Why Agentic Coding Needs Governance at Scale

The DevOps transformation that enterprises began a decade ago remains incomplete, giving rise to a new phenomenon known as agentic coding. Traditional builds, test suites, and release processes continue to be hindered by manual steps, resulting in inefficient and costly operations. Moritz Plassnig, the recently reinstated CEO of CloudBees, shared his observations during a five-year hiatus at Immuta, and returned to the delivery side of the problem, recognizing that agentic coding represents a second chance to fulfill the original DevOps promise.

The issue at hand is not about teams debating the validity of DevOps, but rather the fact that most companies failed to implement it comprehensively, leaving shortcuts that now limit the speed of AI agents. To address this, Plassnig proposes a straightforward solution: encode existing organizational rules, such as two-human code review and individual policies, into versioned policies that a centralized governance engine can enforce across various coding tools like CloudBees, Jenkins, GitHub, or GitLab.

A striking example comes from a major bank, where the technology planning horizon has shrunk from a five-year outlook to two to three years. This rapid change necessitates a move towards standardization and the use of an approved tool menu, controlled by a governance layer. As Plassnig points out, if automated code generation is no longer the limiting factor, the next bottleneck is ensuring safe deployment of software.

He emphasizes that AI-scale development requires AI-scale governance, otherwise, the disparity will become disastrous. The risk of supply chain vulnerabilities grows with each automated step, and companies that continue to rely on manual policies will inevitably learn the hard way that AI agents scale faster than any human review process ever could.