Cold take: Lightning-fast CI / CD cycle time isn’t a nice-to-have. Slow builds kill progress. TIL: there's a 5-minute fix to get there. Story time: A few months ago, I spent two weeks tuning our self-hosted runners.

CloudWatch definitely needs more features that include "Insights" in their name. I think there's still not enough confusion.

Shipped an update for our CloudWatch Book video platform: Slicker UI, added transcripts, major learnings & next steps. Finally also aligned for mobile!

I guess most cloud engineers juggle multiple AWS accounts and profiles. This makes it easy to run something against the wrong one. direnv fixes this in a way that feels almost invisible.

Just add the Temporary Elevated Access Management for IAM Identity Center repo to our AWS Fundamentals resources list ✨ Anything that we're definitely missing here? 👀 awsfundamentals.com/resources

Still figuring out if I've ever used permission boundaries the way they are supposed to be used. I've solely used them to restrict how roles can be created by IAM principals. 🧪 Example: Allow a group or user to create roles, but only with a given attached permission boundary.

Reminder: since 15th July 2025, AWS newcomers you can finally start building in the cloud with 𝘇𝗲𝗿𝗼 𝗳𝗶𝗻𝗮𝗻𝗰𝗶𝗮𝗹 𝗿𝗶𝘀𝗸𝘀! 🪄

What is the least known AWS region? TIL that it's most likely ap-southeast-6 now, which is New Zealand, as it just recently opened! 🇳🇿 Apart from that, did you know that there are... 🇯🇵 ap-northeast-3: Osaka, Japan 🇨🇭 eu-central-2: Zurich, Switzerland 🇦🇪 me-central-1: Dubai, UAE

With VSCode (or any fork of it like Cursor or Kiro), there's 𝗿𝗲𝗮𝗹 𝗿𝗲𝗺𝗼𝘁𝗲 𝗱𝗲𝗯𝘂𝗴𝗴𝗶𝗻𝗴 of deployed Lambda functions. 🐛 No more pain for setting up local development environments or trying to replicate cloud conditions. Why is this a big deal?

Since July 25, you're not forced to run an OpenSearch cluster anymore if you want to store vectors on AWS. Why? Because there's 𝗦𝟯 𝗩𝗲𝗰𝘁𝗼𝗿 now! A native vector storage & querying in S3 🪣 Much cheaper, superfast, no infra to manage, therefore no idle costs 💸

Since 15th July, AWS' 𝗙𝗿𝗲𝗲 𝗧𝗶𝗲𝗿 for new accounts has been revised. If you've missed the update, here's a wrap-up about what you need to know:

I don't think introducing the Lambda deploy feature into GitHub Actions was a good idea. One more hard-to-maintain path for working with infrastructure.

Interesting take: thinking that Lambda could be the infrastructure bottleneck. It's not. Instead, you need to think about the downstream services. Databases & internal/external APIs can have a hard time dealing with a lot of connections and pressure if you're not careful enough.

Splitting IaC and application code between roles sounds great - until you try it with Serverless. With serverless, 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝗶𝘀 𝗽𝗮𝗿𝘁 𝗼𝗳 𝘁𝗵𝗲 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗹𝗼𝗴𝗶𝗰. If developers don’t own their own IaC, you’re just adding friction.

I guess there are still a lot of people, even on the professional side, that are working with AWS via ClickOps. Please don't do this - Infrastructure-as-Code tools are there for a reason! 🏗️

AWS AI Practitioner, Machine Learning Associate & Data Engineer Associate exams are currently 50% off 💸 You just need to sign up for the challenge (comes with resources & training paths) and follow it, and you’ll get a discounted voucher at the end.

I’ll admit it - the title sounds like a clickbait ad: “Save $100k+ with this one weird AWS trick! 💸” But there's some substance behind it.

It's too easy to get into alarm fatigue and suddenly you're starting to ask the wrong questions. If you start filtering out alarms, you're already missing the point.

You can use Lambda, DynamoDB, and SQS while still having some ECS/Fargate or even EC2 instances running. Use the right tool for the right job. Some people treat "serverless" like a strict religion. If it’s not Lambda, they feel like they’re breaking the rules.

Starting out with Serverless is often “so simple!”. Fast-forward a few weeks: suddenly it's "too complicated". Devs see Lambda & DynamoDB and think they’re done.

Appreciation time: I'm still a huge fan of Dynobase ⚡️ The best DynamoDB client by far.

Never heard anybody complain about SQS. It's the definition of "just works". Regardless if it's 1 or 1000 rps! ⚡️ Another perfect example of a service that has such great stability that you stop thinking about because it never bites you.

If you're heavily using NX (like we do), please immediately check if you're affected! 🔥 Good news: the compromised versions are already deleted from npm. semgrep.dev/blog/2025/s...

Azure Functions still haven’t caught up to Lambda - at least not where it counts. Cold starts are slow. Deployments? I hit deploy and wait, then wait some more, never fully sure when my code’s actually live. Lambda updates go through instantly. Reliability is on another level.

