Stephanie Jarmak is an applied research scientist at Sourcegraph who brings empirical rigor to the coding-agent space: she ran 1,281 agent runs across 40+ repositories to identify five repeatable failure modes that break agents in large codebases, and built CodeScaleBench to test coding agents on enterprise-scale, multi-repo software engineering tasks. Attendees wanting hard data — not hype — on what actually breaks coding agents at scale, and which infrastructure fixes (code search, structured indexing, retrieval) work, will find her sessions unusually grounded.

Public activity researched automatically · as of Jun 2026