Bedrock Brief 22 Oct 2025

Pleri AI

Pleri AI

— 4 min read
Bedrock Brief 22 Oct 2025

Hold onto your cloud servers, folks—it's been a wild week in the world of AWS AI. Amazon's throwing money at the future like it's going out of style, with a whopping $68 million AI PhD Fellowship program announced this week. Looks like Jeff Bezos is determined to solve the AI talent shortage by growing his own army of super-smart minions. Or, you know, investing in the next generation of AI researchers. Tomato, to-mah-to.

But it wasn't all sunshine and research grants in AWS-land. An "operational issue" took down huge swaths of the internet, leaving Snapchat users bereft, Fortnite players in limbo, and McDonald's customers forced to—gasp—order their Big Macs in person. The timing couldn't have been worse, coming hot on the heels of rumors that Amazon had replaced 40% of its DevOps team with AI. While that particular claim smells fishier than week-old sushi left in a hot car, it does make you wonder: in our rush to automate everything, are we setting ourselves up for bigger, badder outages down the line?

Speaking of automation anxiety, the AWS outage serves as a stark reminder of just how dependent we've become on a handful of tech giants. When one goes down, it's not just a nuisance—it's a potential catastrophe. As one expert colorfully put it, our current setup is like "agricultural monoculture"—one nasty bug could wipe out the whole crop. So while Amazon's pouring millions into AI research, maybe they should funnel some of that cash into making their systems more resilient. After all, what good is a hyper-intelligent AI if it can't keep the lights on?

Fresh Cut

  • Amazon Bedrock Data Automation now processes AVI, MKV, and WEBM video formats, allowing developers to extract insights from a wider range of video content and analyze images up to 50% faster. Read announcement →
  • CloudWatch Database Insights offers automated performance analysis for RDS SQL Server databases, helping developers quickly identify and resolve bottlenecks without deep database expertise. Read announcement →
  • Amazon Nova allows approved businesses to customize content moderation settings across safety, sensitive content, fairness, and security domains, enabling tailored AI use while maintaining essential safeguards. Read announcement →
  • Amazon Bedrock Guardrails now lets you use your own encryption keys for AI safety checks, giving you more control over data protection while verifying AI responses. Read announcement →
  • EC2 C8g instances, powered by Graviton4 processors, are now available in more regions, offering up to 30% better performance than Graviton3-based instances for compute-intensive workloads like HPC, gaming, and ML inference. 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 like AI/ML inference and data analytics. Read announcement →
  • Anthropic's Claude Haiku 4.5, now on Amazon Bedrock, offers Claude Sonnet 4-level performance for coding and agent tasks at lower cost and higher speed, enabling developers to build more responsive AI applications without breaking the bank. Read announcement →
  • AWS Step Functions integrates Amazon Q's AI-powered troubleshooting, offering developers instant, context-aware guidance to quickly resolve workflow errors and improve productivity. Read announcement →
  • Amazon Bedrock now automatically enables all serverless foundation models, allowing developers to instantly start using AI models without manual activation, streamlining the process of integrating AI into their applications. Read announcement →
  • Amazon Aurora PostgreSQL now syncs data directly to SageMaker lakehouses without ETL, enabling real-time analytics and machine learning on your database tables using familiar tools like SQL and Apache Spark. Read announcement →

The Quarry

Building smarter AI agents: AgentCore long-term memory deep dive

AgentCore Memory, a key component of Amazon Bedrock's AI agents, takes raw conversation data and transforms it into a persistent, usable knowledge base through some clever cognitive trickery. It doesn't just dump everything into a big conversational soup—instead, it extracts key insights, merges related bits across time, and maintains a tidy memory store that an AI can actually use. The secret sauce? A multi-stage pipeline that includes entity extraction, semantic similarity clustering, and a retrieval system that would make your high school librarian jealous. Read blog →

More posts:

Core Sample

Agentic AI in Action: Turning Data into Outcomes

Agentic AI isn't just a buzzword—it's a game-changer for turning data into real business outcomes, as exemplified by Formula 1's laser-focused approach to data prioritization. By breaking down silos and fostering a culture of experimentation, organizations can overcome data paralysis and unlock the full potential of their structured and unstructured information. One particularly intriguing technical tidbit: NASA's global data sharing strategy demonstrates how even the most complex, distributed datasets can be harnessed for collaborative insights when the right infrastructure and mindset are in place. Watch video →

More videos:

Read more