Megan O'Keefe

POV: It’s 2026 and your manual infrastructure is a ticking time bomb. 💣 If you’re still running terraform apply from your local terminal, we need to talk. Production-grade IaC isn't just about "making it work"—it’s about making it unbreakable. The 2026 Playbook for Terraform Elites: 🔒 State Isolation is King: Local state is a myth. Use S3/GCS with DynamoDB locking. Separate your Dev/Prod states or prepare for a $100k "oopsie." 📦 Modular or Bust: Stop copy-pasting code. Build versioned modules. Pin everything. If a provider update can break your stack, your stack is already broken. 🛡️ Zero-Trust Secrets: Hardcoding passwords in .tfvars? Straight to jail. 🚩 Fetch them dynamically from HashiCorp Vault or AWS Secrets Manager. 🤖 Policy-as-Code (PaC): Don't just plan. Use Open Policy Agent (OPA) to auto-kill any PR that tries to open Port 22 to the world. Stop clicking buttons. Start scaling logic. 🚀 What’s your #1 Terraform "lesson learned" the hard way? Drop it below. 👇 #DevOps #Terraform #CloudNative #SRE #InfrastructureAsCode

8

I just published some findings on Claude Sonnet 4's new 1M context window on Amazon Bedrock. TLDR - it's awesome. I did a benchmark with the LongCodeBench QA dataset (https://lnkd.in/ew4fF3VR) and results show consistent, strong reasoning results across different context lengths. The 5x expansion from 200K tokens opens up some real practical use cases - no more chunking for entire codebase analysis or agents with complex long-term memories, and processing hundreds of documents while maintaining full context. Details and benchmarking code are here: https://lnkd.in/eaTmVBD8

27

2 Comments

This is genuinely good. Anthropic is cooking. Introducing Cowork: Claude Code for the rest of your work. Cowork lets you complete non-technical tasks much like how developers use Claude Code. In Cowork, you give Claude access to a folder on your computer. Claude can then read, edit, or create files in that folder. Try it to create a spreadsheet from a pile of screenshots, or produce a first draft from scattered notes. Once you've set a task, Claude makes a plan and steadily completes it, looping you in along the way. Claude will ask before taking any significant actions so you can course-correct as needed. Claude can use your existing connectors, which link Claude to external information. You can also pair Cowork with Claude in Chrome for tasks that need browser access. Cowork is available as a research preview for Claude Max subscribers in the macOS app. Click on “Cowork” in the sidebar: claude.com/download

11

3 Comments

There is a disturbance in the Force. Read this from a Principal Engineer at Google: "I'm not joking and this isn't funny. We have been trying to build distributed agent orchestrators at Google since last year. There are various options, not everyone is aligned... I gave Claude Code a description of the problem, it generated what we built last year in an hour." - Jaana Dogan In 2026, everyone can vibe code and deploy apps. Everyone is realizing that building something is no longer the problem. Distribution is. #vibecoding #ai #genai https://lnkd.in/gp-NNDr9

21

This is a great writeup that summarizes the different storage approaches used by Datadog over the years and then goes into detail about the newest (v6) engine. LSM and timeseries go well together. https://lnkd.in/g8yTkxuA

97

1 Comment

Just published a new hands-on project on how Aurora Digital, a fictional luxury home brand, scales on AWS using CloudFormation. In this write-up, I walk through: - Building a foundational EC2 web server with Apache using User Data - Evolving it into a fully automated, multi-AZ Auto Scaling Group (2–5 t2.micro instances) - Using Infrastructure as Code to bake in scalability, reliability, and repeatability from day one If you’re breaking into cloud/DevOps or want a practical example of “cattle not pets” in action, this project is a great reference for how to design and automate a web tier on AWS. Read the full breakdown here: https://lnkd.in/eDh98J6M #AWS #CloudFormation #DevOps #CloudComputing #LevelUpInTech #VetsInTech #ACTNOW

6

AWS today launched its AI Agents SDK/framework. I talked to Neha Goswami, the Director of Developer Agents and Experiences at AWS, about how Amazon's take is different from its competitors. https://lnkd.in/gcX_778y

18

I just crushed AWS Cloud Quest Lab 10 and decoded the secrets of scalable architecture! 🚀 For anyone prepping for the Cloud Practitioner or just wanting hands-on experience, I've released a new walkthrough covering Auto Scaling Groups (ASG) and Elastic Load Balancing (ELB). You'll learn how to: Automatically adjust EC2 capacity to handle traffic spikes. Distribute loads effectively for high availability and resilience. Implement a fundamental design pattern for production workloads. This is a core skill for any cloud professional! ➡️ Watch the full walkthrough here: https://lnkd.in/dnQ7GA9B What's the most challenging AWS service you've tackled so far? Let me know in the comments! #AWS #CloudQuest #AutoScaling #ElasticLoadBalancing #CloudComputing #AWSCertification #DevOps #TechTutorial

