Build
PagerDuty, GitHub, Slack, Jira as operational products — fail-open publish on the hot path.
Platform engineering and SRE
AI-SRE-ready context plane for incidents, deploys, and on-call evidence as governed dept.* products — agents investigate with lineage, not another alert chat box.
Vertical map (2026)
Platform teams already standardize on OpenTelemetry for service observability. The AI SRE category is flooded with alert-correlation agents; durable gains come from agents that reuse change + incident history under governance (context engineering) — the same rigor as OTel, applied to the agent layer — not one-shot log chat or company-wide vector dumps.
Head of PlatformSRE / Reliability leadDevOps managerCTO of product company
Steer
MCP tools for health, incident trace, and mesh compose with tenancy, lineage, and audit.
Compound
RCA quality evals on your needles; institutional memory of similar incidents across seasons.
| Workflow | Buyer pain | Mesh answer |
|---|---|---|
| Incident investigation | Alerts without change context; humans re-discover the same failure modes every quarter. | Incident link-enrich: page → PR → Slack war-room → prior similar incidents on dept.engineering. |
| Multi-tenant platform ops | Internal platform incidents lack a single operational narrative for agent copilots. | platform-sre use case: run reliability ops on the same mesh customers will run. |
| Deploy risk & change | Deploy and feature-flag context lives outside the incident timeline. | GitHub deployment + check-run streams as first-class operational facts for agents. |
| Agent context without SQL sprawl | Each agent re-queries SaaS APIs or pastes tickets into chat — no shared governed products. | dept.* data products + policy-gated MCP — one fabric, audit trails, tenancy you already trust. |
- Operational mesh
- Knowledge mesh
- Analytical bridge
Learning loop thesis
AI SRE fails when agents only chat logs. Wins come from change + incident + war-room context that compounds under governance — the same operational rigor you already apply with OpenTelemetry, extended to the agent context plane.
- Build: PagerDuty, GitHub deploys, Slack war rooms, and Jira as operational products — fail-open on the hot path.
- Steer: MCP incident trace and health tools with tenancy, lineage, and audit on every invoke.
- Compound: RCA quality on your needles; similar-incident memory that survives model swaps.
Outcomes to measure
- MTTR leading indicators: time-to-linked-PR and similar-incident recall
- Platform incidents as first-class products agents can query via MCP
- Change-aware investigation without pasting deploys into a chat window
- Governed operational context for agents — not another alert-correlation chat box or company-wide vector dump
Connector focus
Connectors on this map should match /product and /data-products. List the tools. Do not print a readiness label.
- PagerDuty
- GitHub
- Slack
- Jira
- Zendesk
- dbt
- Confluence
Related use cases
Packaged workflows and starting points for this vertical map.