# AI SRE and agentic ops, in reading order

> AI SRE agents and agentic ops in reading order: incident investigation agents, MCP servers for ops tools, agentic CI and the IaC and GitOps beneath.

URL: https://www.matthewswong.com/en/ai-devops/ai-sre-and-agentic-ops/

- Author: Matthews Wong (https://www.matthewswong.com/en/about/)
- Reviewed on: 17 October 2026

> **Key takeaway:** This path starts with AI SRE agents that investigate alerts, from HolmesGPT and Grafana to the AWS, Azure, Google, Datadog, Elastic and PagerDuty offerings, then covers the MCP servers that give agents scoped access to ops tools, the agentic workflows that triage and fix CI, and the Terraform, OpenTofu, Pulumi, Argo CD and Flux changes underneath.

The articles on this site about putting AI agents to work in operations, arranged as one path: agents that investigate incidents, the MCP servers that let them reach your tools, agents inside CI, and the infrastructure code they change.

## Read in this order

Each stage builds on the one before. If you already know a stage, skip ahead; the line under each title says what that article adds.

### Stages 1: AI SRE agents on call

Agents that pick up an alert and come back with a hypothesis: what each one does, what it costs to run, and where its guardrails stop.

1. [HolmesGPT Kubernetes Alerts: Runbooks Beat the Model Choice](https://www.matthewswong.com/en/blog/holmesgpt-runbooks-prometheus-slack-investigations/): Start here: a two-person SRE team routes Prometheus alerts through HolmesGPT into Slack, and finds runbooks matter more than the model.
2. [Grafana Assistant Investigations: Alert-Triggered RCA Setup](https://www.matthewswong.com/en/blog/grafana-assistant-investigations-alert-triggered-rca/): Every page arrives with a hypothesis report: alert-triggered investigations in Grafana, scoped and costed safely.
3. [Elastic nightshift AI SRE: Start With the Investigation Engine](https://www.matthewswong.com/en/blog/elastic-nightshift-ai-sre-four-engines/): Elastic splits its AI SRE into four engines; why investigation alone, on existing alerts, is the safest start.
4. [Datadog Scout 1.0: What Cheaper AI SRE Investigations Mean](https://www.matthewswong.com/en/blog/datadog-scout-sre-model-bits-investigation-cost/): A close reading of Datadog's claim of cheaper investigations, the benchmarks behind it, its credits and a budget script.
5. [AWS DevOps Agent: Incident Investigation, Cost and Accuracy](https://www.matthewswong.com/en/blog/aws-devops-agent-incident-investigation-guide/): How AWS DevOps Agent triages and bills incidents, a daily cost alarm, and a way to test its accuracy claim on your stack.
6. [Azure SRE Agent: Tool Permissions, Hooks and Approval Gates](https://www.matthewswong.com/en/blog/azure-sre-agent-tool-permissions-approval-hooks/): Allow, ask and deny rules, hooks and run modes in Azure SRE Agent, and where Ask fails in Autonomous mode.
7. [Gemini Cloud Assist Investigations: Proactive RCA and FinOps](https://www.matthewswong.com/en/blog/gemini-cloud-assist-proactive-agents-finops-rca/): Google's investigations and proactive agents in Gemini Cloud Assist: setup, Eventarc routing, MCP and failure modes.
8. [PagerDuty SRE Agent: Virtual Responder Setup and Guardrails](https://www.matthewswong.com/en/blog/pagerduty-sre-agent-virtual-responder-guide/): Put an AI agent on the escalation policy as a virtual responder, and the approve-before-act gaps to know.

### Stages 2: MCP servers for the ops stack

How agents reach observability, cloud, infrastructure and portal tools, and how to scope that access before they do.

9. [Grafana MCP Server: OAuth Scopes for Claude Code and Cursor](https://www.matthewswong.com/en/blog/grafana-cloud-mcp-server-oauth-read-query-write/): Connect Claude Code or Cursor to Grafana Cloud over OAuth, and what the Read, Query and Write scopes allow.
10. [AWS MCP Server in Claude Code: OAuth, SigV4 and IAM Scoping](https://www.matthewswong.com/en/blog/aws-mcp-server-ga-call-aws-sigv4-proxy/): Reach AWS from an agent through OAuth or the SigV4 proxy, then scope IAM so it cannot delete.
11. [Amazon EKS MCP Server: SigV4, CloudTrail and Write-Mode Risks](https://www.matthewswong.com/en/blog/amazon-eks-mcp-server-managed-cloudtrail-write-mode/): The managed EKS MCP server: IAM tool permissions, what CloudTrail records, and why write mode stays off in production.
12. [Terraform MCP Server GA: A Safe Agent Setup Since 1.1.0](https://www.matthewswong.com/en/blog/terraform-mcp-server-ga-safe-agent-setup/): Run the Terraform MCP server with operations off, an organisation allowlist and a pinned version.
13. [Backstage MCP Actions Backend: Secure AI Access to Your Portal](https://www.matthewswong.com/en/blog/backstage-mcp-actions-backend-ai-agents/): Expose catalog and scaffolder actions to agents through Backstage, with the filters and auth to set first.
14. [Cloudflare Code Mode MCP: Two Tools, 1,000 Tokens, a Sandbox](https://www.matthewswong.com/en/blog/cloudflare-code-mode-mcp-two-tools-1000-tokens/): A whole cloud API behind two MCP tools, search and execute, and the sandbox and OAuth scoping behind them.

