# Don't Boil the Ocean. Give AI a Box. **Author:** WEKA **Published:** September 15, 2026 ![A clear acrylic cube holds an intricate metallic mechanism and black coils, spotlighted on a dark surface.](https://cdn.sanity.io/images/ult5g8gw/production/4a8b9af6c1e045682293330bc7555e926440e746-1584x672.jpg) ## TL;DR Rethink what AI is actually revealing about your data in this episode of Deep Geeks. - Expose the tech debt AI surfaces the moment it touches a "perfect" platform - Understand the gap between a helpful chatbot and a truly autonomous agent - Focus AI on the most valuable business problem instead of boiling the ocean "Not every business problem needs an AI solution. Just because you can doesn't mean you should, because it's not cheap either." That's not an AI skeptic talking. That's a data leader who has spent two decades building the foundations AI runs on. In the latest episode of Deep Geeks, host Dr. Serena Huang sits down with [Veronika Durgin, VP of Data at Exemplar Luxury Group](https://www.linkedin.com/in/vdurgin). Veronika isn't theorizing about AI in retail. She's living it, from clickstream pipelines to the semantic layers that keep AI honest. If you own a data estate and a mandate to "go all in on AI," this conversation is worth your next 30 to 40 minutes. ### The magic is just your clickstream Why do shopping recommendations suddenly feel so accurate? Veronika strips away the mystique: every click, pause, and scroll is tracked, and machine learning has been mining those patterns for years. What changed is scale. More compute, more storage, more data collected. Not new magic. Bigger foundations. She also reframes the privacy debate as economics. You're not selling your data; you're bartering it. > "When people talk about data monetization, they think you're selling it for money. You're not. You're exchanging your data for something else. If you want something in return, don't give it up for free." ### Your chatbot is not an agent The industry calls everything with GenAI inside an "agent." Veronika draws a harder line: an agent is an autonomous system you hand a goal (e.g. book the cheapest flight to Chicago in two weeks) and it finishes the job without you. A chatbot that answers questions is "search on steroids." Helpful. But not an agent. That distinction matters because it's a maturity test. Everyone has chatbots. Very few organizations run truly autonomous agents, and fewer still are comfortable letting one make a big purchase unsupervised. When someone claims they have "a million agents," ask which definition they're using. ### AI found the skeletons AI is a merciless auditor of your data estate. Veronika has worked with platforms she considered genuinely well-architected: thoughtful design, strong engineering. Then AI landed on top. > "You put AI on top of it, and you quickly realize how imperfect your data platform is. Inconsistent naming conventions. Beta, final, final-final tables that should have been removed. All the skeletons humans know to ignore. AI doesn't know that." Humans carry tacit knowledge: which table is stale, which field changed meaning three years ago. AI has no such instincts. As Serena puts it, AI is the kid in the candy store. It sees everything and uses everything. The fix isn't documenting the entire enterprise. It's scope. ### Scope the box, then expand it Veronika's operating principle predates AI: don't boil the ocean. Instead of connecting AI to everything, define the box. Build a semantic layer modeled around the specific questions you need answered, then expand it incrementally. And flip the weakness into a tool. Unleash AI as your QA assistant: reconcile these two datasets, surface the naming inconsistencies, flag the fields that make no sense. AI takes the first pass through every rabbit hole; humans decide which ones are worth fixing. For new AI leaders under pressure to show results, her advice is refreshingly boring: prioritize the most valuable problem, exactly as you always have. ### The one thing she won't outsource Ask Veronika what stays human, and the answer is immediate: thinking. Understanding the problem, designing the solution, troubleshooting. She still draws her architectures by hand before any model writes a line of code. She'll happily accept documentation written by robots. Logic stays hers. Her vision for retail's future follows the same instinct. Once you hear her describe it (think bridal stores, stylists, and samples) you'll understand why she believes the best retail experience gets more human as AI gets better, not less. ### Watch the full episode This one is for data leaders, AI strategists, and anyone inheriting a data estate with an AI mandate attached. You'll leave with a sharper definition of "agent," a saner scoping strategy, and permission to say no to AI theater. **Watch the full conversation:** [One Size Fits None. What Retail Reveals About AI. ](/podcast/one-size-fits-none-what-retail-reveals-about-ai)or listen on [Spotify](https://open.spotify.com/episode/1PiYqCllp1RgvntLGf2pKQ?si=PD9GJNDJRfaEAxh2dkRH8Q) and [Apple Podcasts](https://podcasts.apple.com/us/podcast/one-size-fits-none-what-retail-reveals-about-ai/id1896864991?i=1000789746127). Make sure to subscribe to Deep Geeks to keep up with emerging trends in AI infrastructure. If this changes how you think about AI and data, share it with someone who needs to hear it.