AI coding agents are powerful out of the box, but they don't know your company. They can't navigate your services, understand your frameworks, query your data systems, or follow your organizational processes. At LinkedIn, we faced this gap as we scaled AI-assisted development across the entire organization.
This talk shares how we built CAPT (Contextual Agent Playbooks and Tools), an MCP-based system that started as a way to give coding agents organizational context, and grew into the context platform for all of LinkedIn. CAPT uses a two-layer architecture: a meta-tools layer that lets agents dynamically discover and invoke hundreds of internal tools via tags, and a skills layer: executable workflows that encode institutional knowledge as step-by-step instructions agents can follow autonomously.
What surprised us was who adopted it, and how. Engineers use MCP tools directly with their agents for debugging, code review, querying internal systems, and navigating infrastructure. On top of that, they capture their recurring workflows as skills that reference those MCP tools, turning tribal knowledge into reusable automation. Product managers and designers used it for rapid prototyping and converting their ideas into production code. One of the biggest wins: CAPT enabled anyone (product managers, data scientists, program managers) to do data analysis with natural language, querying LinkedIn's data platforms without writing SQL. The same MCP infrastructure served every role because organizational context isn't role-specific. It's shared.Rather than building a custom AI agent, we bet on the Model Context Protocol (MCP) as the integration standard, making CAPT compatible with any MCP-aware agent while tapping into a growing community ecosystem.
I'll walk through real production use cases across roles, with concrete results: 70% faster issue triage, 3x faster data-to-insight cycles, and company-wide adoption with 500+ community-authored skills. The talk includes a live demo of an agent going from generic to org-aware using CAPT.
You'll leave with a practical blueprint for building your own organizational context layer, not just for engineers, but for your entire company.
Main Takeaways
AI agents can't be effective at companies like LinkedIn without organizational context. Internal tooling, institutional knowledge, and company-specific workflows don't come built into any model. Context engineering is the discipline of making that knowledge available to agents, and getting it right is what turns off-the-shelf agents into productive contributors inside your organization.
Deploying MCP at scale is a different problem than getting it working. We'll cover the real challenges: tool discovery across hundreds of internal services, authentication and access control, keeping tools reliable as underlying systems change, and the meta-tool architecture we built to solve them.
Adoption at scale comes from ease of use. CAPT succeeded company-wide because it required zero configuration, worked with any MCP-aware agent, and let anyone (not just engineers) leverage AI for their workflows without learning new tools or writing code.
