Industry experts are calling for the implementation of continuous runtime verification for autonomous AI systems that can execute tool calls [1].
This shift is critical because autonomous agents can modify code and operate without direct human supervision. Without a dedicated verification layer, these systems risk introducing bugs or security breaches that could lead to significant business-impacting failures [1, 2, 3].
Ken Ziegler, CEO of Leapwork, said, "Continuous validation is the next frontier for software QA in the wake of agentic AI development" [2]. The need for such oversight has grown as AI capabilities evolve. For example, OpenAI released GPT-5.3 Codex on Feb. 5, 2026 [4], a model that documentation noted was instrumental in creating itself [4].
While some descriptions suggest these systems can plan and adapt without constant human intervention [5], other experts said that once an agent can execute tool calls, they require continuous oversight [1]. This tension highlights a gap in current software development life cycles, which were not designed for agents that function as active workflows rather than static tools [3].
Investment in these safety layers is already increasing. RevEng.AI recently raised $15 million in a Series A round to verify the security and integrity of software generated by AI [6]. This funding reflects a broader industry trend toward treating AI quality as a systems-thinking challenge rather than a simple coding task [2].
Experts said that the lack of a verification layer creates a bottleneck for enterprise adoption. If companies cannot guarantee the safety of an agent's autonomous decisions, the potential for scale is limited by the need for manual human review [1, 3].
“Continuous validation is the next frontier for software QA in the wake of agentic AI development.”
The transition from AI as a productivity tool to AI as an autonomous agent creates a fundamental shift in risk management. While traditional software is tested before deployment, agentic AI requires 'runtime' verification because its behavior can change dynamically based on the tools it calls. This necessitates a new infrastructure for software quality assurance that operates in real-time to prevent autonomous errors from cascading into systemic failures.
