Bedrock Brief 25 Mar 2026

Pleri AI

Pleri AI

— 4 min read
Bedrock Brief 25 Mar 2026

Welcome back, cloud cowboys and AI aficionados! It's been another wild week in the land of AWS, where the only thing moving faster than our instances is the pace of AI innovation.

First up, AWS is celebrating its 20th birthday this month. From humble beginnings of renting out computing power for pocket change, they've grown into a $129 billion behemoth that powers half the internet. But as they blow out the candles, there's a new party crasher in town: AI. The rise of artificial intelligence has AWS facing its toughest challenge yet, with competitors nipping at their heels like over-caffeinated Roombas. Will AWS keep its crown, or is it time for a changing of the guard? Only time (and a whole lot of GPUs) will tell.

Speaking of AI shaking things up, Anthropic just dropped a bombshell. Their Claude AI can now control computers directly, opening apps and filling spreadsheets like a ghostly intern. Meanwhile, AWS is cooking up AI agents to automate sales and business development tasks. It's like they're building a robot workforce, but instead of stealing our jobs, they're just... doing them better? The software market is freaking out, with stocks tumbling faster than a jenga tower at an earthquake convention. Is this the dawn of the AI takeover, or just another day in tech? Grab your popcorn, folks – this show's just getting started.

Fresh Cut

  • Amazon Bedrock AgentCore Runtime introduces WebRTC support, enabling developers to create low-latency voice agents for browsers and mobile apps with real-time audio and video streaming. Read announcement →
  • Amazon EKS simplifies AI workload deployment on Kubernetes by automatically managing AWS Trainium hardware resources, allowing developers to focus on model development rather than infrastructure details. Read announcement →
  • Amazon Polly introduces 10 new lifelike voices and a Bidirectional Streaming API, enabling developers to create real-time conversational AI applications with natural-sounding speech in multiple languages and accents. Read announcement →
  • AWS MCP Server now publishes operational metrics to CloudWatch and uses semantic search for Agent SOPs, giving developers better visibility into AI assistant actions and easier discovery of pre-built AWS task workflows. Read announcement →
  • AWS integrates NVIDIA's NIXL with Elastic Fabric Adapter, boosting large language model inference speed and efficiency on EC2 instances through improved KV-cache handling and inter-node communication. Read announcement →
  • Amazon Redshift now lets you use IAM Identity Center for fine-grained access control across multiple AWS regions, simplifying data security management for developers working with distributed data warehouses. Read announcement →
  • EC2 C8gn instances, powered by Graviton4 processors, offer 30% better compute performance and up to 600 Gbps network bandwidth, making them ideal for network-intensive workloads and AI/ML inference across multiple AWS regions. Read announcement →
  • Amazon Bedrock introduces GLM 5 and Minimax M2.5, two powerful AI models designed for complex reasoning and efficient task completion, enabling developers to build more sophisticated applications with advanced language understanding and problem-solving capabilities. Read announcement →
  • Developers can now access NVIDIA's open-source Nemotron 3 Super model on Amazon Bedrock, offering fast and efficient multi-agent AI capabilities for complex, long-running tasks without losing context. Read announcement →
  • Amazon Redshift speeds up new queries by up to 7x using a composition technique, allowing faster execution of BI dashboards, ETL workloads, and real-time analytics without extra cost or setup. Read announcement →

The Quarry

Evaluating AI agents for production: A practical guide to Strands Evals

Strands Evals offers a systematic way to put your AI agents through their paces before unleashing them on unsuspecting users. This open-source toolkit comes packed with pre-built evaluators and multi-turn simulation capabilities, letting you stress-test everything from basic task completion to more nuanced aspects like tone consistency. For the technically curious, Strands Evals uses a flexible "evaluator" abstraction that allows you to chain together custom evaluation logic, making it adaptable to even the quirkiest of AI agent use cases. Read blog →

More posts:

Core Sample

See It in Action: How stp.one Uses AI Agents to Transform Legal Work on AWS

Stp.one's AI-powered legal assistant transforms document management and case preparation by seamlessly accessing and analyzing vast amounts of legal documents. The system, built on AWS technologies like Amazon EKS and S3, uses the Model Context Protocol to provide lawyers with instant document summaries, simulated court scenarios, and relevant legislation searches. Notably, the integration of Aurora PostgreSQL with pgvector enables powerful semantic search capabilities, allowing lawyers to quickly find pertinent information across their entire case library. Watch video →

More videos:

Read more