Education AI agent · Drafting
Teacher & Staff Copilot AI Agent
Draft lesson materials, comments and administrative text from the inputs on file, hold each draft for the educator who approves it, and leave every decision of record to a person.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Educator review
- Draft only
- After launch
- Agent Care
What this agent does
Drafts the document, not the decision
In
- 01
Ingest the standards, the templates and the team-authored source text from supported SIS.
- 02
Normalise each input and carry it forward with the record it came from.
Reason
- 03
Draft lesson materials, rubric-aligned comment options and the administrative text staff ask for.
- 04
Apply the district's templates, reading-level rules and the direct-control limits on how student work may be used.
- 05
Emit artefacts with heading structure, alt text and reading order.
Decide
- 06
Check a team-authored programme draft for missing fields and internal consistency, and flag them.
- 07
Route each draft to the named educator, and put a suspected-abuse disclosure straight to the reporter's own duty.
Out
- 08
Retain the inputs, the draft, the version and the approval for the district, and put none of it to training.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent drafts; a grade of record stays the teacher's act, and an individualised education programme stays the team's decision, made with the parent participating.
Example workflow
One draft, request to approval
- 1Request receivedPlanning tool, LMS, SIS task or a staff message
- 2Inputs assembledStandards, templates and the team-authored text, each with the record it came from
- 3Draft producedDraft text, the inputs behind it, flagged lines and confidence
- 4Policy checks appliedInput checks, template and reading-level rules, decision-language flags and confidence threshold
No human action required
The first four stages run unaided, and nothing is decided at any of them — the agent is drafting, and the educator's lane opens at the confidence gate.
Goes to the named educator to approve.
Picks up a specialist read first.
Educator review
The draft is held with its inputs, its flagged lines and the confidence.
Approve · Edit · Send to a specialist
Approved — released to send
- 6School systems updatedOnly where write access and approval policy allow it
- 7Draft evaluatedEdit distance, flagged-line outcomes, approval latency and corrections after the fact
Revisions
Every educator edit is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Drafting, amending or finalising programme goals, services or placement.
- Assigning or posting a grade of record.
- Determining a student's eligibility for special-education services.
- Filing, delaying or standing between a reporter and a report of abuse.
Automation boundary
Agent acts unaided
- ✓Draft a lesson plan or slide deck against stated standards for the named owner.
- ✓Generate rubric-aligned comment options for the teacher into the review queue.
- ✓Check a team-authored draft for missing fields.
- ✓Track deadlines and reconvene dates.
Any write happens inside the boundaries agreed at implementation, never ahead of the educator.
- Deciding, signing or transmitting a student's disability plan.
- Making a manifestation determination or a discipline recommendation.
- Sending a parent notice, prior written notice or eligibility letter.
- Releasing student work outside the school-official agreement.
Example output
One draft, annotated
Everything the agent drafts is attached to the inputs it was built from.
Drafting output · single document
Illustrative example- Document
- Progress comment
- Drafted line
- Shows the method clearly and needs the final step written out
- Reading level
- Middle-grade
- Source of record
- Teacher's rubric
- Confidence
- 90%
- Approval
- Held for the teacher
As received
Taken from the teacher's rubric and the work on file — nothing on this side is written by the agent.
Inputs used
- Teacher's rubric
- Course standard
- Submitted work
Why this wording
It describes the work and states no mark and a person still decides.
What the score decides
Below the configured threshold the draft picks up a specialist read before the educator sees.
Action
ApproveEditSend to a specialistValue
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 draft
From the request
Draft from the inputs
Draw on the standards, the templates and the team-authored text already on file.
Draft it, do not decide it
Routine lesson materials and comment options arrive already written.
Send the rest to an educator
Decision language, programme content and reporting triggers are flagged and routed to a named person.
Build an evidence trail
Each draft keeps its inputs, its version and the educator who approved it.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Student information systems
- PowerSchool SIS
- Infinite Campus
LMS and gradebook
- Instructure Canvas
- PowerSchool Schoology Learning
Special education
- Frontline Special Education
- PowerSchool Special Programs
Agent
Teacher and staff drafting
- Reads the inputs
- Drafts the document
- Holds for approval
Drafting and grading
- Google Workspace
- Microsoft
- Gradescope
- Turnitin
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 record
Each control contains the next; the map below names the gaps.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a draft may stand
- L3
- Puts the release in an educator's hands
- L4 – L5
- Hold the inputs behind every draft
- L6
- Reverts to outline-only when signals degrade
How Nestack evaluates it
Evaluate the drafting workflow — not only the finished document.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the document the reader receives
- E1Final-output evaluationDid every statement in the draft match an input on file?
- E2Step-level evaluationDid the agent use the right template, standard and source record?
- E3Tool evaluationDid it read the right record and write nothing to the gradebook?
- E4Confidence calibrationDo low-confidence drafts actually attract more educator edits?
- E5Slice evaluationHow does performance change across specific document types?
- E6Business outcomeHow many drafts needed an educator edit or a correction after the fact?
