What Agentic AI DevOps Means for Security
Agentic AI DevOps is the use of autonomous, always-on AI agents that monitor, secure, and modify software delivery pipelines, handling tasks such as secrets management, automated code review, and supply chain security without constant human prompts. This new pattern moves AI beyond bursty code generation into continuous operations: watching for vulnerabilities, enforcing policies, and proposing fixes in real time. Vendors are starting to embed these agents deep inside their platforms. GitLab’s latest release shifts its Duo agents from helping write code to protecting credentials, managing AI merge requests, and scanning dependencies before release. AWS is promoting a similar model with DevOps-focused agents that run in the background, connect to observability tools, and test builds in isolated environments. As these systems mature, they promise faster, safer releases—but also raise hard questions about how much critical infrastructure can be trusted to autonomous AI.
GitLab 19.0: Agentic AI Secrets and Merge Requests
GitLab 19.0 shows how AI secrets management and automated code review are becoming first-class parts of the DevOps toolchain. The new GitLab Secrets Manager, in public beta, keeps credentials in the same platform that runs code and CI jobs, limiting each secret to authorised pipelines and logging every use in the existing project hierarchy. Instead of replacing cloud key stores, it connects to HashiCorp Vault, AWS Secrets Manager, Azure Key Vault, and Google Cloud Secret Manager, so teams can centralise control without changing infrastructure. On the collaboration side, the Developer Flow agent now tracks the full AI merge requests lifecycle: it reads AGENTS.md for project rules, responds to reviewer comments, splits oversized changes, and can resolve conflicts with a Resolve with Duo action that commits a proposed fix. GitLab Duo still respects branch protection, reinforcing the idea that AI should operate inside, not around, governance.
Automating Supply Chain Security and Dependency Scans
GitLab is also pushing agentic AI into supply chain security by making its SBOM-based dependency scanner generally available. The scanner builds a software bill of materials across ecosystems such as Maven, npm, NuGet, PyPI, Go, and Cargo, then flags risky components before release. Automatic dependency resolution generates lockfiles or dependency graphs when projects have not committed them, with manifest scanning as a fallback. Security configuration profiles let platform teams switch on secret detection, SAST, and dependency scanning through policies instead of editing each project’s CI. For self-hosted environments, the GitLab Duo Agent Platform can run on open-source models or connect to providers like Claude and Gemini, allowing agentic AI DevOps even in restricted networks. Together, these features support a model where AI agents continuously monitor dependencies and policies, so human reviewers focus on design choices rather than hunting for outdated libraries.
AWS Bedrock AgentCore and Continuous DevOps Agents
AWS is promoting a complementary vision where agents run continuously across cloud environments. Its DevOps Agent, now generally available, pulls signals from services like CloudWatch and third-party observability tools, then diagnoses incidents, checks release readiness, and verifies builds in an isolated AWS-managed environment. AWS Continuum adds another layer: a set of agents focused on code vulnerabilities and penetration testing, which not only scan for issues but attempt exploits in a sandbox and suggest fixes such as network changes or patches. On the platform side, Amazon Bedrock AgentCore gives enterprises a foundation for custom agents with a managed knowledge base, web search, and controlled access to paid data sources. Matt Wood said the cost of AI “continues to decrease year by year” when normalized for a given level of intelligence, underlining AWS’s expectation that continuous agents will become a routine part of DevOps workflows.
Trust, Safety, and the Future of Agentic AI DevOps
As AI agents take on secrets management, automated code review, and supply chain security, the central question is trust. Enterprises are delegating high-risk tasks—credential protection, pen testing, dependency scanning—to systems that act without a human in every loop. GitLab’s approach is to keep security and automation on the same platform as the code, with shared governance, audit trails, and branch protection rules. Manav Khurana argues that “when security, automation, and governance share the same platform as the code, teams can move fast on AI without losing control of what ships.” AWS is focusing on safe delegation too, with identity-aware data access in Context, isolated test environments for DevOps Agent, and managed guardrails for Bedrock AgentCore. The likely next step is hybrid control: AI agents handle routine enforcement and triage, while humans set policies, review high-impact changes, and decide how far to extend autonomy into critical infrastructure.
