AI intelligence: release and incident summaries grounded in telemetry

[unjica]

Roadmap priority

7

Goal

Provide actionable AI-generated insights for releases and incidents by combining telemetry data, release health, notifications, and performance signals.

The goal is not to build another dashboard, but to help developers understand:

• What changed?
• What is most likely affected?
• When did it start?
• How severe is it?
• What should they investigate next?

Example

AI Release Summary — v1.4.0

Changes detected:
- 15 new errors
- Error rate +320%
- 54 affected users
- Sessions −28%

Most affected areas:
- POST /checkout
- Mobile Safari (90% of failures)
- First observed 3 minutes after deployment

Suggested next steps:
- Investigate checkout API changes
- Compare v1.4.0 against v1.3.9
- Review recent production errors

Current state

Telemetry Tracker already provides (or has roadmap work for):

Capability	Issue / area	Status note
Notifications	#492	Parent / vision roadmap
Alert Rules	#493	Parent / vision roadmap
Release Health	#453	Priority 3
Search	#494	Priority 4
Compare Periods	#495	Priority 5
Performance metrics	#196, #197	Priority 6
Workspace AI brief infrastructure	existing brief-snapshot / related APIs	Present in codebase

There is currently no AI-generated release or incident intelligence surface that combines these signals into grounded summaries and recommendations.

AI intelligence surfaces

Future AI-generated insights may include:

Surface	Examples
Release summaries	Release regressions, adoption changes, new errors
Incident summaries	Error spikes, missing telemetry, quota incidents
Root-cause hints	Common error patterns, affected platforms or endpoints
Recommendations	Suggested investigation paths
Weekly reports	Project and organization summaries
Anomaly detection	Unusual telemetry patterns and regressions

MVP should start with release summaries and incident summaries grounded in existing telemetry; other surfaces can follow without changing the product framing.

Deliverables

AI summaries

• AI-generated release summaries
• AI-generated incident summaries
• Structured summaries based on existing telemetry data (facts vs suggestions clearly separated)
• Confidence-aware recommendations where appropriate

Intelligence signals

AI may combine signals from:

• Releases (Dashboard: Releases Health page — adoption and regression across versions #453)
• Errors / Events / Sessions
• Notifications (Notifications: multi-channel delivery (vision / parent roadmap) #492) and Alert Rules (Alert rules: configurable conditions and destinations #493)
• Performance metrics (Performance: slow routes and slow pages tables #196, Overview: Performance Overview card (Web Vitals snapshot) #197)
• Compare-period results (Dashboard: Compare periods and releases — explicit better/worse UX #495)
• Existing workspace AI brief / snapshot infrastructure

Navigation

• Link AI summaries back to existing dashboard views
• Preserve existing project and scope filters where applicable
• Surface relevant releases, errors, sessions, and performance views from within a summary

UX & quality

• Loading, empty, and error states
• Clearly distinguish telemetry facts from AI-generated suggestions
• Display uncertainty where confidence is low
• API and dashboard test coverage where deterministic behavior is expected (payload shape, grounding inputs, link helpers)

Technical notes

• AI-generated content must be grounded in existing telemetry data (no invented metrics).
• Recommendations should be presented as suggestions, not authoritative conclusions (and not as auto-rollback instructions).
• Reuse existing workspace brief / AI infrastructure rather than duplicating aggregation paths.
• Release and incident intelligence should build on earlier roadmap priorities (Dashboard: Releases Health page — adoption and regression across versions #453–Dashboard: Compare periods and releases — explicit better/worse UX #495, Performance: slow routes and slow pages tables #196–Overview: Performance Overview card (Web Vitals snapshot) #197) rather than inventing parallel aggregations.
• Release identifiers remain arbitrary application-defined strings (no semver assumptions).
• Stay within current project / workspace RBAC scope.

Related

• Notifications: multi-channel delivery (vision / parent roadmap) #492 — Notifications
• Alert rules: configurable conditions and destinations #493 — Alert Rules
• Dashboard: Releases Health page — adoption and regression across versions #453 — Release Health
• Dashboard: Global Search across errors, events, sessions, releases, and users #494 — Search
• Dashboard: Compare periods and releases — explicit better/worse UX #495 — Compare Periods
• Performance: slow routes and slow pages tables #196 — Slow routes and pages
• Overview: Performance Overview card (Web Vitals snapshot) #197 — Performance Overview card

Out of scope

• Autonomous remediation
• Automatic rollbacks or production changes
• AI-only anomaly detection without telemetry signals
• Cross-project intelligence outside the current workspace scope
• Predictive analytics and forecasting
• Building a separate “AI dashboard” that replaces Overview / Releases / Alerts

Base branch

develop

Suggested acceptance checks

1. AI summaries are grounded in existing telemetry data (claims map to metrics / entities the user can open).
2. Release summaries correctly reference associated releases and telemetry metrics.
3. Incident summaries link back to the relevant dashboard views.
4. Recommendations are presented as suggestions and not as facts.
5. Existing project and scope filters are preserved where applicable.
6. Empty and no-data states display appropriate messaging.
7. Deterministic pieces (input snapshot shape, deep links, fact/suggestion separation) have test coverage.

[unjica]

Tracked under public epic AI Intelligence (#567) / milestone v1.18.x — AI Intelligence. Keep this issue product-facing (no prompts, models, or provider implementation details).