Lessons From Building and Deploying a Multi-Agent System at Scale: An...
summary
Deployed a multi‑agent AI platform across multiple European countries for Deutsche Telekom, processing 4.5 million conversations/month and cutting human handovers by 38%; live since 2023 and winner of a best‑bot award in 2025. Platform uses a JVM‑first authoring approach with Arc DSL, exposes agents as an agent as Kubernetes resource for multi‑tenant deployments, integrates MCP tool calling and KK2‑style models, and proves an open, sovereign, production‑grade agentic stack that actually runs at scale.
AI-written summary · about
chapters (11)
- 0:00Multi-agent platform journey
- 2:06Built multi-agent platform at scale
- 3:33Business outcomes and wins
- 5:21Development speed, scale, cost
- 7:09Empowering JVM engineers with ARC DSL
- 8:55Scaling agents with Kubernetes tools
- 10:18Bridging teams with ARC ADL
- 13:32AI infrastructure complexity
- 16:22Live demo: agent orchestration in e-commerce
- 21:58Future of agentic compute and AGC
- 24:07Q&A
related talks
01h 00mBuilding a Polyglot Multi-Agent AI Architecture
50mScalable Innovation: Exploring Multi-Agent Systems
27mMulti-Agent AI: Transforming Business Operations and Value Creation
15mThe Responsible Path to Hyperproductivity with AI Agents
27mFrom Chatbots to Agents: Orchestrating Agentic AI on Existing Enterprise Kubernetes Infrastructure.
13mFrom Monoliths to Microminds: Building Resilient AI Platforms
25mFrontiers of Agentic AI
06mKeynote: Building Agents at Scale