### Stages 3: Agents in CI and code review

Agents that run inside the pipeline, and what the research says about the review load they create.

15. [GitHub Agentic Workflows: Markdown Automation with gh aw](https://www.matthewswong.com/en/blog/github-agentic-workflows-markdown-gh-aw-guide/): GitHub Agentic Workflows compile Markdown into locked Actions YAML; the safe outputs, engines and where it fails.
16. [Agentic Workflow CI Failure Triage with gh-aw CI Doctor](https://www.matthewswong.com/en/blog/gh-aw-ci-doctor-failing-workflow-triage/): A read-only agent that triages failing CI runs into one deduplicated issue per failure.
17. [Copilot Fix Failing GitHub Actions: Setup and Review Rules](https://www.matthewswong.com/en/blog/copilot-cloud-agent-fix-failing-github-actions/): Copilot's fix button for failing Actions runs: admin setup, workflow approval and the review rules to keep.
18. [AI Code Review Bottleneck: What CMU and Stanford Measured](https://www.matthewswong.com/en/blog/ai-code-review-bottleneck-cmu-stanford-silent-approvals/): Output per developer grew faster than review capacity; what a large study of pull requests measured about review feedback.
19. [DORA AI-Assisted Development: Throughput Up, Instability Up](https://www.matthewswong.com/en/blog/dora-2026-ai-throughput-instability-devops/): DORA ties AI adoption to more throughput and more instability; what to measure and gate as agents join CI/CD.

### Stages 4: The infrastructure code agents act on

The safe pattern for agents on infrastructure, then the Terraform, OpenTofu, Argo CD and Flux releases to plan for.

20. [Pulumi Neo Security: Read-Only Scans, Fixes Only as a PR](https://www.matthewswong.com/en/blog/pulumi-neo-security-read-only-assessment-proposed-diff/): The pattern to copy: scan infrastructure read-only and propose fixes only as a pull request.
21. [Terraform 1.16: What's New and How to Upgrade Safely](https://www.matthewswong.com/en/blog/terraform-1-16-whats-new-upgrade-guide/): What Terraform 1.16 adds, from module import blocks to destroy = false, and what to test first.
22. [OpenTofu 1.13: convert, assume and Upgrade Breaking Changes](https://www.matthewswong.com/en/blog/opentofu-1-13-whats-new-upgrade-notes/): The OpenTofu 1.13 breaking changes and new functions to check before you upgrade.
23. [OpenTofu Ephemeral Resources and Write-Only Attributes](https://www.matthewswong.com/en/blog/opentofu-ephemeral-resources-write-only-attributes/): Keep secrets out of state with ephemeral resources, ephemeral variables and write-only attributes.
24. [Argo CD 3.5: mTLS, Source Integrity and Helm 4 Upgrade Guide](https://www.matthewswong.com/en/blog/argo-cd-3-5-whats-new-upgrade/): Argo CD 3.5's repo-server mTLS, source integrity checks and Helm 4 changes, with a pre-upgrade checklist.
25. [Flux 2.9: CLI Plugins, SSA Ignore Rules and OIDC Webhooks](https://www.matthewswong.com/en/blog/flux-2-9-cli-plugins-gitops-features/): Flux 2.9's CLI plugins, server-side apply ignore rules and OIDC webhook receivers, and the upgrade traps.

## More from the AI & DevOps guide

- [MCP & Agent Security](https://www.matthewswong.com/en/ai-devops/mcp-agent-security/): Prompt injection, the 2026 MCP incidents, access controls, sandboxes and vetting agent skills. (25 articles)
- [LLM Inference Serving](https://www.matthewswong.com/en/ai-devops/llm-inference-serving/): vLLM and SGLang, disaggregated serving, llm-d and KServe, autoscaling and cost per token. (25 articles)
- [Kubernetes for AI](https://www.matthewswong.com/en/ai-devops/kubernetes-for-ai/): The 1.36 and 1.37 upgrades, DRA and gang scheduling, Gateway API routing and agents in the cluster. (24 articles)
- [Supply Chain & Regulation](https://www.matthewswong.com/en/ai-devops/software-supply-chain-and-regulation/): npm and PyPI worms, model signing and AIBOM, secrets in agent commits, the CRA and EU AI Act. (23 articles)

---

- HTML version: https://www.matthewswong.com/en/ai-devops/ai-sre-and-agentic-ops/
- Bahasa Indonesia: https://www.matthewswong.com/id/ai-devops/ai-sre-and-agentic-ops/
- Site index for agents: https://www.matthewswong.com/llms.txt
- Sitemap: https://www.matthewswong.com/sitemap.xml
- API description (OpenAPI): https://www.matthewswong.com/openapi.json