Rootly's decision to abandon its small pull request rule is the kind of quiet admission that speaks volumes about where software engineering is headed. For years, the industry treated PR size as a proxy for risk. Small changes meant smaller blast radius, easier reviews, and a tighter feedback loop. It was a sensible heuristic when humans wrote every line. But Rootly's own data now shows that AI agents generate most of its code, and the old rule is no longer doing the work it was designed to do. That is not a failure of discipline; it is a natural evolution of the economics of code review.
The company's shift from measuring PR size to assessing blast radius is the real story here, and it is one worth paying attention to. When a human writes a 200-line PR, the review is as much about intent and context as it is about correctness. But when an AI agent produces that same diff, the reviewer's job changes. The code is likely syntactically sound and consistent with existing patterns. The real questions become: What does this touch? What can break? And how quickly can we undo it if something does? Rootly's answer is to lean on feature flags and rollback capability, which are far more meaningful safety nets than line count. That is a pragmatic, forward-looking tradeoff, and it suggests that the industry is finally moving past treating code review as a line-by-line proofreading exercise.
For our readers, this is not just a niche process tweak. It is a signal that the metrics you use to measure engineering health need to be rethought as AI tools take on more of the mechanical work. If you are still enforcing a hard cap on PR size, you are likely optimizing for a problem that no longer exists. The real risk lives in how a change interacts with the system, not how many lines it touches. Rootly's approach suggests a better question to ask in your own reviews: If this change goes sideways, how fast can we recover? That is a concrete, actionable takeaway. It reframes the conversation from "how big is this?" to "how contained is this?" and it is a much better use of a reviewer's attention.
The open question is whether this scales beyond Rootly's specific environment. Feature flags and solid rollback tooling are not free; they require a mature platform and a culture that prioritizes reversibility over perfection. But the direction is clear. The small PR rule was never sacred; it was a stand-in for something deeper. As AI agents become the primary authors of code, we should expect more teams to follow Rootly's lead and retire the heuristics that were built for a human-only workflow. The details to watch are in how the tooling evolves to support this. If rollback becomes one-click and flags become the default, then the small PR rule will look as dated as a floppy disk. If not, we may see a split between teams that can safely embrace large AI-generated diffs and those that are forced to keep the old guardrails. That is the tension worth tracking.