AI news for tracking tools, models, and market shifts

AWS and authors propose agentic overlays to transform legacy REST services into A2A-capable agents without rewriting code, reducing agent sprawl.

AWS ML Blog post details building a governed, serverless data mesh for agentic AI, emphasizing secure, scalable data foundations.

AWS ML Blog shows how to deploy SeedVR2 on SageMaker AI for video upscaling, covering architecture, deployment, and performance comparisons.

AWS blog details optimizing SageMaker AI training on NVIDIA Blackwell GPUs, covering batch size, precision, and activation checkpointing for models up to 64B parameters on P6-B200 instances.

Loka built a low-latency voice agent using Amazon Nova 2 Sonic, reducing latency and improving natural conversation flow to enhance customer experience and reduce support costs.

AWS and Snowflake integrate semantic views with Amazon Quick for AI-powered BI, enabling natural-language queries on governed data.

AWS ML Blog post demonstrates building a protein research copilot using Amazon Bedrock AgentCore, combining natural language query parsing, vector similarity search over protein embeddings, and AI-generated summaries.

AWS ML Blog post details patterns for production-ready multi-tenant systems using Amazon Bedrock AgentCore, demonstrated through healthcare AI agents serving multiple clinics and hospitals.

AWS blog details deploying ComfyUI workflows on SageMaker AI processing jobs for batch image generation using GPU acceleration and AWS CDK infrastructure.

AWS ML Blog details building a multimodal AI system for searchable aerial imagery using Amazon Bedrock and OpenSearch Serverless. Amazon Nova Multimodal Embeddings achieved highest F1 scores in evaluations.

Ampersend built a pay-per-intelligence routing layer on Amazon Bedrock AgentCore Payments, enabling AI agents to autonomously route tasks to the most effective model, pay per request, and operate within budgets.

AWS announces innovations at Summit NYC to give AI agents safe, scalable access to context across data lakes, warehouses, and streams for trusted decisions.

Agent-EvalKit is an open-source toolkit for systematically evaluating AI agents. It integrates with coding assistants like Claude Code and supports six evaluation phases, demonstrated with a travel research agent on Amazon Bedrock.

AWS ML Blog details an intelligent document processing pipeline using Amazon Bedrock, including BDA for extraction, Strands Agent for coordination, and Knowledge Base for contextual understanding.

Amazon Quick and Cisco Webex MCP servers enable a custom assistant that preps for meetings by reviewing past summaries and transcripts, and follows up by summarizing discussions and identifying action items.

Rocket Close built Supercharger using Strands Agents, LLMs, Amazon Bedrock, and MCP tools to optimize title operations, achieving significant business impact.

AWS ML Blog details building context-rich research agents using Deep Agents and Bedrock AgentCore, enabling isolated execution environments for multi-step AI workflows. The walkthrough includes deployment to Bedrock AgentCore Runtime via CLI.

AWS ML Blog introduces Strands Evals for AI agent failure detection and root cause analysis, providing structured output with categorized failures, confidence scores, causal chains, and fix recommendations.

Amazon SageMaker AI launches container caching for inference, reducing scale-out latency by up to 2x for generative AI models.

Amazon Bedrock Guardrails launches InvokeGuardrailChecks API for applying individual safety checks in agentic AI apps without creating guardrail resources.