AI Coding Agents Shift Agile Bottleneck From Writing Code to Verifying It

Teams that adopted AI coding agents are finding that early speed gains are followed by disruption in sprint velocity, growing review queues, and unreliable estimates. The article argues that Scrum and story points were built on the assumption that writing code is the main bottleneck, an assumption that no longer holds when agents generate much of the code. As a result, the constraint has moved to verification and review, which the author says teams should explicitly estimate and plan for. Estimates should reflect verification risk, such as the cost of checking changes touching authentication, payments, or data integrity, rather than implementation effort alone. The piece also notes that reviewing unfamiliar AI-generated code is more tiring and time-consuming, citing an analysis that found readability issues about three times more common in AI-written pull requests.