What is it about?

Large software platforms are often built across dozens or hundreds of separate code repositories maintained by different engineering teams. As developers push daily updates, individual changes can gradually stray from the system's intended blueprint—a problem known as architectural drift. Our research introduces Sentinel, a system of coordinated AI agents that continuously monitors code updates (commits) across multiple repositories. By grouping scattered changes into broader development initiatives and comparing them against documented architectural decisions, Sentinel automatically detects design mismatches and alerts developers before problematic code is merged.

Featured Image

Why is it important?

Traditional architecture reviews rely on manual oversight or rigid static rules that struggle to keep pace with modern microservices and AI-accelerated software development. Sentinel shifts architectural governance from periodic, after-the-fact audits to continuous, automated intelligence. Tested in a live production environment spanning more than 120 microservices and 20 engineering teams, Sentinel detected real architectural drift within an average of 12 minutes of a code commit with an 83% true positive rate. This allows engineering organizations to ship software rapidly while maintaining long-term reliability and system coherence.

Perspectives

"In large multi-repository organizations, architectural drift rarely happens because of a single bad decision—it happens quietly across dozens of pull requests that each look fine in isolation. We built Sentinel because humans simply cannot read every commit across over a hundred repositories to spot emerging structural mismatches. Seeing Sentinel connect related changes across separate codebases and flag subtle design violations within minutes showed us that autonomous agents can serve as practical, everyday guardians of software architecture."

Uday Bhanu Prasad Kanagala

Read the Original

This page is a summary of: Sentinel: Autonomous Architectural Governance Through Commit Intelligence Across Multi-Repository Systems, May 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3786335.3813195.
You can read the full text:

Read

Contributors

The following have contributed to this page