From Reverse ETL to Answer Engines
Victoria Madrid, founding engineer at AirOps
My next guest is Victoria Madrid, who joined AirOps as one of the first engineers out of Uruguay in 2022, before the ChatGPT boom, before “answer engine optimization” was a category, and when the product was still moving data between spreadsheets. We talked about riding the AI wave from inside a startup, shipping a demo with five days of React experience, and why the team you join matters more than the idea you’re betting on.
Ben: You were at WyeWorks, a software agency doing staff augmentation, a pretty stable situation. Then Berna, AirOps’ CTO, cold LinkedIn messages you. At the time AirOps was a reverse ETL tool moving data between warehouses and Notion. What made that pitch compelling enough to leave?
Victoria: Honestly, I wasn’t looking to make a change at all. It was February 2022. But when I had the conversation with Berna, something clicked. It felt like a once-in-a-lifetime opportunity — I could see myself joining an early team, working side by side with people I’d actually want to grind with. That part was really important to me.
The product at the time was genuinely very different from what AirOps is today. We pivoted significantly. But I said yes anyway, because it wasn’t really about the product. It was about the people and the feeling that this was one of those shots you don’t get twice.
Ben: You joined right as ChatGPT was about to change everything — though nobody knew it yet. AirOps started as a reverse ETL and workflow tool. How did you go from moving data between databases to building an AI workflow engine?
Victoria: The way I always describe it is the ChatGPT boom — I don’t know if that’s a formal term, but that’s what it was. Around late 2022, people started using ChatGPT and you could feel the energy shift. Everyone was talking about it.
Alex, our CEO, is always on the pulse of what’s moving. He came to the team with a new idea: what if we combined what we’d already built with these models? At the time we called them templates — you could pair them with our existing platform, and people were using it for things like SQL generation and data access through AI. It was a pretty disruptive change from what we were doing, but still had that data-focused DNA.
That project became the foundation of what is now our workflow studio — the core of our actions layer. So in a way, the pivot wasn’t a break from what we built. It was an evolution of it.
Ben: As models went from GPT-3 to GPT-4 to multimodal to massive context windows, the infrastructure requirements changed dramatically. You went from handling small data transformations to orchestrating complex, long-running AI workflows. What broke first, and what did you have to rethink?
Victoria: Context windows expanding was the biggest forcing function. Our users — and honestly, ourselves — started discovering use cases we hadn’t anticipated. Bigger outputs, longer chains, more complex workflows. And our early architecture wasn’t built for that.
We had to revisit some foundational decisions and update them. But what I found was that users are always ahead of you. Sometimes you’ll see a memory spike and start digging, and you realize there’s a workflow doing something more sophisticated than you thought was even possible on your platform. They are constantly challenging our own architecture and design. That’s equal parts stressful and motivating.
Ben: There’s a moment every startup has where the product stops feeling like an experiment and starts feeling real. For AirOps, what was that moment?
Victoria: For me it was the grids feature. Before grids, if you wanted to run a workflow at scale, you were uploading CSVs — it worked, but barely. Grids let people run workflows in bulk, creating or refreshing content at scale. It unlocked an entire category of use cases.
When we launched it, you could just see users go. They’d been doing things manually, one workflow at a time, and suddenly they could run hundreds of use cases in a single run. That was a big moment of saying — hey, we’re doing something good here. People really enjoy using us.
Ben: You came from an agency background with structured onboarding, agile processes, paired engineering. The leap to “you’re employee number three, here’s the codebase, demo’s in five days” is massive. What was actually thrown at you in that first week?
Victoria: My first week was a lot. At WyeWorks my main focus was backend — I did some frontend work, but not much. Definitely not enough React experience for what I was about to be asked to do.
There was no onboarding. It was: welcome, let’s go. And my first task was building a filter UI — where users could set conditions like “data is greater than X” or “text contains this substring,” and that gets translated into a SQL query on the backend. So I had to do both the frontend and backend simultaneously, learn React in a real production setting, and design the UX myself because we didn’t have a product team. Just a designer focused on the website.
We had a demo five days after I joined. I had to demo the thing I built. Looking back, that was wild. But it was also the moment I realized this is what early-stage means.
Ben: Four years in, you’re now an engineering manager — which is a meaningful shift from IC work. Was there a conversation where someone said “you’re a manager now,” or did it just happen?
Victoria: It happened pretty naturally, which I think is how it should happen. The team started growing and we needed structure. Berna was handling most of the management early on, but there was a point where that wasn’t enough. More people needed to help with organization, team health, making sure we were going in the right direction.
A product team also came on, and someone needed to bridge that communication effectively. So I stepped into it. The same thing happened with some of my teammates who joined early — they’re now tech leads on frontend and backend. You just fill what’s needed.
What I really appreciated is that Berna always checked in — asked if I was comfortable with a more people-oriented role versus staying purely technical. I felt heard. I wasn’t pushed into something I didn’t want.
Ben: For someone considering joining a very early AI startup today — and the market in 2026 is noisier than ever, every company has “AI” in the pitch deck — how do you cut through and evaluate whether a team is actually worth the risk?
Victoria: The first thing I’d say is: you need the energy. It’s genuinely fulfilling, you’ll grow faster than almost anywhere else, but you need to be ready to put a lot of yourself in.
Beyond that, I’d focus on the team over the idea. With how fast AI is moving, there’s a very real chance the thing you sign up for won’t be the thing you’re building in two years. For us, when I joined, I never would have predicted our product-market fit would be around marketing, content engineering, and SEO. I couldn’t have imagined that. But because the team was right, the pivot worked.
So if you’re talking to founders, ask about the characteristics of the team they’re building. Ask about how they handle being wrong. Ask whether they actually take care of the people working for them. The idea can change. The people are what you’re really committing to.
Ben: Last one — four years in, what’s something that surprised you about where this landed that you couldn’t have called in 2022?
Victoria: Seeing how our product changed other people’s careers. We have AirOps University — cohorts where people learn how to use the platform. And we get messages from people saying things like: “I got promoted because of a workflow I built.” Someone’s job changed because of something we made. I never would have imagined that when I joined.
That’s one of the real advantages of being a founding engineer that doesn’t get talked about enough. You’re close enough to the customer to actually feel that. You see the output of what you build, in real human terms. I really enjoy that part.
Victoria’s story is a reminder that the best bets in early-stage aren’t always legible from the outside. Sometimes it’s a cold LinkedIn message, a product that’s about to become something else entirely, and a team that makes you say — yeah, I’d go through a pivot with these people. Four years later, she’s still there. And the product-market fit they found wasn’t even a category when she joined.