Ericsson's Multi-Agent Code Reviewer Hits 96% Accuracy, With a Third of Its Correct Findings Rated Must-Fix

arXiv 2609.15877·medium signalXBlueskyLinkedInCopy link

Ericsson built and evaluated a multi-agent code review solution in an industrial setting under the Design Science Research Process, combining specialized agent skills with project-specific contextual knowledge to identify antipatterns across readability, maintainability, reliability and performance. Across several commits it generated more than 200 issues that were manually validated by the case company's own developers, reaching 96% accuracy in correctly identifying issues. Roughly 69% of the correct findings were rated important and about 33% severe enough that they must be fixed, which is one of the few published industrial numbers on agentic review precision.