AI Coding Agents Are Breaking Big Codebases — Dan Adler, Sourcegraph

This talk introduces the problem of codebase decay caused by the tidal wave of AI-generated code, particularly in large, long-lived enterprise codebases. It argues that context and large-scale code visibility are critical infrastructure, not just model quality, for effective AI agent deployment. The talk introduces Sourcegraph's Agentic Batch Changes, a system designed to make and audit changes across thousands of repositories from a single prompt, addressing the challenge of managing AI-driven modifications at scale.