A growing queue of AI-generated pull requests does not, by itself, tell a team why software is taking too long to reach production. Adding reviewers may clear one handoff without changing who can propose, implement and deliver a feature. The more consequential question is whether the organization has redesigned that path or simply added coding agents to an existing process. Faster work at one stage is useful only if the rest of the delivery system can absorb it.
Adir Ben-Yehuda, CEO and co-founder of AutonomyAI, joins Mike Vizard to argue that the bottleneck extends beyond code review. Drawing on his company’s conversations with software organizations, he describes a gap between individual engineering speed and company-level delivery. His proposed division of work separates architecture, system scalability and stability, and the application layer. Product managers and designers would take more responsibility for application changes, while engineers concentrate on the foundations that let those changes run safely.
That arrangement depends on more than giving another employee access to a foundation model. Ben-Yehuda says agents need context specific to the organization’s codebase, tools and practices, along with explicit roles and responsibilities. Multiple agents also need boundaries that keep their work from conflicting. In his view, DevOps and platform teams become central to building the infrastructure and guardrails that make broader application creation possible. The task is to establish an environment people can use repeatedly, rather than have an engineer supervise every individual request.
Delivery also creates work beyond the engineering pipeline. Ben-Yehuda points to sales enablement as an example: releasing features more frequently is not enough if customer-facing teams cannot keep up with what changed. He suggests generating short explanations of merged changes as part of the feedback loop. For teams still building confidence, his advice is to begin with a small, measurable task, such as improving product requirements, and expand as trust develops. The immediate opportunity is not to eliminate review, but to decide which changes to responsibilities, context and feedback would make the whole workflow more effective.
Mike Vizard is a veteran IT journalist with more than 25 years of experience covering the technology industry, having previously served as Editor-in-Chief of both CRN and InfoWorld and as editorial director for Ziff-Davis Enterprise, where he oversaw titles including eWEEK, CIO Insight and Baseline. Over his career he has also edited or contributed to a wide range of enterprise technology publications, including IT Business Edge, Channel Insider, ComputerWorld, TMCNet and Digital Review, and he later led editorial for CTOEdge.com. His reporting and analysis span software development, cloud computing, cybersecurity, IT channel strategy and, more recently, artificial intelligence and DevOps practices. A recognized voice in enterprise IT journalism, Vizard is known for tracking emerging technology trends as they move from early adoption into mainstream enterprise use. He now serves as Chief Content Officer for Techstrong Group, where he oversees editorial strategy across the full network — DevOps.com, Security Boulevard, Cloud Native Now, Digital CxO, Techstrong.ai, TechStrong.IT, Techstrong Semi and PlatformEngineering.com — in addition to writing and hosting content for Techstrong TV and the Techstrong Gang podcast.