The InfoQ Podcast Platform Engineering for AI: Scaling Agents and MCP at LinkedIn
Dec 10, 2025
Karthik Ramgopal, LinkedIn's platform engineering lead, and Prince Valluri, an engineer focused on developer experience, dive into the world of AI agents. They discuss LinkedIn's unified agent platform, designed to enhance security and scalability. The duo explains the difference between foreground and background agents, and how these tools reduce developer toil. They also highlight the Model Context Protocol (MCP) as crucial for standardizing interactions across systems. Practical insights on improving developer experience and effective agent orchestration round out this compelling discussion.
AI Snips
Chapters
Transcript
Episode notes
Agents As A New Execution Model
- AI agents are a new execution model requiring platform-level support similar to compute or storage infrastructure.
- Treat agentic systems as shared, reliable, and trustable infrastructure rather than one-off features.
Use Spec-Driven Developer Intent
- Express developer intent as a structured spec that defines tasks, tools, and acceptance criteria.
- Use specs as a contract so agents plan deterministically and reviewers understand expected outcomes.
Safe Remote Sandboxes For Execution
- LinkedIn runs agents in remote sandboxes with limited permissions and controlled access to systems.
- Agents execute with platform-managed authentication and produce pull requests rather than directly committing changes.