Opinions vs. Reps: How I Actually Started Learning AI

A few weeks ago I asked myself an honest question: am I actually learning AI, embracing it in practice, or just reading about it?

The answer was uncomfortable. I had opinions. I didn’t have reps.

So I picked one tool and committed to going deep. I recently completed Claude Code in Action, Anthropic’s hands-on program for building with AI in real development workflows. Not a passive course. The kind where you break things, get stuck, and figure out why. Which is the only way I actually learn.

What hit me hardest isn’t what I learned. It’s what I realized I’d been avoiding.

The Work I Do

I work in financial data engineering. Reconciliation anomalies, pipeline failures, release validation. The daily grind of keeping numbers honest in a large institution. It’s specialized, high-stakes, and deeply pattern-driven. Exactly the kind of work where AI tooling should have real leverage.

And yet I’d been watching from the sidelines. Reading threads. Nodding along. Telling myself I’d get to it.

The Shift

It started as curiosity, but it’s become a habit.

These days I catch myself doing it constantly. Whatever task is in front of me, some part of my brain is already asking: can this be automated? Is there an agent for this? Could I build one in an afternoon? Not out of laziness. Out of wanting to spend my time on the parts that actually require me.

That instinct found a natural home at work. I started exploring AI-assisted test generation, feeding reconciliation logic in and having it produce edge-case scenarios we’d typically write by hand. The tests aren’t perfect yet. But the goal isn’t perfection on the first pass. The goal is more coverage, more confidence, every time a new change ships. In an environment where a bad release means real financial impact, anything that increases system robustness before go-live is worth pursuing.

What Nobody Talks About Enough

AI doesn’t replace domain expertise in specialized fields. It depends on it. The engineers who’ll get the most out of these tools aren’t the ones who know AI best. They’re the ones who know their domain well enough to close the gap.

That’s what I’m working on. Not necessarily becoming an AI engineer. Becoming a financial data engineer who knows how to use and integrate the tools well.

Follow Along

I’m writing about this publicly as I go. The learning, the experiments, the dead ends. Topics will stay close to what I actually work on: financial systems, data engineering, AI tooling in production contexts, and building in public as a practitioner. If that’s your lane, follow along.