Education AI agent · Tutoring
Tutoring Copilot AI Agent
Explain coursework at the help level the teacher of record sets, ground each explanation in the adopted materials, and treat a student's accommodations as a limit the agent cannot move.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Teacher-owned
- Formative only
- After launch
- Agent Care
What this agent does
Explains the work, not what it is worth
In
- 01
Ingest the adopted materials, the scope and sequence and the accommodations record from supported LMS.
- 02
Normalise each passage and carry it forward with the approved source it came from.
Reason
- 03
Explain the concept at the help level the teacher configured for that assignment.
- 04
Apply the standards, the adopted scope and sequence and the integrity ceiling the institution set.
- 05
Read the accommodations on file as a constraint, and raise a request to the case manager rather than varying one.
Decide
- 06
Render mathematics, diagrams and code as text a screen reader can speak, not as unlabelled images.
- 07
Route a contested claim, an integrity question or an accommodation request to the teacher of record.
Out
- 08
Retain the session and the work it produced for the school, and put none of it to training, profiling or advertising.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent explains and practises; the teacher of record owns what is taught and what it is worth, and the school keeps direct control of the records a session creates.
Example workflow
One session, prompt to hand-off
- 1Student prompt receivedLMS assignment, courseware activity or a tutoring session
- 2Prior work assembledAdopted materials, the standard in play and the accommodations on file, each with its source
- 3Explanation draftedWorked explanation, practice items, the help level applied and confidence
- 4Scope checks appliedSource checks, standard and scope rules, help-ceiling and accommodation checks and confidence threshold
No human action required
The first four stages run unaided, and nothing is graded at any of them — the agent is explaining, and the teacher's lane opens at the confidence gate.
Goes to the student at the configured help level.
Waits for the teacher of record first.
Teacher review
The explanation is held with its cited source, its flags and the confidence.
Approve · Edit · Send to the teacher
Approved — released to the student
- 6Courseware and dashboards updatedOnly where write access and approval policy allow it
- 7Session evaluatedEdit distance, off-standard rates, help-ceiling breaches and corrections after the fact
Reframings
Every teacher edit is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Changing an instructional accommodation in an individualised plan.
- Assigning or recording a grade or mastery determination of record.
- Deciding whether help crossed into academic misconduct.
- Approving instructional content outside the adopted materials.
Automation boundary
Agent acts unaided
- ✓Generate practice items inside the teacher's standard for the named owner.
- ✓Explain a concept at the configured help level.
- ✓Give formative feedback on a draft that nothing.
- ✓Log time on task and item-level performance to the teacher's.
Any write happens inside the boundaries agreed at implementation, never ahead of the teacher.
- Altering the scope, the sequence or the pacing of the course.
- Determining that a student needs evaluation for a disability.
- Telling a parent that a student is struggling, as an assessment.
- Resolving a conflict between the tutor's answer and the teacher's.
Example output
One explanation, annotated
Everything the agent explains is attached to the material it was drawn from.
Tutoring output · single session
Illustrative example- Assignment
- Quadratics practice
- Explanation line
- Complete the square first, then read the vertex off the result
- Help level
- Hint only
- Source of record
- Adopted textbook
- Confidence
- 91%
- Accommodation
- Extended time honoured
As received
Taken from the adopted textbook and the plan on file — nothing on this side is written by the agent.
Sources used
- Adopted textbook section
- Course standard
- Accommodations record
Why this help level
The teacher set hints for this assignment and a person still decides.
What the score decides
Below the configured threshold the explanation waits for the teacher of record before.
Action
ApproveEditSend to the teacherValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every session
From the student
Explain from approved materials
Draw on the adopted materials, the standard in play and the teacher's configured help ceiling.
Explain it a second way
A second explanation arrives without waiting for the next class.
Send the rest to the teacher
Contested claims, integrity questions and accommodation requests are flagged and routed to the teacher of record.
Build an evidence trail
The session, its prompts and the work produced stay attached to the student.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Learning management
- Instructure Canvas
- D2L Brightspace
Adaptive courseware
- McGraw Hill ALEKS
- Pearson MyLab
Assessment and progress
- NWEA MAP Growth
- Curriculum Associates i-Ready
Agent
Tutoring copilot
- Reads the materials
- Explains the concept
- Holds for the teacher
Tutoring and rostering
- Khanmigo
- Amira Learning
- Clever
- ClassLink
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the lesson
Every layer wraps the next. What none of them catches is in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an explanation may stand
- L3
- Puts the release in a teacher's hands
- L4 – L5
- Hold the session the feedback rests on
- L6
- Falls back to practice without feedback when signals slip
How Nestack evaluates it
Evaluate the tutoring workflow — not only the explanation that came back.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the explanation the student reads
- E1Final-output evaluationWas the explanation correct and on the adopted method?
- E2Step-level evaluationDid the agent use the right materials, standard and accommodations record?
- E3Tool evaluationDid it read the right roster record and write nothing to the gradebook?
- E4Confidence calibrationDo low-confidence explanations actually attract more teacher edits?
- E5Slice evaluationHow does performance change across specific student cohorts?
- E6Business outcomeHow many explanations needed a teacher edit or a correction after the fact?
