Farewell, AWS

A short note

Today was my last day at AWS. I’m leaving for a new role outside the company, and I wanted to mark it here instead of letting it slide by in a LinkedIn update.

AWS was the first cloud I ever touched. I’ve shipped real things on Azure and GCP since, but AWS is the one I built my career on. Most of the instincts I rely on every day got formed there.

The BMW cluster

My first run-in with AWS was during a co-op in BMW’s IT Innovation group in Greenville. I was an undergrad, and somehow I’d ended up running the group’s 19-node Hadoop cluster. It was the kind of responsibility you probably shouldn’t hand a college student, but there I was, wrestling with node failures and capacity planning on a Tuesday night.

EMR had just come out. I stood up a cluster, ran a job, tore it down, and felt my brain rewire. I didn’t have the words for it yet, but I remember thinking: this is going to eat the world.

GE and SageMaker

Fast forward to GE Digital in Cincinnati. Most of what I did there was infrastructure analytics and infrastructure ML on a platform monitoring thousands of compute assets across the enterprise. Somewhere in that run, I built an event-driven solution that I’m still proud of. The shape of it, turning messy infrastructure signal into something self-healing, is what hooked me on event-driven architectures for good.

SageMaker showed up during that same stretch. AWS had just announced it at re:Invent in late 2017, and our reps were pushing it hard. Before SageMaker, the gap between “we have a model” and “we have a model in production” was enormous. SageMaker collapsed it. I was sold.

From GE onward, I worked almost exclusively in AWS. Not out of loyalty, just because the tools were the sharpest ones on the shelf. By the time I was a few years into my career, working at AWS had quietly become a background dream. Something I told myself I’d get to eventually.

re:Invent 2022

“Eventually” turned into “now” at re:Invent in December 2022. I went as a customer and left knowing I wasn’t just going to keep building on AWS. I was going to go work there.

The session I remember most was on large-scale training clusters for transformer-based networks. I walked in almost by accident. I’d been deep in ML and analytics work, but I hadn’t been tracking deep learning very closely. Transformers were a name more than a thing to me. The speaker (we’re still connected on LinkedIn) laid out where it was all heading, and I went back to the hotel that night and started reading everything I could find.

The timing, in retrospect, was absurd. ChatGPT launched on November 30, 2022, in the middle of re:Invent week. I was sitting in a session about training transformers at almost the same moment the rest of the world was discovering what those transformers could do. It wasn’t a shock. It was a starting line.

What I did there

I joined as an AI/ML Specialist Solutions Architect in Healthcare and Life Sciences. My days were spent designing enterprise AI/ML architectures across computer vision, agentic AI, and generative AI for some of the most regulated environments on the planet. HIPAA, GxP, GDPR. The kind of work where the constraints are as interesting as the capabilities. I got to help customers make the same jump I’d made a decade earlier at BMW, just at a completely different scale.

Why I’m leaving

Working for AWS was genuinely a dream. Not a “LinkedIn post” dream, a real one. I loved the team I got to work with. I loved the impact we had on customers. The kind of problems I got to put my hands on in two years don’t show up in most careers, and the people I got to put my hands on them with don’t either.

But underneath the AWS dream, there’s been a second ambition I’ve been chasing for a long time. I love to build, and anyone who knows me knows that. What fewer people pick up on is that I love the customer-obsession side of the work just as much. Figuring out what to build and why. How it fits a larger product bet. Who it’s for and what they actually need. The closer I’ve gotten to those kinds of decisions over the years, the more energized I’ve felt. Technical product ownership and product strategy is where I’ve wanted to head for a while now.

The new role pulls me a meaningful step in that direction. Closer to the bleeding edge, and closer to where the product decisions actually get made. It’s also a place where the pace of what’s being built and decided is about as fast as anywhere in the industry right now.

I’m not going to name the company here. Anyone who knows me or sees where I land next will figure it out quickly enough. None of this is a critique of AWS. It’s just where I’m pointed next.

What I’m taking with me

Three things stand out.

Builder culture. Amazon has the most robust builder culture I’ve seen anywhere. The primitives, the internal tooling, the reward structure, all of it is pointed at one thing: making it easy to ship. That’s not an accident. It’s what they’ve been optimizing for since day one, and you feel it daily.

Customer obsession. I assumed it was branding before I joined. It isn’t. It’s a real, enforced, daily practice. Meetings get redirected when someone drifts off the customer. Decisions get pushed back when they can’t be traced to a customer problem. Writing starts with the customer. The gravitational pull of that one principle is something I hadn’t seen anywhere else.

Dive Deep. This is the one that hit hardest for me. In most of the places I’d worked before AWS, diving deep was treated as a failure mode. You were “in the weeds.” You weren’t being “strategic.” At Amazon, it’s a virtue. Senior people are expected to drop into the details at any moment and speak to them credibly. Asking for a deeper layer is a sign you’re doing your job, not a sign you don’t trust your team. After years of being told that diving deep was a bad instinct, finding a culture that named it as a strength felt like coming up for air.

I’m grateful to the people I worked with, the customers who trusted me with their hardest problems, and the leaders who pushed me to think bigger than I thought I could.

From a Hadoop cluster at BMW to the inside of AWS itself. Not a bad arc.

On to the next thing.