6 min. read
An interview with Jeremy Castan, Lead Product Manager at OLX — on 10 years in data, building a basketball team of AI agents, and the moment he deleted an entire project and started from scratch.
Jeremy, you’re about to hit 10 years at OLX — all of them in data. How has that journey evolved?
October will be my 10th anniversary. And I think data is probably the area that has changed the most in that time.
When I started, my first role was implementing our tracking and clickstream systems — understanding the clicks and page loads generated by users. Then I moved into data governance and GDPR, supporting the compliance layer. Then OLX built an experimentation platform from scratch, which I became part of later on. Then the machine learning era came — helping data scientists serve their prediction models in a reliable and scalable way.
And now we’re here: GenAI and agentic AI, with a completely new world of possibilities. Before, data was only accessible to technical people — analysts, engineers, data scientists. That’s not the case anymore. We’ve built connectors and MCPs that bridge the gap between a non-technical colleague and our data platform. Now we want to make trustworthy, comprehensible data available to everyone. Product managers. Sales. Customer support. Designers. C-levels. With as few bottlenecks as possible. That’s the journey we’re on.
What strikes me looking back is that OLX kept pace with every one of those shifts — and kept giving me new problems worth solving. That’s not nothing, over ten years.
When AI became mainstream, a lot of people got nervous. Were you?
Never, really. Maybe because I understand how it works — at its core, it’s a system that predicts the next word. That’s not a criticism, it’s genuinely powerful, but knowing that helped me see it clearly rather than fear it.
My instinct was always: this is an opportunity. How can I embrace it? How can I use it to improve my work and help colleagues around me who aren’t as familiar with it?
The message I give my team is the same: I don’t have a crystal ball, but I don’t see strong signals of people being replaced by AI. What I do see is that people who don’t embrace it will face more challenges than those who do.
When did it click that you could actually build things yourself?
It happened gradually, then suddenly.
I started small — using Cursor to create dashboards, pull my KPIs automatically, and build simple static pages. Useful, but limited. Then a colleague said: “Why don’t you use plan mode first? Brainstorm with the AI before you touch the code.”
That changed everything. In Cursor’s plan mode, I could discuss the problem I was trying to solve — no coding, just thinking out loud with AI. And then I realised: this isn’t just going to produce a random HTML page. I can build a real product with this.
That’s when I built Nina Ninja — an agent that translates plain-language questions into clickstream data queries, no SQL needed. And then an agentic platform for my team of Data Product Managers, where each of them is free to create specialised agents of their own. The barrier between having an idea and being able to build it has been completely removed.
What also helped was that the infrastructure was already there. OLX had built the Data Platform MCP — the connection layer that gives AI tools direct access to our internal data. I wasn’t starting from scratch. I was building on something the team had already made possible.
The basketball team — explain it. Why basketball specifically?
I’ve always liked building a little universe around the things I create. When I first joined OLX, I was part of the data team we called the Pyrates — pirates, but with PY for Python. Our slogan was “Your data, our treasure.” That’s just how I think about products.
So when I was building this platform of AI agents, I didn’t want to call them “Data Agent 1” and “Data Agent 2.” Where’s the story in that?
Basketball made sense because it captures everything I wanted the platform to represent: it’s a collective sport, but one that requires tactics. You don’t put a player in a position randomly — it’s part of a strategy. There’s coordination, timing, and each player has a specific role that makes the whole team stronger.
That’s exactly how agent orchestration works. One agent alone is useful. An agent that is part of a coordinated workflow is, in my opinion, ten times more powerful. Basketball was a great metaphor. And I also just like basketball.
What completely failed — and what did it teach you?
The first version of the agentic platform was a disaster. And the reason was simple: I was telling Claude Code what I wanted, not what problem I was trying to solve.
I’d say: “I want a page with a sidebar here, a header here, agents that communicate with each other.” And it would build exactly that. But it wasn’t working. It wasn’t user-friendly or valuable. It was missing things I hadn’t thought of. And then I’d try to fix one thing, and it would break something else. Eventually, the whole project went to trash.
So I deleted everything and started again. This time, I treated it like I’d treat a conversation with a developer from our team. As a product manager, I don’t open a conversation by describing a solution. I give context and describe a problem. I explain why it matters, what I’ve already tried, what wasn’t working.
I went into plan mode, described the problem thoroughly — not the interface I wanted, but the underlying challenge — and let the AI brainstorm solutions with me. I challenged its proposals. Asked questions. Pushed back. Only then did I start building.
This second version worked. And now that’s my method: at least one hour of brainstorming and self-reflection before a single line of code. Every time.
And the thing is — OLX gave me the space to delete everything and start again. That’s not a given. At a lot of companies, a failed first attempt becomes a conversation about risk and process. Here it was just: try again. That culture matters more than people realise.
You’ve said the PM superpower is articulating problems — and that’s exactly what AI needs. Has working with AI sharpened how you do that?
Yes, I think so. Product managers are naturally close to their users, naturally good at scoping problems and articulating them clearly. That’s the muscle we’ve been building for years. And it turns out that’s exactly the input AI needs to produce something valuable.
But I want to be clear about one thing: AI hasn’t reduced the time I spend with my users. I don’t think it’s interesting to use AI to outsource conversations with internal stakeholders. Wouldn’t you feel strange if you wanted to discuss a problem with someone and they sent you an AI form instead of talking to you? When you’re a PM for internal products, your colleagues still need a human on the other side. The AI layer is more impactful after — between discovery and delivery.
What AI has changed is what I do with those conversations once I have them. I can scope faster, challenge my own assumptions more rigorously, enrich with data insights, prototype more quickly, and look at problems from angles I wouldn’t have thought of alone.
Ten years is a long time. What keeps you here?
A lot of things. The mission has been consistent — helping people buy and sell, giving objects a second life. The people. The pace of change.
But the thing I keep coming back to is something harder to name. I feel free here. When something isn’t working, I can say so. When I want to try something nobody has asked me to try, I can try it. The basketball team of agents, Nina Ninja — nobody commissioned those. I built them because I had a problem and the space to solve it.
In the data world especially, we’re genuinely empowered to experiment. Everything new in technology shows up at OLX — and there are people here who actually want to use it, not just talk about it.
Ten years is a long time. I’ve stayed because I keep finding new problems worth solving. And I think that says more about OLX than anything else I could tell you.
Jeremy story is part of our “AI Innovators” series. Across OLX, people are finding new ways to use AI — to create, simplify, learn, and move faster. This series celebrates the colleagues who are sharing their discoveries and inspiring others to experiment along the way.
