Look under the hood of the new Structural Learning platform. Five engines run every time a teacher presses generate — extracting meaning from the topic, finding inquiry angles, sequencing the thinking, aligning the curriculum, and predicting what children will do. Here's how each one works.
Before we walk you through this, a quick honest note. Many of you tried to generate a lesson with us over the past few months and the platform simply fell over. It wasn't a design decision — it was an infrastructure one. Too many teachers hit the engine at once, and we didn't have the capacity to keep up. So we rebuilt it from the ground up.
This piece is your invitation to try it again. It's still in beta — we're sharpening the format and the output every day, and it isn't perfect — but it works, and it's better every week. Take a look at what now runs under the hood when you press generate.
Behind every generated lesson, five small engines run in sequence. None of them is magic. Each one is a deliberate choice that you would probably make yourself if you had two hours instead of ten minutes. The platform's job is to do them in order, in the right way, in less time than it takes to make a cup of tea.
When you type "photosynthesis" into the wizard, the platform doesn't search a database of pre-built lessons. It reads the topic for its conceptual structure. What ideas does this topic actually contain? What is the energy story underneath it? Which inputs, outputs, and scales matter?
For photosynthesis, four concepts surface immediately: energy transfer, the relationship between inputs and outputs, the role producers play in a food web, and the change in scale from a single leaf to a whole ecosystem. These are the things worth teaching with. Concepts are the unit of planning, not topics. This is why a lesson on photosynthesis built by the platform looks different from one built around the keyword alone.
This stage is also where the platform notices the difference between, say, a Year 5 science lesson on photosynthesis and a Year 10 biology lesson on the same topic. Same word, different conceptual depth. Concept extraction is the bit that makes the lesson feel age-appropriate without you having to spell that out. If you want to read more about why concepts are the unit we plan with, our piece on concept-based learning walks through the pedagogy.
Concepts on their own don't make a lesson. You need a question that gets children curious about the concept. So the next engine generates inquiry angles — four routes the lesson could take, each one a different cognitive door.
For energy transfer, the platform might offer: "What if a plant lost all its leaves overnight?" (a counterfactual), "How does a tree feed itself without a mouth?" (a mechanism question), "Why are most plants green and not blue or red?" (a cause question), or "How is a leaf like a tiny factory?" (an analogy). You pick the one that fits your class. The rest stay parked for the next time you teach the topic, so the second lesson is not a copy of the first.
This is where the Thinking Framework comes in. The framework is a four-colour shared vocabulary for cognitive work in lessons. Green is Know It — retrieval and prior knowledge. Orange is Define It — vocabulary precision. Blue is Map It — organising and externalising thinking. Red is Apply It — the higher-order step where children stretch and explain.
Once you have picked an inquiry angle, the platform sequences the thinking children will do across the lesson. It also runs pedagogical rules underneath. A Red step always needs a Green step behind it, and either a Define It or Map It before it. No two consecutive steps share the same colour. Year 5 caps at two Red steps; EYFS doesn't have Red at all. You don't have to think about those rules — the engine simply won't generate a lesson that breaks them.
This is the stage that surprises teachers most. Every thinking step in the sequence is matched against the curriculum your school teaches. Over 32,000 curriculum objectives are indexed across the major frameworks — UK National Curriculum, Cambridge Primary and IGCSE, the three IB programmes (PYP, MYP, DP), British Columbia K to 9, and AQA GCSE. International colleagues get their own framework matched in the same lesson.
When the platform builds your photosynthesis lesson, it is not guessing. It matches "recall what plants need to grow" against UK NC Y5 Science 2b. It matches the inputs-and-outputs diagram against 3a. It matches "explain a real plant in trouble" against 4c. You see the objective codes in the teacher's notes column, so your planning folder can carry the evidence straight into the next book scrutiny or curriculum review.
The last stage is the one parents and inspectors see. The platform predicts the artefacts the lesson will produce — the diagram in the book, the sequence on the table, the explanation overheard between two children. Every Map It, Build It, and Say It mode produces something visible. That visibility is the assessment loop. You can mark a diagram in seconds. You can hear an explanation in ninety seconds. You can photograph a sequenced set of cards before you collect them back in.
The platform also generates the teacher-side outputs alongside the children's: a printable lesson page, a key vocabulary panel, teacher's notes, and a slide deck for the whiteboard. Everything stays in your account, so the lesson is there next year when the same topic comes round. Our piece on graphic organisers covers more of the artefact side, if you want to see the kinds of outputs we are designing for.
Plenty of "AI lesson planners" exist now. Most of them treat the topic as a prompt and ask a model to write what it can. What you get is a lesson that is grammatically correct but pedagogically random. Sometimes there is prior knowledge activation; sometimes there isn't. The vocabulary might be at the right level; it might not. There is no system underneath.
We wanted the system to be the bit teachers stop having to do. Not the writing itself — most teachers can write a clear lesson page in fifteen minutes if they are not also planning the cognitive arc. The bit that takes the hour is making sure the lesson moves children from prior knowledge to new understanding, with the vocabulary scaffolded, the thinking sequenced, and the curriculum quietly aligned in the background. That's the architectural work, and that's the part the platform now does. You stay in the driving seat, but the cognitive load on the planner drops sharply. If you want the longer version of why planning matters this much, our lesson planning piece sets out the case.
The platform is in early beta. It changes most weeks based on what teachers tell us is missing. If you haven't taken it for a spin yet, the easiest entry point is to pick a lesson you'd rather not plan from scratch — something you've taught before and don't enjoy preparing — and use the wizard for that one.
Four minutes from now, you'll know whether the system saves you time, or whether your version is still better. Either answer is useful to us. If something is missing, tell us. Reply to this email and the message lands with the small team that builds the platform. We read every one.
