A two-layer framework for building AI tutors that actually teach.
Most AI-powered learning tools do one of two things: they answer questions like a search engine, or they deliver scripted content like a slideshow with a chatbot bolted on. Neither one teaches. They give you information and hope learning happens on its own.
That's like handing someone a textbook and calling it a teacher. Information is not instruction.
ScaffoldedAI is a framework for building AI tutors grounded in established learning science — tutors that guide learners through understanding rather than handing them answers. These tutors use Socratic questioning, check for misconceptions, adapt to what the learner already knows, and build knowledge step by step.
The framework is built on a two-layer architecture:
-
Layer 1: Pedagogy. A universal set of instructional principles that govern how the tutor teaches — things like Socratic dialogue, scaffolded complexity, formative assessment, and productive struggle. This layer stays the same no matter what you're teaching.
-
Layer 2: Domain Content. The subject matter, learning objectives, learning path, common misconceptions, and key concepts for a specific topic. This is the layer you swap out to create a new tutor.
The result: you can build a tutor for any subject by writing a Layer 2 document and pairing it with the universal Layer 1. No coding. No AI expertise. Just clear thinking about what you want to teach and how learners typically struggle with it.
- A Claude Pro or Team account with access to Claude Projects
- About 15 minutes to set up your first tutor
- Curiosity about how AI can do more than answer questions
If you want to see what a ScaffoldedAI tutor feels like before building anything, start here. We've built three sample tutors that demonstrate the framework across different types of learning:
| Sample Tutor | What It Teaches | What It Demonstrates |
|---|---|---|
| Impressionism & Post-Impressionism | Key movements, artists, techniques, and the historical shifts between them | Teaching conceptual knowledge — how the tutor builds understanding of ideas and connections |
| Photography Exposure Basics | Aperture, shutter speed, ISO, and depth of field | Teaching an applied skill — how the tutor helps you learn settings that interact with each other |
| Logical Fallacies | Common reasoning errors and how to spot them in real arguments | Teaching analytical reasoning — how the tutor presents flawed arguments and guides you to find the problem |
Each sample includes everything you need: setup instructions, the Layer 1 and Layer 2 files, and suggested first prompts to get started.
Ready to create a tutor for your own subject? The Tutor Builder is itself a Claude Project — you give it your source material (documentation, guides, training manuals, textbooks, or even just your own expertise written out), and it helps you generate a Layer 2 document tailored to your content.
Two ways to start:
-
Use our starter kit. We've included a sample source document so you can walk through the full build process with guided content before using your own. Think of it as a practice run.
-
Bring your own content. If you already have documentation or subject matter you want to teach, skip the kit and jump straight in. The Tutor Builder will walk you through everything.
-
Start with Layer 1. This file contains the pedagogical instructions that tell Claude how to be a tutor. You add it to a Claude Project as part of the project instructions. You never need to modify this file.
-
Add a Layer 2. This file contains your subject matter — the learning objectives and path, key concepts, common misconceptions, and anything else the tutor needs to know about your topic. You add this to the same Claude Project.
-
Start a conversation. Open a new chat within the project and begin learning. The tutor will assess where you are, guide you through the material, and adapt to your responses.
That's it. Two files in a Claude Project. No code, no plugins, no platform to learn.
ScaffoldedAI was created by Marion Jensen, an author and instructional designer with 25+ years of experience building learning programs at companies like American Express, Discover, Brex, and Milliman. Marion's work sits at the intersection of instructional design and AI — exploring how large language models can move beyond content delivery into genuine, adaptive teaching.
This framework grew out of real-world deployment: AI tutors built using this architecture have been used for employee onboarding, professional development, and skills training in live corporate environments.
Marion is currently pursuing a PhD in Instructional Technology & Learning Sciences at Utah State University, with a research focus on AI-facilitated learning.
This project is licensed under the Creative Commons Attribution 4.0 International License. You are free to use, adapt, and share this framework for any purpose, including commercially, as long as you provide attribution.