ODSP2245 Pre-recorded On demand
Give AI agents grounded context for complex codebases
Ray Ploski
Summary
From Microsoft's AI summary
Ray Ploski of CoreStory described how AI coding agents lack system-wide context in large codebases. CoreStory ingests source code, builds an intelligence model grounded in file and line references, and exposes it through MCP, HTTP, command-line tools and a web interface. A benchmark of six coding agents on 45 difficult SWE tasks showed GitHub Copilot rising from 62% to 78% with CoreStory context. A live demo used an account suspension change request, identifying authentication and billing impact and the rule to revoke active sessions. The session also covered modernisation and three operational recommendations.
Session description
Microsoft's catalogue description
AI coding agents can move quickly, but they need reliable system context before they modify a complex codebase. In this session, see how CoreStory builds persistent, code-grounded intelligence across architecture, behavior, and intent and makes it available to people and AI agents through MCP. Learn how teams use that shared context to assess impact, understand constraints, and create implementation-ready specifications for maintenance and modernization work.