Skip to content
AI News HubLIVE

This edition’s highlights

Robotics

How NVIDIA Open Source Tools Are Helping Researchers Build Next-Generation 6G Networks

  • NVIDIA AI Aerial and Sionna see 10x developer adoption growth in one year, with over 1 million downloads of open source tools and 8,000+ institutions using them.
  • King's College London uses Sionna for goal-oriented semantic communications, reducing robot task completion time by up to 52.6%.
NVIDIA BlogIn-site articleHow NVIDIA Open Source Tools Are Helping Researchers Build Next-Generation 6G Networks

How Greenlight Financial built an autonomous SOC solution on Amazon Bedrock AgentCore

  • Luna uses a tiered supervisor architecture with Tier 1 and Tier 2, orchestrating more than 30 specialized agents to handle 3,000-4,000 cases per month.
  • AgentCore memory enables episodic and long-term memory, boosting autonomous resolution rate from 83% to over 90%.
AWS Machine Learning BlogIn-site articleHow Greenlight Financial built an autonomous SOC solution on Amazon Bedrock AgentCore
Chips

SageMaker Inference now supports Multi-Instance GPU to maximize GPU utilization for model endpoints

  • SageMaker Inference now supports GPU partitioning using NVIDIA MIG, creating hardware-isolated partitions on a single GPU.
  • Two optional API fields, AcceleratorPartitionLayouts and AcceleratorPartitionType, let you place one model per partition within inference components without managing the underlying platform.
AWS Machine Learning BlogIn-site articleSageMaker Inference now supports Multi-Instance GPU to maximize GPU utilization for model endpoints
Agents

Govern AI agent access with Okta Cross App Access and Amazon Bedrock AgentCore – Part 1

  • Okta XAA acts as the governance plane, determining which agent can act for which user against which resource.
  • AgentCore Identity is the execution plane, securely redeeming short-lived access tokens without exposing client secrets to agent code.
AWS Machine Learning BlogIn-site articleGovern AI agent access with Okta Cross App Access and Amazon Bedrock AgentCore – Part 1

Beyond code generation: A developer’s guide to agentic engineering

  • More commits do not automatically translate to proportional productivity gains; review bottlenecks can grow.
  • Agentic engineering spans planning, building, verifying, and observing, with Copilot support at each stage.
GitHub AI & MLIn-site articleBeyond code generation: A developer’s guide to agentic engineering

Quoting Nilay Patel

  • Patel says AI products do not yet deliver value that justifies data center noise and other costs
  • He questions whether agents are real or being used as anyone intends, and notes companies spend more than they make
Simon Willison's WeblogIn-site articleQuoting Nilay Patel

Extend Claude Code to remote sessions on Amazon Bedrock AgentCore with security controls

  • Each task runs in its own AgentCore runtime microVM with sanitized memory, and model calls use a private Amazon Bedrock endpoint signed by the runtime IAM role.
  • GitHub access is managed through AgentCore Identity with short-lived tokens (8-hour expiry), avoiding long-lived credentials and limiting access to authorized repositories.
AWS Machine Learning BlogIn-site articleExtend Claude Code to remote sessions on Amazon Bedrock AgentCore with security controls

Taxonomy of agentic primitives in Amazon Quick

  • Agentic work is decomposed into four primitives: triggers (When and What), agents (Who), and capabilities (How), with learned context accumulating over time.
  • Triggers include human-initiated, agent-delegated, scheduled, and event-driven types, each conditioning execution behavior.
AWS Machine Learning BlogIn-site articleTaxonomy of agentic primitives in Amazon Quick

Serverless MCP server for AI agent data lake access on Amazon ECS Fargate

  • Architecture separates concerns into client, MCP server, tools, and knowledge-base refresh layers, each independently scalable.
  • Metadata-driven approach eliminates per-dataset integration code; adding a dataset requires only a JSON file.
AWS Machine Learning BlogIn-site articleServerless MCP server for AI agent data lake access on Amazon ECS Fargate
Tools

Read, don't copy: replacing file and API feeds with OpenSharing - Part 1 (The Why)

  • Traditional SFTP/API feeds create pipeline maintenance, backfill, and duplicate storage costs as customers grow.
  • OpenSharing uses open table formats like Delta Lake and Iceberg for in-place reads and change data feeds.
Databricks BlogIn-site articleRead, don't copy: replacing file and API feeds with OpenSharing - Part 1 (The Why)

You have reached the end of this edition.

Your reading list
Other updates (2)
Agents

Govern AI agent access with Okta Cross App Access and Amazon Bedrock AgentCore – Part 2

In Part 2, you scale the governed agent pattern to an enterprise: Amazon Bedrock AgentCore Gateway becomes the single front door and redeems Okta's grant, Cedar policies bind each tool to the agent, a second agent runs on Amazon EKS, and AgentCore Observability traces every allow and deny beside Okta's log in Amazon CloudWatch.

AWS Machine Learning BlogIn-site articleGovern AI agent access with Okta Cross App Access and Amazon Bedrock AgentCore – Part 2

Taxbit Assist: A grounded, role-aware AI agent on Amazon Bedrock AgentCore

Taxbit built Taxbit Assist, a cited, permission-aware AI assistant embedded in its enterprise dashboard, on Amazon Bedrock AgentCore. This post shows how they grounded answers in a curated knowledge base, enforced role-aware and region-compliant safety, and continuously measured answer quality on live production traffic.

AWS Machine Learning BlogIn-site articleTaxbit Assist: A grounded, role-aware AI agent on Amazon Bedrock AgentCore