Agentic AI
GitLab survey reveals governance gap in agentic AI development
Most organizations are adopting AI coding tools faster than they can govern them, creating infrastructure and risk management challenges at machine scale.
The New Stack
A GitLab survey of 1,500+ developers and tech leaders reveals a critical tension in enterprise AI adoption: while 60% report that AI coding ROI has exceeded expectations and 78% see faster code commits, only 21% see productivity gains beyond code generation itself.
The underlying problem is infrastructure mismatch. Most organizations bolted AI tools onto systems built for human-scale development. Git backends, CI/CD pipelines, and governance frameworks can't keep pace with agent operations running at machine scale. The result: platform reliability breaks down, security exposure widens across dependencies, and costs spiral as agents consume tokens inefficiently.
Governance lags adoption#
The survey found that 80% of organizations adopted AI tools faster than they developed policies to manage them. Even more striking: 82% worry that AI-generated code is creating "a new form of technical debt" their teams aren't prepared to handle. Only 28% say their development tools are fully integrated with shared data and workflows.
Four pillars of agentic infrastructure#
The path forward requires building "agentic infrastructure"—not just agentic code—across four layers:
-
Machine-scale execution: Git and CI/CD systems designed for production incidents that take days to debug need to become systems where root cause is traceable in minutes.
-
Context traveling with code: Agents perform only as well as their context. A unified graph connecting code, work items, pipelines, security findings, and production signals prevents agents from operating with "artificial confidence."
-
Governance in the flow: Agent actions must be tied to identity, logged against policy, and auditable. Low-risk changes move fast; higher-risk ones trigger review. This becomes critical for regulated industries like automotive.
-
Orchestration: A coordination layer that determines which agents run, in what order, and how failures are managed—turning separate capabilities into a coherent system.
The shift ahead is telling: 85% of survey respondents expect the next phase of AI in software to focus less on code generation and more on governing it. That reflects maturing enterprise thinking—AI as a foundational, trustable capability rather than just a productivity boost.
