Good Morning!
I was at the AWS NYC Summit last week, where I learned a lot about Agentic AI and squat about much else, because that’s where all of the attention has gone. Increasingly I’m wondering who AWS events are for, because they’re not really for me anymore…
From the Community
I said that Route53 databases contain multitudes. That also includes the location of the ISS.
This source code analysis of Amazon Kiro is surprisingly fun to read. I’m deeply curious what led to "DO NOT discuss ANY details about how ANY companies implement their products or services on AWS or other cloud services" being included in the system prompt. It’s clear that Kiro has struck a chord; people like it!
The tale of how Medium (motto: "Medium: because we’re neither rare nor well done") pulled off a financial turnaround. I still try to avoid linking to it, just because the giant half page popup whining for you to sign in isn’t particularly respectful of you, the reader.
Last week I linked to a report of an AWS Lambda silent crash, and got more reader feedback than anything else has in at least three years. There were a lot of folks looking into this, but I think this writeup by AJ Stuyvenberg does the best job of encapsulating it. AWS Support could have responded better to the original author of the post, but frankly it seems that the author is more interested in (badly) marketing himself than he is in learning something new about the system he’s attempting to use.
My beloved Route53 database contains multitudes, including in this case a bunch of malware.
Podcasts
Last Week In AWS: Sovereign Cloud, Now With Slightly More Pretend Sovereignty
Choice Cuts
Announcing Model Context Protocol (MCP) Server for AWS Price List – You can now talk to the AWS Price List. Personally, I’ve been screaming at that thing for years.
AWS Price List API now supports four new Query Filters – "What’s wrong with you," "why are you like this," "who hurt you," and "are you kidding me" are the four queries of the AWS Pricing Apocalypse.
Proven Practices for Succeeding with a Multicloud Strategy – I am shocked by this; I was fully expecting it to be a very AWS-serving post, but it’s almost point for point what I would advise customers to do. Even the most AWS-advantaging point (pick one primary provider that gets workloads by default) is one that I’ve made repeatedly to customers. This is very well written, and a far cry better than I expected it to be.
The Amazon S3 console now displays an external access summary for all your buckets – This is surprisingly helpful, and a nicely granular change from the simple fit over open buckets it used to pitch.
Coming soon: Exam update and new name for cloud operations certification – "AI is changing EVERYTHING!" proclaims the company that’s rolling out a new version of its trivia-based closed book test questions.
AWS Free Tier update: New customers can get started and explore AWS with up to $200 in credits – At last. At long, long last.
Amazon S3 Metadata now supports metadata for all your S3 objects – "Phew, I just finished reuploading all 40 petabytes of my S3 data to now be annotated with tags. It took months, but it’s finally done! Now to take a big sip of coffee and read this next Last Week in AWS newsletter item…"
Introducing Amazon S3 Vectors: First cloud storage with native vector support at scale (preview) – I’m not entirely clear on what the pricing here is, nor are any of the customers I’ve spoken with. Yes, it’s 6¢ per GB, but… for what, exactly? Is that the data itself, or just the vector embeddings? If the latter, what’s the relationship between the embeddings and the actual data, size-wise? Nobody seems to have a good sense of it yet.
Amazon S3 Tables reduce compaction costs by up to 90% – This kinda speaks to either "they’re seeing competitive pressures" or "they mispriced this and now know a lot more about their economics and can lower customer costs." Which is it? Choose your own adventure.
Under the hood: Amazon EKS ultra scale clusters – I do appreciate these technical deep dives, but man are these some overwrought clusters. Notable that they’re not using etcd the way you or I would use etcd, since unlike MongoDB it’s apparently not web scale enough.
Introducing Amazon Bedrock AgentCore: Securely deploy and operate AI agents at any scale (preview) – "AgentCore" sounds like a very specific niche of adult entertainment, which is how you know it’s absolutely something AWS itself named.
Introducing AWS AI League – Oh, how delightfully AWS of them to take something that was actually fun to watch: tiny cars careening around tracks while engineers nervously clutched their laptops, aka "every DeepRacer tournament ever" and transform it into what’s essentially a glorified homework assignment with a participation trophy. It’s like a duller version of the math team.
AWS API MCP Server now available – Apparently what the world desperately needed was yet another way to make API calls to AWS services through Claude and other AI assistants, since "the AWS CLI" was insufficient?
Amazon EBS now provides visibility into EBS volume initialization status – Oh man, this would have made the last decade so much easier with regard to explaining to customers why newly created volumes from snapshopts are slow as death initially. Better late than never, but this is pretty late.
AWS Cost Anomaly Detection improves accuracy with model enhancements – How?! How did this improve? It was already quietly EXCELLENT; it surfaced things I couldn’t find when I went looking for them. It’s been an absolutely fabulous service that not enough people know about, but I’m mystified about how it could be more accurate than it has been in my (extensive) experience with it.
Accenture scales video analysis with Amazon Nova and Amazon Bedrock Agents – I saw a demo of this in New York and honestly? It’s kinda spooky how effective this is.
… and that’s what happened Last Week in AWS.
