Agentic DevOps
Agentic DevOps is the shift from static automation (CI/CD pipelines, scripts) to autonomous, goal-oriented AI agents that can observe, reason, and act within a DevOps environment.
The 4-Layer Stack
According to the Agentic Readiness DevOps Assessment, a team’s readiness is mapped across four layers:
- Foundation: Data quality, observability, and existing automation maturity.
- Platform: Cloud infrastructure and developer platform capabilities.
- Agent Infrastructure: The hosting environment and runtime for AI agents.
- Agent Intelligence: The LLMs and reasoning engines driving the agents.
Governance & Security
A “governance-first” approach is critical to prevent security breaches (e.g., the Vercel/Context tool incident).
- Agent Boundary Design: Agents must have strictly defined IAM boundaries.
- Permission Matrix: Every agent must be audited for:
- READ: What data/logs can it access?
- WRITE: Where can it commit code or change configs?
- EXECUTE: What commands or pipelines can it trigger?
Evaluation Framework (Agentic TDD)
Applying Test-Driven Development (TDD) principles to agents involves creating explicit evaluation scenarios:
- CI Failure Scenarios: Flaky tests, OOM kills, dependency conflicts, missing secrets.
- Rules of Engagement: Explicitly defining what an agent should not do (e.g., “do not delete the database to fix an OOM kill”).
Tooling Landscape
- Reasoning/Coding: Claude Code, GitHub Copilot Agent, OpenAI Codex.
- Orchestration: LangChain, Argo Workflows.
- DevOps-Specific: kagent.
References: