Differentiation was always the hour nobody paid you for.

AI-assisted differentiation means the teacher decides what each group needs and the tool produces the versions. The decision is the teaching. The production is the hour that disappears on a Sunday. Keeping those two apart is what separates a differentiation tool that helps from one that quietly starts deciding what a child can handle.

What does AI-assisted differentiation actually mean in practice?

A teacher plans one lesson and needs four artefacts: the core task, a version at a lower reading age, a version with scaffolding for the pupils who need the structure made visible, and an extension for the three who will finish in ten minutes. The intellectual work is knowing which pupils need which, and why. The remaining work is retyping.

A tool that takes the core task and the teacher's instruction and returns the other three, in the same format, ready to be corrected, has removed the retyping. It has not touched the judgement. That is the whole of the useful case, and it is a bigger case than it sounds, because retyping is where the evenings go.

Which parts of differentiation should stay human?

Deciding what a pupil is capable of, above all. A system that infers a child's level from their previous output and quietly serves them easier work is not differentiating, it is lowering expectations with an audit trail. Guidance already rules out AI making judgements about attainment, and this is the same principle wearing different clothes.

• The teacher decides the grouping. The tool does not infer it from past performance.
• The teacher decides what is simplified. Reading age is a formatting change, not a curriculum change, and the two are easy to confuse.
• The teacher checks the science, the history and the maths. Adapted material introduces errors precisely where the adaptation happened.
• The teacher keeps the ceiling. Every version should be able to reach the same learning objective by a different route.

Does personalised learning with AI actually help pupils?

The honest answer is that the evidence is thinner than the marketing. Adaptive systems have been sold to schools for a long time on outcome claims that did not survive independent evaluation, and the current generation is younger than its claims. Anyone quoting you an attainment uplift from a tool this new is quoting something they cannot support.

What can be said with confidence is narrower and still worth having: producing differentiated material takes teachers a great deal of time, and a tool that produces it faster gives that time back. Whether the resulting differentiation improves outcomes depends on the teaching around it, which is where it always depended.

What to check before you trust a differentiated resource

Read the simplified version first, not the original. Adaptation is where meaning gets lost, and the pupil receiving the adapted version is usually the one least able to tell you it stopped making sense. Check that the learning objective survived, that no factual claim was introduced, and that the scaffolding supports thinking rather than replacing it.

The Teaching and Planning module of Aevum Teacher Manager is in integration, producing differentiated resources with multimedia support, on the school's own hardware, for the teacher to review. In development. No outcome claims, because we have not earned any yet.