Building a Polyglot Multi-Agent AI Architecture
summary
Typeform scales its AI from one feature to a company platform, moving from self‑managed Kubernetes to Amazon Bedrock AgentCore to host a supervisor agent plus domain agents (forms, contacts, automations) communicating over MCP/ATA, built in about four months and launched GA in April; the system handles ~2 million interactions/month, solves durable execution with microVMs with 8‑hour sessions instead of pods, starts agents in Python/LangGraph but supports polyglot HTTP agents, and adds tracing to Arize plus an LLM gateway for cost observability.
AI-written summary · about
chapters (9)
- 0:00Introduction and context
- 3:36From UI to AI-native
- 7:14Expanding AI capabilities
- 10:00Architecture and Conway's law
- 18:23Solving technical challenges
- 27:45Adopting Agent Core
- 30:32Outcomes and benefits
- 35:26Closing thoughts
- 37:34Q&A session
related talks
30mLessons From Building and Deploying a Multi-Agent System at Scale: An...
26mAI powered automation & multi-agent orchestration
15mThe Responsible Path to Hyperproductivity with AI Agents
13mFrom Monoliths to Microminds: Building Resilient AI Platforms
26mBuilding Agentic Workflows: Patterns for Orchestrating Intelligent Systems.
17mAI TechOps: Data and Infra for AI Foundation to Scale and Manage 1000s of Cohesive Agents
24mAI Agent Swarms Are Amazing
24mAI Agents in Production: Why the Platform Matters