500 Skills, Zero Fine-Tuning: LinkedIn's Playbook for AI Agents — Ajay Prakash, LinkedIn
Summary
LinkedIn addressed the challenge of coding agents (LLMs) lacking context within massive, proprietary enterprise codebases. The solution involves 'contextual agent playbooks and tools' managed by an internal MCP (Model Context Platform) server. Instead of feeding all tools into the context, the system uses three meta-tools—Search, Get Schema, and Execute—to scale to thousands of tools and playbooks. Playbooks provide self-contained, structured instructions, enabling agents to perform complex, multi-step tasks reliably, and incorporating a self-improving loop where agents update stale documentation.
Key takeaways
-
Focus on Reliability and Quality from Day One
20:00
The system's success was predicated on prioritizing quality and reliability over mere productivity, ensuring the infrastructure does not degrade as the organization scales its use of AI agents.
-
Build Dedicated Infrastructure for Agents
20:10
In a large enterprise, simply providing the latest AI models and tools is insufficient; a dedicated, robust infrastructure is required to manage and guide agent operations within the internal system context.
Technical details
-
Contextual Agent Playbooks and Tools
410s
Playbooks are instructions published as tools, invoked like any other tool, and kept self-contained. They allow agents to access organizational context and service knowledge, solving the problem of scattered, stale documentation (wikis, Slack threads).
-
Scaling Agent Context (MCP)
1030s
To scale beyond the 30-40 tool limit of traditional MCPs, LinkedIn replaced the full surface with three meta-tools: Search (to find relevant tools/playbooks), Get Schema (to retrieve details), and Execute (to run the tool/playbook). This allows scaling to thousands of tools and playbooks.
-
Agent Workflow Example
0s
An agent can handle an on-call incident end-to-end: fetching debugging instructions, retrieving logs/metrics, identifying the root cause, proposing mitigation steps, applying actions (upon human confirmation), updating the incident system, and creating a PR for the underlying fix, all in minutes.
-
Playbook Principles
600s
Playbooks must be self-contained (addressing one specific task) and should be broken into multiple smaller, referenced playbooks. This ensures reusability and progressive discovery of context.
-
Self-Improving Loop
800s
The system incorporates a self-learning flywheel: agents are encouraged to identify outdated information or discrepancies when using a playbook, then automatically check out the repository, update the playbook, and create a PR to keep the knowledge base fresh.
Mentioned resources
- Cloud Code
- GitHub Copilot
- MCP (Model Context Platform)
- Jira
- Slack
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.