Good Morning!

A strange pooching of a Google Cloud customer, AWS gets haughty around the idea of AI cautions, and it’s the most wonderful time of year: not RSA Conference week in San Francisco anymore.

From the Community

Kevin Beaumont conducts a fantastic analysis of Microsoft’s recent claimed pivot to putting cybersecurity as their top priority. Personally I think it’s a bit of a sham; they talk a lot more about GenAI than they do security–and unlike their corporate peers (and unlike their own security) they’ve actually got something to show for it.

Someone at UniSuper or Google Cloud absolutely screwed the pooch. A weeklong outage, Google has said NOTHING, the UniSuper blog posts throw Google under the bus for the outage. Honestly, description here of a magic issue, that only affects one customer, is deeply catastrophic, and is met with a "Can we see it?" "No," is remarkably Steamed Hams of them. Now there’s a statement from the Google Cloud CEO on UniSuper’s blog, with no post on a Google property, but has been confirmed by Google to be authentic. What… what is happening down under?

Someone has tricked AWS CloudFormation into powering an API. I wish someone else would trick it to doing its damned job in less time than it takes to walk to the airport from my house.


Ever consider becoming a Cloud Economist / FinOps Engineer here at The Duckbill Group? We’re hiring contractors / part time folks if you think you’d like to apply your SRE / DevOps background to solving interesting billing problems for our clients. Hit the link and see if it perhaps appeals to your interests…


Last Week In AWS: A Blast From the AWS Past

Screaming in the Cloud: AI’s Impact on the Future of Tech with Rachel Stephens

Screaming in the Cloud: Inside Amazon’s Quest For Global Dominance With Dana Mattioli

Choice Cuts

Announcing Amazon Bedrock Studio preview – I wonder if Bedrock Canvas is going to be next. That’s what SageMaker did; those thieves still owe me $260 in credits for their predatory pricing model.

Amazon Cognito introduces tiered pricing for machine-to-machine (M2M) usage – On the one hand, this is charging for a use case that used to be free. On the other, now that that use case costs customers money, I think it’s a lot more sustainable long term for those customers. Put slightly differently, if I use Route 53 as a database, I can expect future service enhancements to potentially disrupt my insane workflow. But if AWS starts charging for database dimensions on Route 53, I can be assured that someone’s paying attention to my needs and I can build atop it with confidence. Let me know if you’re using Cognito’s M2M for things today.

Amazon EC2 Inf2 instances, optimized for generative AI, now in new regions – "For Generative AI." Okay, does that mean that EC2 is considered "a generative AI service?" In their earnings call, Amazon committed to making billions in revenue from GenAI services–which is easy to do if you just stuff everything from EC2 (this post) to S3 (a now deleted AWS job posting) under the GenAI umbrella…

AWS Amplify Gen 2 is now generally available – Six months after being announced at re:Invent, it’s available. When I get back around to my Amplify project I’ll see how it’s improved. Do we like this or no, folks?

AWS Cost Anomaly Detection reduces anomaly detection latency by up to 30% – "By up to 30%." The post says "With this new capability, AWS Cost Anomaly Detection analyzes cost and usage data up to three times a day, instead of daily." That’s cool and all, but it’s tremendously ambiguous. When it runs on its now 8 hour cycle, how long must the anomaly have existed in order for it to get flagged? If it’s still the "several days to consistency" model of the rest of the billing system, then this doesn’t appear to help all that much–and is in fact heavily misleading. If it takes three days for the billing data to show up, then another 24 hours historically for the detector to trigger, and you drop that to 8 hours for the triggering? You’ve dropped the latency by a bit under 10%, not 30%.

Amazon DynamoDB introduces configurable maximum throughput for On-demand tables – It’s not promoted this way, but make no mistake: this is a cost control measure. Nice work.

Creating an organizational multi-Region failover strategy – The first architectural diagram here demonstrates a bunch of resources in a region, including S3. The S3 bucket becomes impaired, so it falls back to an S3 bucket in another region. That’s great, but if S3 goes away for a bit in a region, basically nothing else is going to work properly there either, no?

Reimagine customer experiences with AWS at Customer Contact Week 2024 – Meanwhile as a part of Customer Full Contact Week, Broadcom put out a blog post saying that just because AWS won’t sell VMware Cloud on AWS anymore, the service isn’t going away. But who’s gonna be able to afford what they’ll doubtless do to the renewal pricing?

List unspent transaction outputs by address on Bitcoin with Amazon Managed Blockchain Query – Focus, AWS, focus! You’re targeting grifters with GenAI now, not cryptocurrency.

Generative AI: Getting Proofs-of-Concept to Production – "The media is filled with stories of AI hallucinations, toxic speech, bias, and inaccuracy." So is Amazon Q! So is every experiment I’ve run with Bedrock! This post is about mitigating risk, but you can’t deploy GenAI willy-nilly and then act shocked when it attempts to write a limerick about one of the sex toys you sell.

… and that’s what happened Last Week in AWS.

Newsletter Footer

Sign up for Last Week in AWS

Stay up to date on the latest AWS news, opinions, and tools, all lovingly sprinkled with a bit of snark.

"*" indicates required fields

This field is for validation purposes and should be left unchanged.
Sponsor Icon Footer

Sponsor a Newsletter Issue

Reach over 30,000 discerning engineers, managers, and enthusiasts who actually care about the state of Amazon’s cloud ecosystems.