🦸 Serverless Superhero

Our serverless superhero this week is Lydia Delyova, senior cloud architect at Deutsche Telekom IT Solutions and AWS Community Builder. Lydia does extremely cool work with AI, regularly sharing her expertise and functional examples through community talks and her YouTube channel. She makes great, relatable designs that help make the advanced concepts feel simple and approachable. Thank you for everything you do, Lydia!

💯 Spotlight

You probably noticed the internet wasn’t quite working right on Thursday afternoon (Texas time). There were issues with Spotify, Discord, Cloudflare, npm, GCP, and a ton more online providers. Funnily enough, I was streaming at the time and it started causing faults with the AI agent I was building and left my empty stream hanging for two hours 🫠 Turns out, the root cause was GCP - specifically an invalid automated quota update to their API management system. It’s absolutely crazy to think about the blast radius Google has, where something as seemingly small as a quota update can cripple the world for a couple of hours 🤯

🔥 My Favorite Content

I love it when people can make me think about something in a completely different way. It’s the basis for innovation. Harshil Agrawal recently gave a lightning talk where he claimed you could make serverless apps stateful - and no, not just by persisting state in a database 😅 By using Cloudflare Durable Objects, you can take your serverless app and actually give it stateful characteristics. I’ve been using Cloudflare a lot recently and have been loving the innovation coming from them. While Durable Objects aren’t new, this talk is a great refresher on what they can do.

On my personal “production-ready checklist” for any API I build is a simple task: creating a custom domain. This allows for consistency, discoverability (both internal and external), and honestly, it’s just nicer to look at. This goes for private APIs, too. Last week, I read an article from Ricardo Cino talking about his adventure in putting a private AppSync API behind a custom domain. This apparently was not easy to do, but he got it figured out and shared exactly how to do it yourself. Thanks, Ricardo!

Generative AI makes an appearance in this newsletter pretty regularly, not because it’s the current hype - but because in many instances it’s completely serverless. When Andres Moreno and I do our biweekly livestreams on AI, our two primary goals are to educate people with real software and to do it all serverlessly. I was really excited last week to cover the A2A protocol in a serverless environment. We discuss how A2A is different than MCP, go over the major components of it, look at some code, and demo a working server agent that connects to our garden MCP server. Overall, I was very happy with this stream - it’s a wonderful primer into the world of A2A.

If you don’t know Mark Sailes, it’s time to get to know him. Last week he launched his YouTube channel, Java in the Cloud where he talks about using AWS with Java. He published two videos last week, one on processing S3 files with Java and another on secure passwords for Lambda applications. These are great videos coming from an expert in the field. Keep up the great work Mark and congrats on launching the channel!

Looking for a cool example of serverless in production? Look no further than Jon Goodall’s talk from last week at the AWS User Group UK meetup about how he built a serverless podcast workflow with AI components. Let me clarify - this is not a vibe coded workflow. It’s an agentic workflow that uses AI to significantly reduce the time and effort of publishing an episode of his podcast. This is definitely worth the watch, it has great tips and is delivered wonderfully.

💡 Tip of the Week

A bit of humor for you from last week from Tobias Schmidt. I’m glad I’m not the only one who does this!

🐣 New Releases

Reminder - AWS News is the best source for AWS-related service announcements. For all releases and summaries of what happened, head over there!

Amazon Verified Permissions reduced their price by 97%, going from $150/1M requests to $5/1M requests. Why was it so expensive before? I don’t know. But at least now it’s a considerable option to use.

DynamoDB Streams added support for Kinesis Client Library 3.0. Best I can tell is that this is a Java client that has gotten a big boost.

Lambda Powertools introduced a Bedrock Agents Function utility that makes it easier to create Lambda functions that serve as agent tools. Pretty slick!

Last Words

Were you affected by the outage last week? Does it make you consider a multi-cloud failover or an in-general multi-cloud strategy? You never know when disaster will strike, but is it good to be prepared for a one-in-a-million? You tell me.

If you’d like to make a recommendation for the serverless superhero or for an article you found especially useful, send me a message on Twitter, LinkedIn, or email.

Happy coding!

Allen