Fifteen years making systems reliable.
Now I'm making them reason.

Hi, I'm Srinath.

I didn't set out to become an AI engineer. I spent the first decade and a half of my career deep inside enterprise platforms — Salesforce systems of record, the unglamorous machinery that businesses actually run on. My world was deterministic: a record either saved or it didn't, an integration either fired or it didn't. I learned to love that predictability, and to engineer carefully for the moments it broke.

That was the work for years. Migrations off brittle legacy systems. Event-driven architectures moving millions of transactions a day. The slow, careful craft of making complex platforms reliable at scale — at Mphasis, at QBurst, at Autodesk, and for the last five years at Airbnb, where I lead the engineering for our claims platform. Systems of record teach you a specific discipline: the data is sacred, every change is consequential, and "mostly works" is not a passing grade.

So when generative AI arrived, my first reaction wasn't excitement — it was the reflexive skepticism of a platform engineer. Where are the guarantees? What happens when this thing is confidently wrong on a record someone's business depends on? The demos were dazzling and the reliability was nowhere. It looked like everything I'd spent fifteen years learning to distrust.

But the longer I looked, the more I realized the hard part of AI wasn't the model at all. It was everything around it — permissions, idempotency, audit trails, evaluation, designing for the wrong answer. That's not a new discipline I had to go learn. That's platform engineering, pointed at a new and stranger kind of component. The leap I thought I was facing turned out to be a continuation.

So that's what I do now: architect agentic AI on top of the systems I used to build. I led a migration to refactor a legacy data model so it could actually support reasoning, and built a dual-review framework where a custom AI Judge and human experts grade every output before it counts. I mentor my teams on shipping faster with AI coding agents without lowering the bar. The systems of record are the same ones I always cared about. What's changed is that now they can think — carefully, and on a leash.

I write here about that intersection: making AI reliable on the systems that matter. Not the hype — the guardrails, the human-in-the-loop, and the engineering that turns a clever demo into something you can trust in production. It's the same job I've always had. The material just got a lot more interesting.

Views and opinions here are my own and do not represent my employer.