more enthusiasm. It is evidence, published with its protocol, and a supplier willing to say what their tool cannot do. We publish the protocol before the result for this reason, and we have withdrawn a figure we were proud of when we found the measurement behind it was broken.

What should a teacher do about it now?

Learn enough to be in the room when the decisions get made, which is a different goal from becoming an AI expert. The teachers whose working conditions will be best protected are the ones who can ask a supplier where the data goes and know whether the answer was any good.

• Know the three things guidance already forbids. They are your strongest argument and they are free.
• Try a tool on a task you know well, so you can see where it fails rather than being told.
• Ask where the data goes, every time, until suppliers expect the question.
• Push for the released time to be named. Time that is not accounted for is time that will be reallocated.

Aevum is built by someone who taught for twenty years, first in Portugal and then in England, and the design assumption throughout is that the tool drafts and the teacher decides. Everything is in development. Own the model. Keep the data. Inspect the reasoning.