The Data Huddle by Yuki is a podcast about modern data leaders at global companies, the people who make data work but rarely get the spotlight.
Honest conversations about their careers, wins, mistakes, and how they see the future of AI and data.What separates the companies that actually succeed with AI from everyone else?
In our first episode, Dana sits down with Hannan Kravitz, a data leader with 15+ years in the industry, to talk about career pivots, building delivery-first teams, and what AI changes (and what it absolutely does not) when you’re responsible for production.
On this episode
Most conversations about data and AI live in the land of “strategy” and “transformation”.
This one lives in production. Where dashboards are “fine” until the CEO opens them.
In this episode, Hannan gets real about:
- Shipping over never-ending projects
- AI as an accelerator, not a magic wand
- Why governance and testing are the unsexy heroes of “AI agents everywhere”
- Learning the hard way that “improving” a working system is a risky hobby
The guest
Hannan Kravitz is a long-time data practitioner with 15+ years of experience. He’s currently Head of Data Engineering and Science at Rounds, and he joined just two months before this recording, so yes, he was still figuring out where all the skeletons are stored. He’s held management roles at Playtika (BI Team Lead), Wix (Data Engineering TL), and Artlist (Data Engineering Team Leader, then Data Engineering Group Manager).
Episode highlights
Your data career won’t stay one job
Hannan started in a BI world where one person often did everything: sources, modeling, dashboards, stakeholder requests, and emotional support. Over time that single “BI person” role split into specialties, and he evolved with it: developer → team lead → data engineering focused on pipelines and production readiness.
Takeaway: titles change, the work changes, the people who win adapt.
AI tools, pick them like you’d pick a teammate
Hannan pushes his team to use AI tools because not using them creates a real speed gap. But he keeps it practical: start with the business goal, then test if AI helps you hit it. If it becomes noise, bench it.
Want Hannan’s attention? Don’t pitch. Show the messy reality.
Hannan gets a lot of inbound. What works isn’t big claims, it’s a demo tied to daily pain:
- codebase organization
- code review
- downstream break detection in pipelines
- and the ability to try it hands-on before buying
Takeaway for builders: if your demo doesn’t touch real problems, it’s just a fancy slide deck.
Make vs buy is a maturity and skill-set decision
Hannan’s framework is refreshingly adult:
- If your team isn’t oriented to build tools, buy to fill the gap.
- If you have motivated builders and it fits business needs, building can be part of growth.
- Product maturity matters too, sometimes “buy” means “adopt something stable and stop reinventing it”.
Data leaders become gatekeepers for AI in production
Hannan is candid: AI is impressive, and that creates real concern. His answer isn’t “don’t worry”, it’s: managers become gatekeepers. Let AI suggest, but don’t let it execute unsupervised near production. (Production does not need more courage.)
He predicts new roles will rise (prompting, AI engineering), some roles may shrink, and within two years most companies in the space will use AI as part of production workflows.
“Agents everywhere” only works if your data is aligned
If companies want reliable AI agents, they need a clean foundation and a semantic or business layer. Otherwise the same question gets different answers because nothing is stable.
Biggest failure: improving what already works
Hannan’s most honest section: he’s broken working systems by pushing improvements too far. “If it works, don’t break it” became a rule after paying the cost.
He also talks about hiring mismatches and why setting expectations early saves everyone time and pain.
What he’s proud of: shipping data products people actually use
Hannan is business-oriented: infrastructure is only valuable if someone uses it. He cares about validated products that let stakeholders get answers without constantly pulling engineering in.
Translation: if nobody uses it, it’s not a platform, it’s a hobby.
If you want the real story of data leadership, the decisions, the trade-offs, the mess, and the lessons, watch the episode.
Full video is available now.
