It differs from general-purpose AI use because every output is anchored to two things: an explicit instructional model such as Bloom’s taxonomy or Depth of Knowledge, and a named curriculum standard such as a TEKS student expectation or an NGSS performance expectation. The teacher edits and signs off, and the system records that they did.
The problem in U.S. classrooms
Generative AI reached American teachers before any district had a policy for it. Two failure modes followed within a semester.
In the first, teachers use a consumer chatbot to generate lesson material that reads well and teaches badly — six activities at the same cognitive level, no retrieval practice, no check for understanding, and no traceable relationship to the standard the teacher is accountable for. Fluency is exactly what makes this hard for an instructional coach to spot.
In the second, the district bans the tools, teachers use them anyway on personal devices, and student data ends up in a consumer product with no contract behind it.
Pedagogy with AI is the narrow middle position: the model drafts, the teacher decides, and the record shows who decided.
Definition
Pedagogy with AI is the use of artificial intelligence to design, adapt and evidence instruction while the teacher retains professional judgment and accountability. Three conditions separate it from unstructured AI use:
- Model-anchored. Output is generated against a stated instructional model — Bloom’s taxonomy, Webb’s Depth of Knowledge, Understanding by Design, the 5E model, SOLO — not to a free-text prompt.
- Standard-anchored. Output maps to a specific standard code, so a lesson traces to the exact expectation it serves and coverage gaps become visible at the objective level.
- Teacher-owned. The teacher edits, approves and signs off, and the approval is recorded. Accountability does not transfer to the model.
How it differs from adjacent products
| Approach | What the AI does | Who is accountable |
|---|---|---|
| AI tutoring | Interacts directly with the student | Ambiguous — the model mediates learning |
| Adaptive learning | Sequences content by prior performance | The algorithm sets the path |
| Consumer chatbot use | Produces material on request | Teacher, with no structure to check against |
| Pedagogy with AI | Drafts against a stated model and standard code | The teacher, explicitly and on the record |
What it looks like in a U.S. classroom
1. Lesson design against the standard you are accountable to
A teacher selects a standard — a TEKS student expectation, an NGSS performance expectation, a Georgia Standards of Excellence element, a B.E.S.T. benchmark — plus a class and an instructional model. Edves drafts a lesson sequence with the cognitive demand of each task tagged, differentiation for the prior-attainment spread in that specific section, IEP and 504 accommodation prompts where flags exist on the roster, and the formative check that will evidence the standard.
The teacher edits it. What ships is the teacher’s lesson.
2. Real-time adaptation
During instruction, responses to checkpoint questions indicate whether the class has met the expectation. The system proposes an adjustment — reteach with a different representation, extend, or move on — and the teacher chooses.
3. Evidence for observation and MTSS
Because lessons carry standard codes, the record produced is the evidence an instructional coach, an MTSS team or an evaluator would otherwise have to assemble by hand. See Teaching Observation with AI.
4. Assessment that closes the loop
Items generated against the same standard code and cognitive level, with results feeding back into what the next sequence must address. See assessment.
The instructional models
Edves operationalizes a defined set of 14+ models rather than treating pedagogy as a free-text field. Each carries its own task shapes, scaffold logic and rubric structure.
| Model | What it structures |
|---|---|
| Bloom’s taxonomy | Cognitive-level tagging across tasks |
| Depth of Knowledge | Complexity calibration against state assessment demand |
| Project-based learning | Scaffolded milestones and rubric feedback |
| Problem-based learning | Scenario-driven scaffolds |
| Inquiry-based learning | Questioning sequences and investigation design |
| Experiential learning | Hands-on cycles with structured reflection |
| Competency-based learning | Mastery tracking and personalized progression |
| Design thinking | Iterative prototyping and peer critique |
| Spaced learning | Retrieval scheduling for retention |
| Micro learning | Short bursts with immediate feedback |
| Blended learning | Synchronous and asynchronous rotation |
| Gamified learning | Engagement mechanics tied to mastery, not activity |
| Special education pedagogies | Scaffolds for autism spectrum, dyslexia and ADHD profiles |
| Assessment for learning | Formative cycles with next-step planning |
| Career pathways | CTE-aligned progression with advising analytics |
Why the anchoring is the whole point
An unanchored model produces a plan that reads like a good lesson and is not one. Anchoring to a model and a standard code turns a plausibility engine into something a department chair can audit — and turns lesson planning into the coverage evidence a district needs anyway.
It also protects the teacher. When an evaluation conversation turns to whether a standard was taught, a record showing which lesson addressed which expectation, and that the teacher approved it, is a better position than a folder of documents and a memory.
Data protection
Student information used to personalize lesson design stays within the district’s tenant, is not used to train general-purpose models, and is subject to the same role-scoped access and audit logging as the rest of the platform. Full detail: FERPA and student data privacy.
Getting started
Most schools begin with lesson design alone, because the time recovery is immediate and the risk is low. Observation and CPD follow once staff trust the framing. Assessment comes last, because it touches the most stakeholders. Attempting all three in one semester is the single most common reason implementations stall.
State-specific detail: Texas, California, Florida, New York, Georgia, or all 50 states.