Shipping Enterprise AI Agents with Amazon Bedrock AgentCore
summary
Amazon Bedrock AgentCore offers production primitives: agent core runtime runs serverless agent sessions (scale-from-zero, sessions up to 8 hours, multimodal payloads); agent core identity handles inbound/outbound OAuth2 flows with Okta Azure AD Cognito and issues short‑lived creds; agent core gateway exposes OpenAPI and Lambda targets, indexes tool catalogs (360 tools → semantic search returning ~4 relevant tools) to cut context and cost; agent core memory stores short‑term (configurable up to 365 days) and extracts long‑term preferences; plus browser, code interpreter (JS TS Python), CloudWatch/OpenTelemetry observability; preview as of Sep 5.
AI-written summary · about
chapters (12)
- 0:00Introduction and Overview
- 0:36The Evolution of Agent AI
- 3:06AWS Vision for Agents
- 3:56Fundamentals of Agent Core
- 8:34Agent Core Runtime
- 12:21Agent Core Identity
- 15:50Agent Core Gateway
- 20:35Agent Core Memory
- 23:52Agent Core Browser
- 26:34Agent Core Code Interpreter
- 28:19Agent Core Observability
- 30:08Conclusion and Next Steps
related talks
44mAgentic AI at Sage with AgentCore
29mBuild and scale reliable multi-agent systems on AWS
24mBuilding GenAI applications with Amazon Bedrock
47mAgents and Memory Architecture
19mBuilding Agents at Cloud Scale
39mBuilding Safe AI Agents
15mAgents Don’t Ship Themselves
31mAI Agent Orchestration for Enterprise Arch with AWS Bedrock, Flyte & LangGraph