Agentic AI DevOps: From Code Suggestions to Autonomous Security Decisions
Agentic AI DevOps is the use of autonomous, goal-driven AI agents that continuously watch code, infrastructure, and pipelines, then decide and act on changes themselves, covering tasks from credential protection and code review to dependency and supply chain security without waiting for manual prompts. Recent releases from AWS and GitLab show this shift. AWS Bedrock AgentCore and related DevOps agents are built to run in the background, scanning environments, validating builds, and proposing or applying fixes to outages and vulnerabilities. GitLab 19.0 moves its DevOps AI agents beyond code generation into AI secrets management, autonomous code review on merge requests, and supply chain security automation through SBOM-based scanning. Both platforms aim to embed DevOps AI agents where developers already work, so AI decisions are logged, reviewable, and constrained by existing controls. Yet trust in autonomous decisions remains the main barrier to wider adoption.
AWS Continuum and AgentCore: Continuous Agentic DevOps in the Background
AWS is pushing toward continuous agentic DevOps with its Continuum agents and Bedrock AgentCore platform. Continuum for code vulnerabilities, now in closed preview, scans AWS environments, prioritizes vulnerabilities that are reachable in production paths, and even demonstrates exploits in a sandbox before proposing fixes such as network changes or patches. The existing AWS Security Agent is being folded into “Continuum pen testing” and “Continuum code scanning,” reinforcing the focus on ongoing security continuity. The AWS DevOps Agent, already generally available, brings release management features that run software in isolated AWS-managed environments to assess readiness and verify builds. According to AWS chief AI and technology officer Matt Wood, the company sees AI tools “operating continuously in the background, rather than being used on demand,” signalling a move from interactive helpers to always-on DevOps AI agents grounded in observability data and external tools.
GitLab 19.0: Agentic AI for Secrets, Merge Requests, and Supply Chain Security
GitLab 19.0 pushes agentic AI deeper into security-sensitive workflows. The new GitLab Secrets Manager, in public beta, centralizes AI secrets management by storing credentials in the same platform that runs code and CI/CD, tying each secret to the jobs authorized to use it and logging every access. It integrates with HashiCorp Vault, AWS Secrets Manager, Azure Key Vault, and Google Cloud Secret Manager, rather than replacing them. On merge requests, the Developer Flow agent now handles reviewer feedback, splits oversized changes, resolves conflicts, and reads an AGENTS.md file so autonomous code review follows project-specific standards. A Resolve with Duo button (beta) evaluates both branches, commits a proposed fix, and leaves a summary comment. On the supply chain side, an SBOM-based dependency scanner is now generally available across major ecosystems, backing supply chain security automation with automatic lockfile and dependency graph generation.
Trust, Governance, and the Path to Safer DevOps AI Agents
Both AWS and GitLab are confronting a central problem: enterprises do not yet fully trust autonomous AI in security-critical DevOps workflows. GitLab’s Manav Khurana notes that “AI made it faster to generate code, but it didn’t make it easier to trust or secure it at scale,” highlighting why controls and auditability now matter more than raw model power. GitLab anchors DevOps AI agents inside its existing project, group, and policy model so secrets, merges, and scans inherit current access rules and audit trails. AWS, meanwhile, is building guardrails into AgentCore and Continuum through identity-aware data access, isolated execution environments, and tool integrations rather than unconstrained actions. Agentic AI DevOps is shifting from code generation toward autonomous decision-making, but the platforms that succeed will be those that treat explainability, rollback, and human oversight as first-class features, not optional add-ons.
