curated · link-verified · 2026-10-17

awesome-agentic-devops

The shift from deterministic, script-driven DevOps to autonomous AI agents that reason, act on infrastructure, self-correct, and collaborate with humans across the software delivery lifecycle — every resource that matters, in one place.

0 resources 15 categories 6 seed sources 128 GitHub links API-verified 65 web links checked 0 archived · flagged
▸ github.com/adventurewave-labs/awesome-agentic-devops

What is Agentic DevOps?

Four definitions this directory is built on — from the vendors who coined the term, the analysts who measure it, and the theorists who frame it.

Microsoft · Build 2025

AI-powered agents operating as members of dev and ops teams — Copilot agents plus Azure automating, optimizing, and accelerating the entire software lifecycle.

azure.microsoft.com →
Opsera

The shift from static, human-managed scripts to agents that reason, self-correct, and manage the pipeline — autonomous decisioning, not just autonomous execution. Upstream + downstream together = the AIDLC.

opsera.ai →
Superagentic AI

Four theoretical pillars: automation→autonomy, intelligent orchestration, continuous learning, and human-agent collaboration — with Development, QA, SRE, and Optimization agent taxonomies.

super-agentic.ai →
What it is NOT

Not traditional CI/CD (deterministic vs. probabilistic). Not just LLM code generation (no agency). Not AIOps — AIOps surfaces anomalies; agentic DevOps reasons through causes and executes remediation.

opsera.ai →

C-P-A agent anatomy

Perception (logs/metrics/traces → embeddings) → Memory (RAG over runbooks & incident history) → Reasoning (ReAct plans) → Tool use (verified CLI execution). — Harness

Three adoption horizons

1 Augmented Operator (sidecars, human-in-the-loop) → 2 Agent Swarms (specialized teams, human-on-the-loop) → 3 Autonomous SRE (policy-managed Tier-1). — Harness

AI is an amplifier

DORA 2025: AI magnifies existing org strengths and weaknesses; unstable AI adoption correlates with reduced throughput and stability.

The lethal trifecta

Private data + untrusted content (PRs, issues, web) + outbound access in one sandbox. GitHub's gh-aw answer: six defense layers + credit caps.

Protocols made it real

"MCP is how agents use tools. A2A is how agents collaborate." — StackOne. MCP is now Linux Foundation governance.

The review bottleneck

AI code cuts PR-open time 58% — but PRs sit 4.6× longer in review. Agentic governance is the missing half. — Opsera 2026 benchmark

The Directory

Filter by search above. Badges: OSS OFFICIAL VENDOR COMMUNITY ARCHIVED · stars as of 2026-10-17.

The Podcast Index

agenticdevops.fm · all 14 episodes extracted

Agentic DevOps: AI Engineering for Infrastructure — Bret Fisher, since May 2025. "Real-world AI agents for DevOps without the hand-wavy hype." Full episode index:

#EpisodeGuestDate

Talks & Video

Safety First — operator rules

These agents can change infrastructure. Distilled from OWASP, GitHub's gh-aw architecture, and the DevOpsAIguru123 safety model.

Read-only or proposal mode first. gh-aw's staged: true, read-only MCP modes (K8s, DBHub, Flux), approval gates before any write action.
Beware the lethal trifecta. Private data + untrusted content + outbound access must never share one sandbox. GitHub runs six defense layers; run at least one.
Policy-as-code beats vibes. Deterministic checks (OPA, protected approvals) gate probabilistic agents. Block terraform destroy-class actions outright.
Scope credentials per agent. Diagnosis agents get read-only; remediation agents get namespace-scoped writes; secrets via Doppler/Vault MCPs — never in prompts.
Record everything. Black-box recorder / compliance evidence packets for every production-adjacent run. See the OWASP MCP Top 10.