6

Ever wondered how running a major port like Oakland compares to managing modern data applications? In my latest blog post, I break down the surprising parallels between the Port of Oakland and tools like Docker, Kubernetes, and Argo Workflows. Whether you're into tech, logistics, or just curious about how containers power both shipping and software, this analogy might change the way you see both. Check it out! https://lnkd.in/gs_t_4FW

8

2 Comments

Queue-based architectures are one of the most powerful things in the world—they transform how systems handle scale and flexibility. Just recently, I dove into AWS, set up a dedicated namespace in our message broker for a key workflow, crafted producers and consumers, and built a topic-based exchange. Once it went live, it was pure flow—no bottlenecks, just fire-and-forget. When system design turns real, that’s when the magic happens. #AI #SystemDesign #QueueBasedArchitecture #RabbitMQ #Async #Scalability #Kafka #ProducerConsumer

14

If I had an L5 System Design interview at Google next week, and only 2 or 3 good days to revise, here’s exactly what I would go over. Because trust me you don’t want to design a sharded, geo-replicated, tier-cached monster on the whiteboard and then explode when the interviewer asks a simple question like “So what happens when the user forgets their password?” People love to spit out fancy jargon. But simple almost always scales better than complex. And the first rule of distributed systems is to never build one if you can solve the problem without it. They’re messy, expensive, slow to debug, and the source of most outages. Btw, before we get into it, give me a follow if you like system design insights like this and if you’re studying system design for the first time, I’ve got a platform to help you reach your goals. The best way to learn system design is to learn it by doing, and that’s exactly what layrs.me helps you to do. You get:  – 60+ interactive problems – Option to add your own problem – Real-time, first-principles feedback – A community that shows up for each other This is my short revision plan.. – Networking Essentials (You need these to survive half the follow-up questions) • How the network stack actually works • TCP vs UDP vs QUIC • DNS resolution, TLS handshakes, HTTP lifecycle • REST vs RPC vs gRPC • Idempotency, retries, backoff, timeouts, and rate limiting • ETags, conditional requests, and how caching works • Load balancer basics (L4 vs L7, health checks, stickiness) – Data Storage Foundations (If this is shaky, everything you design collapses) • SQL vs NoSQL and when to choose which • B+ Trees, LSM Trees, WAL, MVCC • Isolation levels and common anomalies • Multi-tenant data layouts in SaaS apps – Distributed Systems (Treat this section like a weapons locker) • Sharding strategies • Replication models • Strong vs eventual consistency • Leader, follower, quorum • Time, clocks, drift, and ordering events in distributed systems If these concepts are clear, any big design, URL shortener, messaging queue, social feed, notifications, payment pipeline. becomes far easier to reason about. If they’re weak, even a tiny follow-up question can blow up your entire architecture. Keep it simple. Pick the right components. Explain tradeoffs. And don’t jump into distributed systems unless you absolutely must.

552

16 Comments

This is a good example of the challenges introduced by PostgreSQL’s MVCC design—and why it’s not ideal for write-heavy workloads. When backfilling a table, be conservative with the write rate and keep it low.

42

Ever since leaving AWS and spending more time in smaller markets like Honolulu, I've been thinking about how AI will reshape the workforce in places that don't look like SF or NYC, and how much time those communities have to get ahead of it. A recent Anthropic report analyzed data showing what is at stake: the most AI-exposed workers earn 47% more on average than those with zero exposure and are 16 percentage points more likely to be women. This isn't just a conversation about preserving jobs. It's about whether local talent retains access to the highest-earning rungs of the economy, or whether those rungs quietly migrate to places that built the pathways first. For a market like Hawaiʻi, the risk is specific: brain drain accelerates as entry-level white-collar roles thin out, high-earning mainland remote workers fill the vacuum and compound the housing crisis, and the wealth gap widens for Native Hawaiian and Pacific Islander communities as the next wave of opportunity bypasses them. I don't think this is inevitable, but the window is real, and it won't stay open indefinitely. Hawaiʻi's economy runs on hospitality, healthcare, and construction, industries with low AI exposure today, which protects existing jobs but may also mean fewer high-productivity knowledge roles over time. The opportunity is to build on that foundation: identify the tasks most likely to be augmented—documentation, analytics, communications, and administrative workflows—and train the next generation to own them. A hotel revenue manager doing the work of an analytics team. A construction PM using AI for documentation and compliance. A nurse practitioner offloading triage to focus on care. The regions that win won't be the ones trying to replicate Silicon Valley. They'll be the ones that help local talent apply AI inside the industries that already define their economy. The window is still open, but building the ladder takes longer than the window stays open. That's why this conversation needs to happen now. Link in comments to the full report by Anthropic.

