Flagship product

Teaching Observation with AI

Observation where every rating has to cite what actually happened in the room — and the teacher can answer back on the record.

Direct answer
Teaching Observation with AI produces short, evidence-anchored classroom observations scored against your district’s evaluation rubric, in which every rating requires cited evidence from the observed lesson.

It aligns to Danielson, Marzano and state frameworks including T-TESS, TKES, TEAM, NEPF, TPEP, APPR, OTES and Compass, as well as district-built models. Hedged language is stripped from feedback, coaching prompts are job-embedded, and the teacher holds a documented right of reply.

Why most observation produces nothing

American teacher evaluation has a well-documented distribution problem: in most systems the overwhelming majority of teachers are rated effective or better, which means the instrument is not discriminating and the ratings carry almost no information. The result is a process that consumes enormous administrator time and changes very little practice.

The diagnosis matters, because the usual fixes make it worse. Tightening the rating distribution turns observation into a compliance threat and teachers stop taking risks in observed lessons. Adding more observations without changing what they contain just multiplies a low-value artifact.

What actually changes practice is specificity: feedback that names something concrete that happened, at a specific moment, and states what to do differently. That is a design problem, and it is the one this product addresses.

How it works

Evidence is required, not optional

An observer cannot record a rating on a rubric component without attaching evidence from the observed lesson — what was said, what students did, what was on the board, a timestamped note. A rating with no evidence does not save. This single constraint is what separates useful observation records from a checkbox exercise.

Hedging is stripped out

Draft feedback is checked for the language that makes observation feedback useless: “perhaps consider,” “you might want to think about,” “generally good but.” The system surfaces these and asks the observer to state the specific action instead. Ambiguous praise and ambiguous criticism are both removed.

Coaching prompts are job-embedded

Each rubric component links to next-step actions and short professional development the teacher can act on before the next lesson, rather than a course scheduled for a PD day in March.

The teacher has a right of reply

Every observation record carries a teacher response field that becomes part of the permanent record. This is not a courtesy. It is what makes a record defensible if it is ever contested, and it is what makes teachers willing to be observed honestly.

Short and frequent beats long and rare

The product is built around brief walkthrough observations conducted often, rather than a formal annual observation. Ten short observations across a year produce a trend; one produces an anecdote.

Framework alignment

Edves does not impose a rubric. It runs on the one your state or district already uses.

FrameworkWhere used
Danielson Framework for TeachingWidely adopted nationally and as the basis of many district models
Marzano evaluation modelAdopted in numerous districts and several state systems
T-TESSTexas
TKESGeorgia
TEAMTennessee
APPRNew York
OTESOhio
CompassLouisiana
NEPFNevada
TPEPWashington
PERAIllinois
SEEDConnecticut
District-built modelsConfigurable rubric builder for local frameworks

Full list and structure: teacher evaluation frameworks.

What administrators get

  • Trend data by teacher, grade, department and campus — where practice is strong, where it is thin, and whether coaching moved anything.
  • Inter-rater reliability signals — whether two assistant principals score the same lesson the same way, which is usually the first thing to fix in a district evaluation system.
  • Coverage of the observation cycle — who has been observed, who has not, and where the calendar is slipping, before it becomes a compliance problem.
  • Link to instruction — observation records connect to the lessons and standards from Pedagogy with AI, so a conversation about instructional rigor can reference what was actually planned.

What teachers get

  • Feedback that names something specific rather than a number.
  • A next step small enough to try on Monday.
  • A record they can respond to, and a history that shows growth rather than a single annual verdict.

Where AI is and is not used

This distinction matters for union conversations and for district counsel, so it is worth being exact. AI in this product drafts feedback language, surfaces relevant rubric components from the observer’s evidence notes, flags hedged phrasing, and suggests coaching resources. It does not assign ratings, does not make employment recommendations, and does not observe autonomously. A human observer is in the room and a human observer scores the rubric.

Districts adopting AI-supported observation should say this explicitly in their evaluation handbook and in any collective bargaining discussion. Ambiguity here is the fastest way to lose staff trust in the whole system.

Implementation

StageFocus
1Load the existing rubric; do not change the instrument in year one
2Calibrate observers — multiple administrators score the same recorded or live lesson and compare
3Run short walkthroughs only, framed explicitly as developmental and non-evaluative
4Introduce trend reporting once enough observations exist to be meaningful
5Connect to formal evaluation only after staff trust the developmental use

Districts that begin at stage five get compliance and no practice change.

Higher education

The same engine runs for colleges and universities as Spot Observation, with frameworks appropriate to postsecondary teaching. See higher education.

Common questions

Frequently asked

Which teacher evaluation frameworks does Edves support?

Danielson, Marzano and state frameworks including T-TESS (Texas), TKES (Georgia), TEAM (Tennessee), APPR (New York), OTES (Ohio), Compass (Louisiana), NEPF (Nevada), TPEP (Washington), PERA (Illinois) and SEED (Connecticut), plus a configurable rubric builder for district-built models.

Does AI assign the teacher's rating?

No. AI drafts feedback language, surfaces rubric components from the observer's evidence notes, flags hedged phrasing and suggests coaching resources. A human observer is in the room and a human observer scores the rubric. AI makes no employment recommendations.

What does 'evidence-anchored' mean?

An observer cannot save a rating on a rubric component without attaching evidence from the observed lesson: what was said, what students did, what was displayed, or a timestamped note. Ratings without evidence do not save.

Can teachers respond to an observation?

Yes. Every observation record includes a teacher response field that becomes part of the permanent record, which is what makes the record defensible if contested and what makes honest observation possible.

How does this help with inter-rater reliability?

The platform surfaces scoring variance between observers on comparable lessons, which lets a district see whether two administrators evaluate the same practice the same way. In most districts this is the first thing worth fixing.

Is this compatible with our collective bargaining agreement?

Edves runs on the rubric and observation cycle your district already negotiated rather than imposing its own. Districts should still state explicitly in the evaluation handbook where AI is and is not used, and Edves provides that language during implementation.

See it run on your own standards.

A 20-minute walkthrough using your state’s standards, your evaluation framework and your reporting requirements.

Schedule a demo

Keep reading