Differentiation is the promise on the box of every AI teaching tool, and Claude for Teachers makes it better than most: its teaching skills will happily tier a task three ways, suggest scaffolds for your EAL students and draft extension questions for the fast finishers. But there is a hard limit built into the format, and it is worth naming precisely. An assistant can give you one document describing thirty adjustments. It cannot hand your thirty students thirty different things to do. That second thing - one consumable lesson per student - is what actually collapses the differentiation workload, and it is what a generator produces that an assistant structurally cannot.
The two bottlenecks assistants don’t remove
Ask any teacher why full differentiation rarely happens and you get the same two answers, and neither is a shortage of ideas.
- Production. Thirty students need thirty artefacts, or at least five or six genuinely different ones. Claude for Teachers will describe them beautifully, but each described version still has to be prompted, formatted and produced by you. The advice is instant; the artefacts are still hand-made.
- Marking. Every extra version you produce is an extra version to check. Differentiate a worksheet five ways and you have quintupled your marking surface. This is the quiet reason one-size-fits-all worksheets survive: not because anyone believes in them, but because uniformity is the only thing that keeps marking sane.
One lesson per student changes the unit of work
A consumable lesson - single-serve, generated for one student, consumed directly by that student - attacks both bottlenecks at once. Production stops being your job: describe the concept once, name a different interest and level per student or group, and generate. Each student gets the same curriculum outcome wrapped in their own world, which is the most reliable engagement lever we know of. And marking stops scaling with versions, because the practice in every version checks itself and gives hints as the student works. Five versions, ten versions, thirty versions: the marking load is the same, which is to say, near zero for the practice layer.
This is the argument we made in differentiated instruction at scale taken to its conclusion: differentiation was never a knowledge problem, it was a manufacturing problem, and generation solves manufacturing.
What this looks like in practice
Take a mixed Grade 4 class on equivalent fractions. In Sprout you write the concept once and generate: a lesson set in netball scores for one student, baking measurements for another, Pokémon card trades for a third; a version with more worked examples for the two students who are still consolidating, and a stretch version for the one who is ready for it. Each is a live interactive page or a ten-slide deck, each is aligned to the same outcome in your chosen framework - Australian, US or Canadian - with the codes recorded in the footer, and each marks itself. Generating the set takes minutes, not a weekend.
Keep the assistant for the thinking
None of this makes Claude for Teachers redundant, and the fair comparison is spelled out in Sprout Lessons vs Claude for Teachers. Deciding how to differentiate - reading the diagnostic data, spotting who needs what, planning interventions - is exactly the kind of judgement-heavy work a frontier assistant supports well, and for verified US teachers it is currently free. The division of labour that works is: assistant for the differentiation strategy, generator for the thirty artefacts the strategy calls for.
If the production bottleneck is the one keeping your differentiation theoretical, try Sprout free and generate the same concept three ways for three real students. The difference is easier to feel than to read about.
FAQ
Can Claude for Teachers differentiate lessons?
It differentiates plans very well: its teaching skills can tier a task, suggest scaffolds and draft extension work. But its output is a document describing the adjustments. Each described version still has to be prompted, formatted and produced by the teacher, and every extra version produced is an extra version to mark. The strategy is instant; the artefacts and the marking remain manual.
Why does differentiation usually fail in practice?
Two bottlenecks, neither of which is a shortage of ideas: production (thirty students need many genuinely different artefacts, which are slow to make by hand) and marking (every additional version multiplies the checking load). Uniform worksheets persist not because teachers believe in them but because uniformity keeps marking manageable.
How does one lesson per student become practical?
Generation removes the production cost: describe the concept once, name a different interest and level per student or group, and each student gets an interactive lesson on the same curriculum outcome wrapped in their own world. Self-checking practice removes the marking cost: whether there are five versions or thirty, the practice layer marks itself with hints, so the marking load stays flat.
Is Sprout Lessons a replacement for Claude for Teachers?
No, they split the job. Claude for Teachers (free for a year for verified US K–12 educators) is strong at the judgement-heavy half: reading diagnostic data, deciding who needs what, planning interventions. Sprout Lessons produces the thirty artefacts the strategy calls for: consumable, curriculum-aligned, self-marking lessons, one per student, in any of seven frameworks across Australia, the US and Canada.