🦸 Community Superhero
Our community superhero this week is Diptanu Gon Choudhury, founder and CEO of Tensorlake. Before starting Tensorlake, he worked on Facebook’s ML platform and ran massive container workloads at Netflix before a stint at HashiCorp, so he comes from genuine production experience at scale. Tensorlake just launched a serverless runtime for agentic workloads, complete with durable execution, code sandboxing for untrusted LLM-generated code, and built-in observability. Thank you for everything you’re building, Diptanu!
💯 Spotlight
In EDA, it’s considered good practice to send failed messages to a DLQ. But that’s just the beginning. I remember a time in my career where putting messages in a DLQ was the final stop. We never monitored them, and if we did, we certainly didn’t know what to do with them. If only I had Cyril Bandolo’s article from last week on automating DLQ triage with Bedrock and Step Functions, things would have been different. Cyril shows us how to automatically handle messages in a DLQ and classify them as redrivable or if they need a ticket written up for work. This is an awesome practical build for both EDA and AI.
🔥 My Favorite Content
When we got Lambda Managed Instances at re:Invent last year, there was a mixed set of reactions about them. They’re ideal for steady-state workloads vs bursty or intermittent use. If this sounds like something you need, Lucas Vera published a great article last week discussing the use case in detail. He provides some great advice on gotchas and assumptions when getting started with them that will likely save you a bunch of time. This is a solid article with practical insights on developing with Lambda. Great job, Lucas!
We’ve all tried to track ip addresses of API callers at one point or another (at least I have). David Behroozi shared an extremely useful blog last week telling us how we’re accidentally tracking them wrong (my words, not his). His blog talks about the problem he was trying to solve with Lambda function urls, and includes a detailed table of where to pull the ip address from under certain scenarios. I had no idea there was this much to it, and I absolutely love it.
Abdulai Yorli Iddrisu wrote up a great article on a portable AI memory layer with MCP that he made. While the core functionality mimics what we see in mem0 and zep, it’s a nice option if you want total control over everything. It also comes with a neat Chrome extension to automatically inject context when you’re using Claude, ChatGPT, or Gemini. It’s a genuinely good idea and useful build.
S3 turned 20 last week, wow! What I’d consider the first serverless service, S3 has come such a long way - especially in recent years. Sébastien Stormacq wrote a blog post last week talking about the sheer scale and engineering behind it that is just mind boggling. There’s nothing new in this one, but it’s cool to look at the numbers behind the rockstar service.
💡 Tip of the Week
There continues to be round after round of layoffs across our industry this year, and it feels flattening. I like what Peter Hanssens is doing and encouraging those affected by the layoffs to continue showing up. Continue to be present, be visible, and keep learning. And of course, if there’s anything I can do to help you if you’ve been affected, please let me know 💙
🐣 New Releases
Reminder, all releases from AWS can be found on AWS News by Luc van Donkersgoed. Below are my favorite from last week.
S3 now supports account regional namespaces for general purpose buckets. No more hunting for globally unique names, just lock a namespace to your account and region, and you’re off to the races. Yes, this is probably what you were already doing, but it’s built-in now!
Bedrock AgentCore Memory now streams to Kinesis every time a memory record is created or modified. This now lets you trigger downstream workflows whenever memory is updated without having to poll. Nice!
Last Words
Details matter more than we think. IP address in the wrong header, DLQ messages that are retryable yet unchecked, a shift to a steady-state Lambda model vs on-demand. These are big impact problems that take your systems from good to great, when done right. Our content creators this week did a great job framing that and giving us the answers to build better.
That’s my take on the week, but what’s yours?
What did I miss? What made you nod along (or 🙄)? Hit reply if you’re reading the email. Prefer socials? Ping me on Twitter, LinkedIn, or email.
Happy coding!
Allen