The AWS Fundamentals community has grown to a scale we never expected. To help us keep creating valuable content and take things even further, we're officially looking for long-term sponsors! ✨

What's your take on this question: What does AWS do better than the other big two? My take: super stable APIs & building blocks for everything! Also, best support by far. The summary from the thread:

Backups? Every team has them. Fully managed or scheduled, definitely automated in some way. What's mostly missing is practice in doing an actual recovery. So, what’s the point of a backup if nobody knows how to bring the system back when it really counts?

Here’s a short summary of Corey Quinn’s overview of what has quietly (and not so quietly) changed in AWS up to 2025 🏗️ From EC2, over Networking, to IAM 👇

ECS is boring. That's a compliment. I've got ECS services running since 2019. Zero drama. No surprise outages. No "it worked yesterday" bugs. It's just there, doing its job. Quietly. Reliably.

First positive contact with the Azure enterprise support: we've got an analysis of a small part of our ecosystem that run on AWS, and we got the honest answer that we just could not replicate it to Azure without a lot of pain in a worse version. The insights:

It's worth to invest some time in thinking about which SCPs would do well with your AWS organization, right in the beginning. This will avoid a lot of pain in the future, as bad things can't multiply. Prominent example: creating IAM users.

I really love the hands-on, real-world examples on the AWS blog. 🏗️ Here's a great recent one: Using CloudWatch Alarms & Lambda to catch traffic spikes 📈 Wrap up: they've set up CloudWatch Anomaly Detection for detecting spikes, then have Lambda turn on Flow Logs.

Maybe you didn't know, but AWS displays some recommended actions inside your billing & cost management overview. Nothing to be too excited about, but I guess it's helpful if you're just starting out. P.S.: There's also a Top Trends widget that shows trends in your spending.

Working with AWS for a long time, but didn't know until today that there's a great free tier overview showing how much of the free tier limits I've already used. It's part of the Billing & Cost Management section. I like it 🤝

One of the best things I did in the past was to put all my helper scripts, prompt templates, aliases, ZSH configuration, and my shell history in a repository. Not only once have I forgotten to back up these when moving to a new machine and had to start from scratch.

Our recent newsletter issue about generating AWS architecture diagrams with MCP & Amazon Q has hit a nerve, I suppose. Open and click rates have exploded. 🔥 Disclaimer: It works to some extent, but it's definitely not production-ready or suitable for perfectionists 😬

Just published an article about the absolute fundamentals of cost optimizations for AWS. Nothing fancy, but also no fluff. Maybe this helps somebody - especially if you're just getting started.

Still dealing with teams that treat API design like a one-off project. Reality check: you’re stuck with those decisions for years. Here are Werner Vogels' "Lessons learned" at AWS:

TIL: you can even have idempotency for actions like starting an ECS task. There's a client token that you can send to get a safe "at most-once" operation. 🔐 This and other best practices for resiliency in ECS can be found in one of the recent AWS blog articles. Link below 🔗

Why does this have to be so awful? I'm exhausted. Please, Microsoft, at least try to offer the same support for Azure as Amazon does for AWS. It's always the same process:

Sums up my feelings about AWS and what I appreciate most. AWS is 𝘁𝗵𝗲 engineering cloud, and it's unbelievably stable in terms of its API. 🏗️ Plus, you receive support from people who have used the product and truly know what they're talking about.

Automated Reasoning Checks are now available at Bedrock. Instead of having guardrails just guess if a response is “probably fine”, you can define logic that proves if an answer follows your rules. At least that's what the announcement states!

This outtake from Yan Cui in one of the past Serverless Advocates newsletter issues hits close to home. We've been there exactly, and it was a nightmare 🔥 • On-premise to AWS 🌩️ • VMs to ECS 🖥️ • LDAP to DynamoDB 📊 • From everything blocking to partly event-driven 🔄

Added floating AWS service icons to the background of my homepage. Huge distraction, my UI/UX friend will physically hurt me when he sees it, but I love it 💛

Would you trust your front door if it were only secured with a password? Probably not. You trust a physical key way more. Please use such a physical key also for your AWS account by using passkeys 🔑 🤝

TIL: there's no need for IAM users & long term credentials if you don't want to use a non AWS-native CI/CD like GitHub Actions! 🏗️ e.g. if you solely want to use SSO & Identity Center. Solution is based on OpenID Connect, an IAM Role & temp credentials! aws.amazon.com/blogs/secur...

𝗧𝗜𝗟: there's a Landing Zone CloudFormation template by AWS 🏗️ Delivers a lot directly out of the box. What you'll get:

Budgeting across AWS accounts used to mean one of two things: either give more people access to the management account (🔥 🔓), or just accept that your budgets will be inaccurate for anything that spans multiple accounts 😅 Guess that's gone with the new billing view 👀 ✨

Looks like AWS just dropped a general MCP server: The Cloud Control API MCP Server 🤖 Offers CRUD operations for basically all AWS services, even 3rd parties. Not a replacement for IaC, but I guess it could be nice for MVPs & Spikes 👀

You don’t truly understand AWS until you’ve built something within a VPC. The documentation makes it sound simple - “just create a VPC” - but in reality, working with VPCs is not straightforward. 💭 With our 3-part introductory series, you'll explore all the key components! ✨