DevOps & SRE AI Platforms Atlas 2025

A vendor-neutral, community-driven wiki – Last updated: November 12, 2025

GitHub GitHub GitHub

What This Is

This atlas provides comprehensive, evidence-based analysis of AI agent platforms across the DevOps/SRE/ITOps ecosystem.

Key Features:

  • Vendor-neutral analysis – No hype, just capabilities
  • Community-driven – Contributions welcome
  • Structured data – Machine-readable platform information
  • Actionable insights – Real implementation guidance

Quick Navigation

Executive Summary – DevOps & SRE AI Platforms (Oct 11, 2025)

The headline: AI agents for DevOps/SRE have progressed from “chat over dashboards” to actionable co‑workers spanning triage → RCA → guided remediation → post‑incident learning. No single vendor covers the full lifecycle; the 2025 winning pattern is composable adoption: anchor on an observability‑first triage brain, add a workflow orchestrator for approvals/audit, and layer in guardrailed provisioning for safe apply/rollback.

Market map at a glance

Archetype Best fit Representative platforms Core strength Typical limits
Observability‑First Fast triage, RCA, incident comms Dynatrace Davis AI; Cisco (Splunk) AI Agents; Datadog Bits AI & Agents; Elastic AI Assistant; New Relic AI Multi‑signal correlation, causal/hypothesis reasoning, narrative updates Usually no direct apply beyond playbooks; relies on external approvals
Provisioning‑Focused Safe, repeatable infra changes Azure Copilot (Agent Mode); DuploCloud; Qovery; Kuberns Generate/apply IaC with approvals and rollback Lighter AIOps correlation; observability via integrations
Developer‑Centric & Frameworks Code/PR changes; build your own agents AWS Strands SDK; Atlassian Rovo Dev; GitHub Copilot Coding Agent; Zencoder; JFrog Fly; Azure AI Agent Service (Foundry) Planning/tool orchestration, CI/CD fixes, AgentOps Not a runtime ops console; direct infra apply limited
Enterprise Orchestrators Cross‑team workflows, audit, CMDB ServiceNow AI Agent Orchestrator; Salesforce Agentforce (OpsAI); PagerDuty AIOps Ticket/change graph, approvals, runbooks Deep telemetry depends on observability tools
Data & MLOps AI quality, data pipelines Databricks Agent Bricks; Snowflake Cortex Agents; Dataiku AI Agents Evaluations/guardrails, lineage, model operations Infra ops out of scope
Specialized Domain Deep expertise for a niche IBM AskIAM (IAM); Solo.io Kagent (K8s) Accuracy within narrow scope Limited breadth by design

Top 6 takeaways for 2025

  1. From chat to action: leaders pair explanations with verifiable evidence (queries/logs/traces) and suggest the next safe step (runbook, PR, or controlled apply).
  2. Human‑in‑the‑loop is default: approvals, RBAC and audit trails are table stakes for production change.
  3. AgentOps matters: tracing, evaluations and policy hooks (data boundaries, PII controls, prompt safety) separate pilots from production.
  4. Event ontology wins RCA: platforms with typed entities + change context consistently outperform generic LLMs on root cause.
  5. IaC is the safety rail: even where agents can apply changes, diffs/Bicep/Terraform + approvals remain the safest path.
  6. Compose the stack: the best outcomes pair one observability brain + one orchestrator + optional provisioning.

Quick picks by job‑to‑be‑done

  • Rapid triage & RCA with evidence: Dynatrace; Cisco (Splunk); Datadog Bits AI — complement with PagerDuty AIOps for comms/response.
  • End‑to‑end incident coordination & comms: PagerDuty AIOps; ServiceNow; Salesforce — complement with your observability suite.
  • Direct, approval‑gated infra changes: Azure Copilot (Agent Mode); DuploCloud — complement with Qovery for migration IaC.
  • Build governed, bespoke ops agents: Azure AI Agent Service (Foundry); AWS Strands — complement with Dataiku/Databricks for evaluations.
  • Repo/CI‑centric fixes (PRs/tests/docs): GitHub Copilot Coding Agent; Atlassian Rovo Dev; Zencoder — complement with JFrog Fly for release policy.
  • Kubernetes deep‑dive fixes: Solo.io Kagent — complement with APM/logs.
  • Identity requests & compliance: IBM AskIAM.
  • Data & MLOps guardrails/evals: Databricks Agent Bricks; Dataiku; Snowflake Cortex Agents.
  • Diagnostics: more change‑aware correlation and hypothesis testing; observability tools add post‑mortem drafting and Slack/Teams updates by default.
  • Activities & history: richer incident timelines and action replays (e.g., “apply last good remediation”), strongest in orchestrators.
  • Provisioning: clear split—some stay advisory; others (Azure Copilot, DuploCloud) execute with diffs, approvals, rollbacks.
  • Event ontology: convergence on OTel semantic conventions and CMDB/entity graphs; better joins across alerts, services and changes.
  • Agent observability: first‑class traces, evaluations, safety scores (Foundry, Dataiku, Databricks, Salesforce Command Center).

Reference architecture (what “good” looks like)

  1. Telemetry backbone: OTel + your observability suite (Dynatrace/Splunk/Datadog/Elastic/New Relic).
  2. Orchestration & audit: ServiceNow or Salesforce; add PagerDuty AIOps for real‑time response.
  3. Provisioning lane: Azure Copilot (Agent Mode) or DuploCloud for controlled applies; Qovery for migration IaC.
  4. Developer loop: GitHub Copilot Coding Agent / Atlassian Rovo Dev / Zencoder; JFrog Fly for release policy.
  5. AgentOps layer: Azure AI Agent Service (Foundry), Dataiku, or Databricks for tracing, evaluations and safety.
  6. Guardrails: RBAC, approval workflows, drift checks, read‑only dry‑runs by default and immutable audit logs.

Risks & guardrails to enforce

  • False confidence / silent failures: require evidence links for every diagnosis; block summary‑only outputs.
  • Unsafe changes: enforce two‑person approval and automatic rollback plans; prefer IaC diffs over ad‑hoc commands.
  • Vendor lock‑in: favour platforms with MCP/A2A, exportable traces and open schemas.
  • Cost surprises: track agent run counts, tool invocations, LLM usage; set SLOs & budgets for agents.

Bake‑off checklist (pair with Atlas KPIs)

  • TTFC/TTRC: measure on a fixed incident set, with linked evidence and a verification plan per RCA.
  • Actionability rate: percentage of cases with a safe next step (rollback/fix or runbook).
  • Closed‑loop rate: percentage of incidents where the platform proposed and executed a remediation under approvals.
  • AgentOps quality: presence of traces/evals/guardrails, red‑team tests, data‑boundary controls.
  • Interoperability: MCP/OTel support; OTel‑CMDB mapping if you use an orchestrator.

Join the Community

This is a living document! Help us keep it current:

  • Contribute updates via GitHub
  • Join conversations via GitHub Issues (GitHub Discussions is not enabled)
  • Report bugs or feature requests via the same Issues link above.
  • Meet the team & community at our Bucharest XDevOps Meetup

This project is open source and maintained by the xDevOps community.