Crafting an Enterprise‑Scale Agent Mesh
summary
An enterprise agent mesh is a working platform not a paper design: Grid Dynamics reports a production system used daily that started bottom‑up with agents for sales and client profiling and grew into a agent marketplace plus an agent orchestrator that routes requests, enforces token delegation / SSO, provides unified memory, sandboxes generated code, favors A2A agent-to-agent connections over MCP tooling, and adds dynamic UI personalization and recursive context requests for missing variables.
AI-written summary · about
chapters (11)
- 0:00Enterprise agentic systems intro
- 2:03Why enterprise AI fails
- 3:47Building agent mesh bottom-up
- 6:14Non-negotiable agent controls
- 7:15AI harness core components
- 8:59A2A vs MCP for workflows
- 11:27Sandboxing and security
- 12:52Agent delegation and tokens
- 14:15Role-based access controls
- 14:57Dynamic agent UI
- 15:17Recursive LLMs for context
related talks
29mAgentic AI Meets Event-Driven Architecture
45mAn Agent Walks Into a Bar... and Meets Another Agent: Multi-Agent Systems in Action
43mArchitectural Blueprints for AI Agents
22mHOW TO BUILD YOUR OWN AI AGENT: THE COMPLETE ARCHITECTURE FOR ENGINEERING TEAMS
40mBuilding Agents with Agent Bricks and MCP
41mPatterns & Practices for building Multi-Agent Systems
24mAI Agent Swarms Are Amazing
15mEngineering Production-Grade Multi-Agent Systems in the Cloud