65

9 Comments

In the first part of our 3-part series on self-service K2K replication, I break down Kafka Connections, a feature in Lenses 6.1 that lets platform engineers define secure, scoped access paths to Kafka. Developers get to deploy K2K without ever touching a secret. No YAML credential sprawl. No ticket loops. No over-privileged defaults. The short version: your observability layer and your application workloads should never share credentials. Kafka Connections fix that with clean isolation, topic-level scoping, and full audit trails. This post covers the "how" and "why" of Kafka Connections, including IAM patterns for MSK, SASL/SSL secret management in Kubernetes, and how Lenses IAM keeps the whole thing locked down by role. Parts 2 and 3 walk through the hands-on setup and running an actual migration. Stay tuned. https://lnkd.in/ebsxnytx #kafka #datastreaming #kubernetes #lensesio #platformengineering #devx

5

1 Comment

Spent the last few days trying to move AWS SES out of the sandbox and honestly—it’s been exhausting. I started with SendGrid (it worked fine), but my plan expired and continuing would cost around $20/month, which feels too expensive for a small site sending only ~100 transactional emails per month. So I moved to AWS SES for the pay-as-you-go pricing. I set up everything properly in eu-north-1: verified domain and From email, DKIM/SPF/DMARC, Custom MAIL FROM, SNS topics for bounces/complaints, an SES configuration set, and I even provided screenshots (suppression list enabled + SNS event proofs). The use case is strictly transactional only—no marketing, no newsletters, no cold outreach. Almost every step I have taken was based on documentation review and extensive independent research. Yet after multiple submissions, AWS Trust & Safety keeps replying: “We’re unable to grant your request. Unfortunately, for security purposes we cannot provide specific details that led to this decision.”- that's it no specific reason or clear next steps. Finally, their statement that there would be no further response brought the matter to a complete halt, leaving the issue unresolved and deeply frustrating. I’m sharing this because I’ve used SES before, but I never realized how difficult the production approval process can be. In comparison, SendGrid felt drastically easier. If anyone has practical tips on what actually helps get approved (beyond the standard checklist), please share. I’d really appreciate it. #AWS #SES #EmailDeliverability #Laravel #CloudComputing

1

1 Comment

I'm adding diff-cover to my GitHub Actions workflow to make sure that my changed Python code always has test coverage. I'm requiring 90% coverage, since often there's a subset of code paths that are never expected to be hit and harder to test. What threshold do you all require? PR: https://lnkd.in/gBGzDrm6 Diff-cover:

8

1 Comment

AWS SAP Studies Continued: Amazon RDS I’m in the databases section of my AWS Solutions Architect Professional prep. What RDS is: A managed database that lives inside your VPC. Put it in private subnets and connect from your app tier or from your private network (VPN/Direct Connect). You can make it public, but that’s rarely smart for production. Architecture essentials --- Many engines (MySQL/Postgres) can host multiple logical databases on one RDS instance. --- Storage: each instance uses its own EBS. Multi-AZ Cluster style shares a synced cluster volume across AZs. --- Backups: stored in AWS-managed S3 (you won’t see the bucket). High availability (two flavors) --- Multi-AZ Instance: primary + hidden standby in another AZ. Standby isn’t readable; failover takes a short pause. --- Multi-AZ Cluster: one writer + two readable standbys. Faster failovers and built-in read scaling (use the reader endpoint). Backups & restores (your safety net) --- Automated backups (PITR): daily snapshot + transaction logs let you restore to an exact second within your retention (up to 35 days). --- Manual snapshots: you trigger them; first is full, later are incremental; they don’t expire and outlive the DB until you delete them. --- Performance impact: Instance mode usually backs up from the standby; Single-AZ/dev may see brief I/O pauses; Cluster minimizes impact. --- Restores: always create a new DB with a new endpoint—plan the cutover and practice it. Read replicas (scale + DR help) --- Asynchronous copies to offload reads and help cross-Region scenarios. Great for reports/dashboards. Expect lag; don’t use them for strict read-after-write paths. DOCS: https://lnkd.in/eyAR8iVv https://lnkd.in/ekwxi5G7 #AWS #Solutions #Architect #Professional #RDS #Databases #Scaling #Performance #DisasterRecovery #RPO #RTO #Backups #Snapshots

13