After large language models, the next breakthrough in generative artificial intelligence could well come from autonomous agents capable of performing end-to-end tasks. In a DevOps pipeline, these agents will no longer simply write or complete code—they will orchestrate workflows, detect anomalies, fix bugs, optimize performance, and even manage infrastructure changes.

Contrary to recurring fears, the emergence of agent-based AI is not intended to replace developers or DevOps engineers, according to the software platform GitLab. Rather, it is meant to allow them to refocus on high-value-added activities: architecture, quality, security, and design. A transformation of teams is on the horizon: less repetitive engineering, and more oversight, planning, risk management, and innovation.

Automated Pipelines

This new generation of AI agents will initially be used in CI/CD (continuous integration/continuous deployment) pipelines. Human intervention will no longer be required to run tests or deploy updates. AI will soon be able to manage the entire process while learning from each release. The expected result: faster, more reliable, and more consistent pipelines. This will lead to shorter software delivery cycles and higher-quality software.

The same logic applies to infrastructure. Thanks to real-time monitoring, agents will be able to automatically adjust resources based on workload, anticipate failures, and maintain service continuity without constant supervision. This promises cost optimization and increased resilience.

However, it is in incident management that these systems could truly realize their full potential. Rather than generating a flood of alerts, agents will analyze anomalies, assess their impact, and take immediate action—such as rollbacks, reconfigurations, or redeployments—thereby reducing downtime, streamlining development, and ultimately saving time.

Finally, security should also benefit from this intelligent automation. By detecting and correcting vulnerabilities at an early stage, agents will provide active, continuous protection capable of stopping threats before they spread.

The Challenge of Human Acceptance

But this shift also presents new challenges. The first concerns governance: how can we ensure that an agent complies with security policies, compliance requirements, and internal standards? To address this issue, GitHub highlights safeguards such as branch protection and the requirement for human approval before CI/CD workflows are executed.

On a larger scale, the rise of “self-governing” systems is raising questions about their reliability. Academic research is therefore calling for the establishment of an “AgentOps” framework dedicated to monitoring, validating, and correcting agent behavior. The goal: to ensure transparent, controlled, and secure execution.

Finally, there is the issue of integration. Deploying agents into existing environments—which are often complex, heterogeneous, and sometimes aging—will require a high level of DevOps maturity, as well as efforts to address internal concerns. Beyond the technical challenges, human acceptance will indeed be one of the main challenges of this new era of augmented DevOps.