Floor — the work that reaches the classroom
Failure modes
Where each failure originates in the agent
Seven modes across the drafting lifecycle.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- IS-03
Closed document retrieved
Last year's goals are pulled in place of the active plan.
Stage gathers
Standards, templates, with the source each came from
02 · Reasoning
2 modes
- IS-04
Trend written as a decision
A progress trend enters the draft as though the team agreed it.
- IS-06
Fluency scored as quality
A rubric criterion tracks surface fluency and down-scores some writers.
Stage proposes
The draft, the inputs behind it and confidence
03 · Tool / write
2 modes
- IS-02
Scores posted before review
Suggested marks reach the gradebook ahead of the review queue.
- IS-05
Notice sent unreviewed
A generated parent notice goes out through the mail merge.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- IS-01
Translation that inverts a right
A translated notice reverses a clause about parent rights.
Stage returns
The document the parent or student receives
05 · Change / Version
1 mode
- IS-07
Silent template regression
A template change shifts phrasing mid-year with nothing tied to a version.
Stage tracks
Model, prompt, template and routing config
- Sev-1 · acts outside the boundary
- Sev-2 · a wrong document reaches a record
- Sev-3 · input degrades, draft routes to review
Affected slices
Overall quality can hide one document type
The baseline is the rate across all drafts, and read that way it looks manageable. Read by document type 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 |
|---|---|---|---|---|
| Programme and plan documents | 5.3% | 3.8× | Review | |
| Machine-scored written work | 4.6% | 3.3× | Review | |
| Translated parent notices | 2.1% | 1.5× | Watch | |
| General formative comments | 0.7% | 0.5× | Normal |
Bar: error-rate lift vs. the all-draft baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A cycle ends in a standing case
What closes a cycle is a case in the suite, not agreement about what went wrong That suite is what the following detection is measured against..
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Error rate rises in one document type.
- 02
Diagnose
Which changed — the template, the input record or the routing? The drafts are read back until one answer holds.
- 03
Improve
Each change is version-stamped and linked to the drafts that surfaced it.
- 04
Verify
The release waits on the affected cases passing again.
- 05
Learn
It becomes a permanent case and a change to the drafting rules.
Learn → Detect
The return edge. The next detection is measured against a longer suite.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, inputs, drafting workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Drafting workflow discovery and boundary definition.Week 1
- 02SIS and planning-tool assessment.Week 1, Week 2
- 03Template, standards and reading-level mapping.Week 1, Week 2
- 04Input ingestion and normalisation.Week 2
- 05Drafting logic and input binding.Week 2, Week 3
- 06Confidence scoring and approval routing.Week 3
- 07Educator review workflow.Week 3, Week 4
- 08SIS and planning-tool integration.Week 5
- 09Draft regression cases.Week 4, Week 5
- 10Guardrails and approval controls.Week 4
- 11Draft-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 school, one team | ProductionProduction school systems | AdvancedMultiple schools / districts |
|---|---|---|---|
| Introduced at Pilot | |||
| Drafting from your inputs | ✓ | ✓ | ✓ |
| Educator review | ✓ | ✓ | ✓ |
| Draft-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by document type | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Planning-tool integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-district template sets | — | — | ✓ |
| Multi-stage educator approvals | — | — | ✓ |
| High drafting volume | — | — | ✓ |
| Enterprise drafting 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 templates and document standards
→Input ingestion and template mapping
Week 102Representative drafts from your staff
→Drafting baseline, input binding and style extraction
Week 203Your approval routing and named approvers
→Template, routing and reading-level mapping
Week 104Access to relevant APIs, feeds or exports
→SIS, LMS and special-education assessment, then integration setup
Week 205Drafts that should not have gone out
→Draft cases and the evaluation suite
Week 406Which drafts may never leave without an educator
→Confidence scoring, approval routing, guardrails and review controls
Week 307Named educators to approve held drafts
→Educator 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
Phases occupy real weeks, and evaluation overlaps launch in the fifth rather than being stretched.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Drafting workflow discovery, template mapping and the automation boundary
W2Source integration and the drafting baseline
W3Drafting workflow, confidence logic and approval controls
W4Evaluation suite, accessibility checks and document-type error testing
W5SIS integration, a pilot document set and targeted corrections
W6A unit of materials approved by the teacher, then handover
- Reading the band
- Each band covers the weeks its work is named in and no others. The doubled fifth week is real.
- At the end of W6
- A verified unit in production, then monitoring sits with Agent Care.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Education AI agent
Build a drafting agent around your school's approval chain.
Show us your templates, your document standards and who signs each one. Which of your drafts can never leave without a named educator? We map the routing from that answer, set the automation boundary and hold the rest behind a person.
- Typical delivery
- 4–6 weeks
- Deployment
- Your SIS and LMS
- Educator review
- Draft only
- After launch
- Agent Care available
Nestack Agents · Teacher and staff copilot
AGT-ED-08 · Agent Care available after launch