Epic
AI-Graded Exercise
Allow exercises to reach completion using AI Practice Coach scoring alone, without requiring a human grader.
Client: ZoomInfo (contracted)
Phase: Decide
Scope: Exercise · Task Completion · Config
Business case
Today, an exercise task never reaches completion until a human grader creates a scorecard. This creates two problems:
- Grading queue burden. Practice exercises jam up the grading queue when the exercise is just for rep practice, not formal evaluation.
- Course completion blocked. Courses containing exercises can't be completed without human graders, blocking learning program progress.
ZoomInfo has contracted for this feature. Additional prospects have cited the grading bottleneck as a barrier to exercise adoption.
Core change
AI Practice Coach already auto-generates scorecards using the exercise's scorecard template. The core change is narrow: task completion logic must accept an AI-created scorecard as satisfying the grading requirement when AI-Graded mode is enabled.
Key workflow distinction
The learner sees the AI score before handing in. Hand-in is a confirmation step, not a scoring trigger.
Human grading (today)
- Learner records response
- Learner hands in
- Waits for human grader
- Gets score
AI grading (new)
- Learner records response
- AI scores automatically
- Learner reviews score & feedback
- Hands in (or re-records)
Exercise completion logic
- No passing threshold → handed in + AI scorecard = finished (complete)
- Passing threshold met → pass (complete)
- Passing threshold not met → redo required (not complete; learner re-records)
- Course completion: no pass/fail at course level — just checks all required items are complete
Decisions
D1
Access control: Two-layer gate
Company setting must be ON and the user must have the permission to set an exercise as AI-graded. When the user lacks the permission, the AI-only option is visible but locked, with a tooltip identifying the required permission. Users without the permission can view but not change the AI-graded status on exercises set by others.
D2
Exercise properties UX
"AI-only Assessment" is a new option in the "Who will evaluate final submissions" selector. When selected, human grader fields are hidden/disabled. An inline acknowledgment toggle appears below: "I understand that AI-only may not be used where it could result in employment decisions. I am legally authorized to use this setting." Help bubble links to regulatory guidance. The toggle is one-time per exercise — once acknowledged, it stays visible and on for subsequent edits. Both the option and the acknowledgment are required to save. An audit trail records who acknowledged and when.
D3
Hand-in validation: Client + server
Client-side: hand-in button disabled with inline explanation until an AI scorecard exists. Server-side: API validates the response has an AI-generated scorecard instance of the scorecard template assigned to the exercise. The server check covers all hand-in paths (exercise page, content tile, mobile) and returns a displayable message on failure.
D4
Reporting: Fast-follow
Reporting updates ship after V1. Exercises function correctly and completions happen, but reports may show grader identity incorrectly until the reporting update ships. See the Reporting section below for the full surface list.
D5
Toggle-off: No special UX
When the company setting is turned off, AI-graded exercises behave like exercises with no grader assigned — a valid state today. No migration flow needed. Existing completions remain valid.
D6
Copied exercises: Permission-gated
When an AI-graded exercise is copied by a user who lacks the permission, the AI-graded setting is stripped and the exercise reverts to human-graded.
Admin workflow
- Company admin enables the AI-Graded Exercise company setting.
- User admin grants the AI-Graded Exercise permission to appropriate users.
- Permitted user opens exercise properties.
- Under "Who will evaluate final submissions," selects "AI-only Assessment."
- Human grader fields are hidden/disabled.
- Acknowledgment toggle appears below. User toggles it on. Help bubble links to regulatory docs.
- User saves. Audit trail records who acknowledged and when.
- On subsequent edits, acknowledgment toggle is visible and already on.
Learner workflow
- Learner records a response.
- AI Practice Coach automatically scores using the exercise's scorecard template.
- Learner reviews the AI score and feedback.
- If satisfied, learner hands in.
- Button disabled until a matching AI scorecard exists on the response.
- Server validates the scorecard before accepting.
- If not satisfied, learner re-records and gets a new AI score.
- Task marked complete per completion logic (finished, pass, or redo required).
Feature breakdown
-
Cloud
Company setting and user permission
- Company-level setting to enable/disable AI-Graded Exercise
- User permission to control who can set exercises as AI-graded
- When company setting is off, AI-graded exercises behave as no grader assigned
-
Cloud
Task completion logic
- Accept AI-created scorecard as satisfying grading requirement when AI-Graded is enabled
- Apply existing completion logic (finished/pass/redo) using AI scorecard
-
Cloud
Hand-in validation
- API validates response has AI-generated scorecard of the exercise's scorecard template
- Reject with displayable error if validation fails
- Covers all hand-in paths
-
Cloud
Acknowledgment audit trail
- Store who toggled the legal acknowledgment and when for each exercise
- Retrievable for compliance/legal review
-
Web
Exercise properties UI
- "AI-only Assessment" in evaluator selector with inline acknowledgment toggle
- Visible but locked for users without the permission (tooltip explains)
- Toggle one-time per exercise; stays on after initial acknowledgment
-
Web
Learner hand-in experience
- Disable hand-in button with explanation until AI scorecard exists
- Updated hand-in text indicating AI-only scoring
-
iOS / Android
Learner hand-in experience
- Disable hand-in until matching AI scorecard exists
- Updated hand-in text indicating AI-only scoring
-
Web
Copied exercise behavior
- Strip AI-graded setting when copying user lacks the permission
- Reverts to human-graded
Scope
| Item | Status |
| AI scorecard satisfies task completion | V1 |
| Company setting gate | V1 |
| User permission gate | V1 |
| Exercise property — AI-only grading with acknowledgment | V1 |
| Acknowledgment audit trail (who, when) | V1 |
| Hand-in validation — client + server (template match) | V1 |
| Updated hand-in text (web + mobile) | V1 |
| Copied exercise permission check | V1 |
| Re-trigger scoring via transcript edit | V1 (existing) |
| Reporting — AI grader identity across all surfaces | Fast-follow |
| Document Response exercises | Out |
| Dispute / appeal mechanism | Out |
| AI draft, human confirm | Out |
| Manager completion notifications | Out (separate feature) |
Reporting (fast-follow)
The following surfaces need to display AI grader identity and feedback correctly after V1 ships. AI-graded scorecards already record AI as the grader — the work is ensuring all downstream consumers render it correctly.
- Reports
- Any report showing grader identity must display "AI Practice Coach" for AI-graded exercises
- Dashboards
- Completion and exercise performance views must reflect AI-graded completions
- Exports
- CSV/Excel exports of exercise data must include AI grader identity
- Data feeds
- Integrations surfacing grader information must handle AI as the grader
- Scorecard views
- AI grader displayed correctly outside exercise context (user profile, manager views)