Floor — what the learner actually understands
Failure modes
Where each failure originates in the agent
Seven ways a session goes wrong, by stage.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- TU-03
Method outside the adopted text
A general-web procedure replaces the one the class was taught.
Stage gathers
Adopted materials, standards, with the source each came from
02 · Reasoning
2 modes
- TU-04
Mastery read from a short streak
Scaffolding a plan requires is withdrawn after a few right answers.
- TU-06
Graded work treated as practice
A submitted item is answered in full, past the help ceiling.
Stage proposes
The explanation, the help level applied and confidence
03 · Tool / write
2 modes
- TU-02
Feedback posted to the gradebook
A comment lands in a field no teacher has read.
- TU-05
Student work sent onward
An essay reaches a service outside the school-official agreement.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- TU-01
Contested claim asserted
A disputed point is taught as settled fact.
Stage returns
The explanation the student learns from
05 · Change / Version
1 mode
- TU-07
Silent help-level regression
A model update moves the default from hint-first to solution-first.
Stage tracks
Model, prompt, help ceiling and materials config
- Sev-1 · acts outside the boundary
- Sev-2 · a wrong explanation reaches a student
- Sev-3 · source degrades
Affected slices
Overall quality can hide one cohort
Break the session error rate down by cohort and it names itself Nestack reports it by slice rather than in aggregate The cohorts that carry it are named, not averaged away..
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Students with plan accommodations | 4.2% | 3.5× | Review | |
| English learners and dialect writers | 3.6% | 3.0× | Review | |
| Elementary read-aloud sessions | 1.8% | 1.5× | Watch | |
| Secondary STEM practice | 0.6% | 0.5× | Normal |
Bar: error-rate lift vs. the all-session baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Each miss becomes a case
The loop shuts when the miss is a case in the suite, not when someone has explained it.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Error rate rises in one student cohort.
- 02
Diagnose
The transcript, the cited passage and the help-ceiling setting are laid against each other until one accounts for it.
- 03
Improve
Whatever changes ships against a version, with the sessions that prompted it attached.
- 04
Verify
Nothing ships until the affected cases pass a second time.
- 05
Learn
The suite grows by one case; so does the instructional guardrail set.
Learn → Detect
The return edge. Detection next time runs against a longer suite than this one.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, materials, tutoring workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Tutoring workflow discovery and boundary definition.Week 1
- 02LMS and courseware source assessment.Week 1, Week 2
- 03Standards, adopted-materials and help-ceiling mapping.Week 1, Week 2
- 04Materials ingestion and passage normalisation.Week 2
- 05Explanation logic and source binding.Week 2, Week 3
- 06Confidence scoring and escalation routing.Week 3
- 07Teacher review workflow.Week 3, Week 4
- 08LMS and courseware integration.Week 5
- 09Explanation regression cases.Week 4, Week 5
- 10Guardrails and scope controls.Week 4
- 11Session-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne course, one teacher | ProductionProduction courseware | AdvancedMultiple schools / subjects |
|---|---|---|---|
| Introduced at Pilot | |||
| Explaining from your materials | ✓ | ✓ | ✓ |
| Teacher review | ✓ | ✓ | ✓ |
| Explanation-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by subject | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved dashboard write-back | — | ✓ | ✓ |
| Courseware integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-district materials sets | — | — | ✓ |
| Multi-stage teacher approvals | — | — | ✓ |
| High session volume | — | — | ✓ |
| Enterprise instruction controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your adopted materials and scope and sequence
→Materials ingestion and passage mapping
Week 102Representative tutoring sessions
→Explanation baseline, source binding and method extraction
Week 203Your help ceiling for each assignment type
→Help-ceiling, standards and scope mapping
Week 104Access to relevant APIs, feeds or exports
→LMS, courseware and rostering assessment, then integration setup
Week 205Explanations that went wrong
→Explanation cases and failure-mode testing
Week 406Where help stops and doing the work begins
→Confidence scoring, escalation routing, guardrails and review controls
Week 307Named teachers of record to review held work
→Teacher review workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The bands follow real work rather than a plan, so evaluation and pilot genuinely share the fifth week.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Tutoring workflow discovery, scope mapping and the automation boundary
W2Materials integration and the explanation baseline
W3Tutoring workflow, confidence logic and escalation controls
W4Evaluation suite, accessible-notation checks and cohort error-rate testing
W5Courseware integration, a pilot subject and targeted corrections
W6A term of tutoring reviewed by the teacher of record, then handover
- Reading the band
- A band covers the weeks its work is named in, and nothing else. The doubled fifth week is real.
- At the end of W6
- Once the term validates, Agent Care owns the running agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Education AI agent
Build a tutoring agent around your teacher of record.
Show us your adopted materials, your standards and what counts as acceptable help. If your integrity line moves by assignment, we will configure it that way, set the automation boundary and name what stays with the teacher.
- Typical delivery
- 4–6 weeks
- Deployment
- Your LMS and courseware
- Teacher-owned
- Formative only
- After launch
- Agent Care available
Nestack Agents · Tutoring copilot
AGT-ED-03 · Agent Care available after launch