Register or log in to access this video
Building AI products sounds exciting, until reality hits. I’ll share honest lessons from a year of scaling AI at a FinTech startup: what broke, what burned, and what we would absolutely do differently next time.
When I set out to build an AI engineering org at GoodLeap, a high-growth FinTech startup, I thought it would be straightforward: hire smart engineers, pick good models, set clear goals. I couldn’t have been more wrong. Turns out, AI isn’t just another layer you slap onto a tech stack. It changes everything: how you hire, how you build, and how you operationalize and support production systems. If you underestimate those shifts (we did), you’ll pay for it later (we did).
This talk is a field guide to the real, messy work of building AI systems in a fast-growing company. I’ll share:
- Hiring AI talent: How hiring brilliant ML specialists without strong engineering instincts led to major gaps, and what an ideal hiring framework should have looked like. You’ll leave with practical frameworks you can use to hire AI engineers who can build and own systems, not just models.
- Operationalizing Proofs of Concept (POCs): How launching a promising POC without real production planning caused model drift, hallucinations, and a lot of late-night firefighting. You’ll learn how to operationalize POCs the right way, and when to say “not yet” to production.
- Integrating AI as infrastructure: Why treating AI like a “feature” instead of an infrastructure change set us back, and what engineering leaders should do instead. This includes how to prep your engineering org for AI ops challenges before they hit.
- Driving AI adoption: How our best AI tech demos failed to win buy-in until we embedded AI inside real product teams (and stopped trying to “sell” AI separately). You’ll discover how to create AI adoption by solving real user and business problems first.
If you’re leading teams through AI adoption or thinking about it, this is the talk I wish someone had given me a year ago.
Key takeaways
- Mistakes are inevitable, but they can become the engine for faster, more sustainable AI scaling if you learn deliberately.
- AI initiatives need different hiring strategies: Focus on systems builders over pure researchers.
- Moving POCs into production demands discipline, not just technical optimism.
- Operational resilience matters more than flashy model performance.
- Winning business buy-in requires embedding AI inside real workflows, not running it as a parallel innovation track.
