In this wide-ranging conversation, Manasi Vartak, Chief AI Architect at Cloudera, discusses the growing pains of generative AI adoption, the company’s strategy to “move AI to the data,” and the complex ethics behind AI deployment. Using her memorable metaphor that “GenAI is like a two-year-old kitten,” Vartak emphasizes both the promise and the immaturity of today’s large-language-model landscape — and offers a roadmap for how enterprises can responsibly, and profitably, move from experimentation to production.
Core Takeaways
-
- Generative AI is still in its infancy. Enterprises should treat it as a young technology that needs time and experimentation to mature, though the potential is enormous.
-
- Cloudera’s differentiator is “AI to the data.” Instead of moving vast, regulated datasets to the cloud, Cloudera’s architecture runs AI where the data already resides — whether on-prem or across multiple clouds.
-
- Smart experimentation drives value. Companies should identify high-ROI use cases and assess data readiness before scaling deployments.
-
- Ethics and responsibility must lead AI growth. Vartak stresses that enterprises need clear policies on data use, bias, and privacy — especially in sensitive areas such as mental health, finance, and child-focused applications.
Key Quotes
“Generative AI is a Two-Year-Old Kitten”
“The analogy came about because we were talking about how many POCs actually make their way to production. The number going around is maybe 95 percent aren’t [getting there]. That’s not necessarily a bad thing — it shows how early we are. Generative AI only came to prominence at the end of 2022.”
“For enterprises, that means giving this technology time to grow. You have to experiment, learn where it works and where it fails, and then build from those lessons. We shouldn’t view the low production rate as failure.”
“Move AI to the Data — Not the Data to AI”
“Cloudera’s strength has always been bringing compute to where data lives — at the edge, in the cloud, or on-prem. Our next step is applying that principle to AI. If your data can’t leave Asia Pacific due to regulations, we’ll run the model there. If your customers are in the U.S. on AWS or in the Middle East on GCP, we’ll run AI there too.”
“It’s expensive and often illegal to move petabytes of data across regions, and the latency ruins user experience. So instead of forcing data to travel, we bring the AI to the data — that’s the future of hybrid AI.”
“Controlling AI Costs Starts with Smart Use Cases”
“When CIOs tell me they’re overspending on AI, the first step is to find the right business use cases. Our professional-services teams run workshops to identify which use cases have high ROI and whether the necessary data exists and is ready. Only then does it make sense to deploy.”
“And for large-scale users — anyone running over a hundred models — it can actually be cheaper to run those workloads on-prem rather than in the cloud. The right balance between cloud and on-prem depends on workload complexity, data sensitivity, and cost control.”
“Ethics and Responsible AI Are Business Imperatives”
“AI ethics is multifaceted. In consumer use cases like mental-health chatbots, the risks can be life-threatening. In the enterprise, the stakes are different but just as serious — for instance, bias in loan decisions or inappropriate data use.”
“Responsible AI means asking: should I even be using this data for this purpose? Am I introducing bias? Am I protecting privacy? Enterprises that bake responsibility into every step — from data governance to model oversight — will be the ones that thrive as AI matures.”
