Small-Group Instruction at Scale, with an AI Planning Partner
Differentiation, mastery-based learning, and small-group instruction reliably help students — and reliably cost teachers time they don't have. Claude for Teachers doesn't replace the practice; it becomes the planning partner that makes it feasible.
Education has a strange problem for a field so anxious about evidence: we already know a lot about what works. Small-group instruction. Differentiation across readiness levels. Mastery-based progression, where students move on when they've actually got it rather than when the calendar says so. These are not fringe theories or the latest app. They are durable, research-backed practices that reliably move learners — and they have been sitting in teacher-prep courses for decades.
So why doesn't every classroom run this way? Because knowing a practice works is not the same as having time to run it. Every one of those practices is expensive in the currency teachers are poorest in — planning hours. Small groups need differentiated plans for each group. Differentiation needs multiple versions of the same lesson. Mastery needs constant reassessment and re-grouping. The research says do it; the week says you can't. That gap — between proven practice and feasible practice — is the specific thing Claude for Teachers is built to close, and it closes it by acting as a planning partner rather than a substitute teacher.
The practices we already know work
There is a reason the launch names these three by name. Differentiation, mastery-based learning, and small-group instruction are among the most reliable moves a teacher can make, and they share a common logic: they meet students where they actually are instead of where the average student is assumed to be. A whole-class lecture aimed at the middle is, by construction, mistimed for most of the room — too slow for some, too fast for others, exactly right for a lucky few. The evidence-backed practices break that compromise by shrinking the unit of instruction: a small group, a tier, an individual's next competency.
None of this is controversial among teachers. Ask any of them whether four kids at a table with a purpose-built task learn more than thirty kids facing a whiteboard, and you'll get a tired "obviously." The disagreement was never about whether it works. It was about whether a single human being can plan and run five differentiated groups, reassess them weekly, and still have a life. For most teachers, most weeks, the honest answer has been no.
"Known to work" is not the same as "happening"
Here is the quiet tragedy of good pedagogy: the practices with the strongest support are also the most labor-intensive, so they are the first to get cut when time runs short. A teacher under pressure doesn't abandon differentiation because they stopped believing in it. They abandon it because it's Tuesday, three preps are due, and the differentiated version of tomorrow's lesson would take two hours they simply do not have. The default — one lesson, one level, aimed at the middle — wins not on merit but on cost.
The best practices in teaching are the ones that scale worst by hand. That is not a coincidence to shrug at — it is the entire bottleneck. Anything that lowers the planning cost of a proven practice changes how often it actually happens.
That reframes what "help" even means. The teachers running these practices already know how. They don't need to be taught differentiation; they need the differentiated materials to appear without costing an evening. The intervention that matters is not another workshop on why small groups are good. It's a partner that makes the planning cheap enough that the good practice survives contact with a real week.
The bottleneck is planning, not teaching
It's worth being precise about where the time goes, because that's where a tool can and can't help. Look at each practice and the pattern is identical: the classroom execution is human work that no one wants to automate, but the preparation is repeatable, structured labor that eats hours.
| Practice | What it reliably produces | Where the planning time goes | What an AI planning partner takes on |
|---|---|---|---|
| Small-group instruction | Instruction matched to a handful of similar learners | Grouping students, then designing a distinct task per group | Drafting per-group tasks from one lesson; suggesting groupings from class data |
| Differentiation | The same core content at below / at / above proficiency | Rebuilding one lesson into multiple leveled versions | Producing tiered versions that hold the core content constant |
| Mastery-based learning | Progress gated on demonstrated understanding | Reassessing constantly and re-planning what comes next | Mapping the next competency and drafting the reteach or extension |
Notice what stays with the teacher in every row: the judgment, the relationships, the read of the room. What moves to the partner is the mechanical multiplication — turning one plan into five, one lesson into three tiers, one diagnostic into a next step. That division of labor is the whole design.
An AI planning partner, not an AI teacher
This is the distinction that keeps the practice healthy. Claude for Teachers is deliberately a tool for teachers, not a tool aimed at students — a choice grounded in the evidence that student-facing AI is mixed and implementation-dependent, while teacher-facing AI can strengthen instruction. The small groups are still taught by the teacher. The partner's job is upstream: make running them feasible.
In practice that looks like handing Claude a folder of class data — roster, diagnostics, attendance, notes — and letting it build a picture of where each student stands, so grouping starts from evidence instead of a hunch. You control exactly what's shared, and nothing shared is used for model training. From that picture, you ask for the tiered lesson, and it drafts the differentiation plan plus per-proficiency student materials for you to revise. Because the standards and their underlying learning progressions come in through the Learning Commons connector, the groups' tasks come out aligned and scaffolded rather than improvised. A helper like lesson-plan-studio, or a fuller kit of education agent skills, does the analogous work when you're building this stack yourself.
Small groups become schedulable
The last piece is that this doesn't have to be a one-off heroic effort. Because Claude for Teachers includes Claude Code and Cowork, the recurring parts of running small groups can be scheduled and handed off. A daily pass over exit tickets can flag which students should move groups tomorrow. A weekly job can regenerate the leveled materials for next week from this week's diagnostics. The work carries forward on its own, which is what turns small-group instruction from a practice you attempt in your best weeks into a practice you can actually sustain.
That sustainability is the real unlock. We were never short on knowledge about what helps students. We were short on the hours to deliver it. Give a teacher a partner that absorbs the planning multiplication, and the proven practices stop being aspirational bullet points in a district plan and start being what Tuesday looks like. From here it's worth reading how the same partnership supports mastery-based learning, and the concrete mechanics of the differentiate-for-every-learner workflow. The research already told us where to go. This is finally a way to afford the trip.
Part of the Claude for Teachers series. Related: Mastery-Based Learning with an AI Partner · Differentiate for Every Learner in the Room. Browse AI tutoring skills or more builder insights.