Closing Panel: Data Platforms Were Built for Humans - What Do Agents Actually Need?

Unlock Instant Access

Complete the form to start watching.

Session Overview

Explore how data platforms must evolve to support non-human workloads as AI agents replace traditional dashboard consumers.

In this panel discussion, moderator Kundan Vyas (StreamNative) leads a conversation with Smruthi Mukund (JPMorgan Chase), Connor Clark (Harvey AI), Abhinav Wagle (Together AI), and Bonnie Xu (OpenAI) on the structural demands AI agents place on enterprise data architectures. Discover the challenges of managing 24/7 ad-hoc query surges, designing capability-based permission models, assembling layered context engines, and building strict audit trails to govern automated agents.

Key Takeaways

Agent vs. human query patterns: Handling sporadic, 24/7 bursts of thousands of ad-hoc queries without degrading warehouse performance
Freshness vs. governance: Why data governance and auditability can't be treated as an afterthought or a "tax" on streaming pipelines
Capability-level permissions: Moving past static IAM roles to pass-through authentication and task-scoped sandboxes
Context engineering: Breaking business knowledge into distinct enterprise, customer, and real-time layers to prevent LLM hallucinations

Chapters

00:00 — Introduction & meet the panelists
05:15 — What agents need from enterprise data platforms
08:08 — Balancing freshness, reliability, and governance
12:02 — Platform gaps & missing infrastructure capabilities
13:52 — How agent query loads differ from human traffic
16:15 — Rethinking permission models & auth for AI agents
21:30 — Context engineering & skill mapping
25:15 — The future of agentic workflows & closing Q&A

About Speaker

Abhinav Wagle Abhinav Wagle is a Staff Software Engineer with over 15 years of experience building high-performance data and AI-driven systems at scale. Currently based in the San Francisco Bay Area, he brings deep expertise in real-time streaming, experimentation platforms, and distributed systems to some of the technology industry's most data-intensive environments. Abhinav currently serves as a Staff Software Engineer at Together AI, where he led a company-wide infrastructure transformation from AWS Kinesis to Kafka/Confluent Streams, reducing real-time decision-making latency from hours to minutes. Prior to that, he served as the sole technical authority at Stealth Startup, designing a cloud-native supply chain data platform built on AWS, Spark, and Delta Lake. His most formative engineering work came during nearly five years at Roku, where he was a founding member of the Experimentation Platform, an initiative he built from the ground up to support over 1,000 concurrent experiments processing 100+ petabytes of data. Earlier, at Yahoo, he architected a reinforcement learning platform using multi-armed bandits that increased click-through rates by 3–4% across properties reaching hundreds of millions of users work that formed the basis of his independently authored arXiv research paper on batched multi-armed bandits. Abhinav holds a Master of Science in Computer Science from the University of Southern California, where he conducted research in NoSQL performance benchmarking. He is a published author in both IEEE and arXiv, and is passionate about bridging rigorous academic thinking with practical, production-grade engineering.

Bonnie Xu Bonnie Xu is a software engineer and tech lead of the Data Productivity team at OpenAI. Before joining OpenAI, she spent four years building data infrastructure at Stripe and held engineering roles at Meta and Google. Her work focuses on building reliable, scalable systems that bring AI and data together to make analysis faster and more accessible.

Connor Clark Connor Clark is a Senior Data Scientist at Harvey, the legal AI company. He joined as the third analytics hire and has since built much of the analytical foundation for the business side of the company, along with the product analytics foundations for Harvey's embedded experience team. Most recently he drove the creation of Harvey's internal Data Agent: a background LLM agent that anyone at the company can tag in Slack to get analytics questions answered — SQL written, run against Snowflake, and returned with sources in minutes. Before Harvey, Connor worked in product analytics at DraftKings and as an analytics engineer at Verkada, where he rebuilt the data foundations across dbt and Looker.

Smruthi Mukund Dr. Smruthi Mukund is an AI and Machine Learning leader, educator, and technologist with more than two decades of experience driving innovation at the intersection of advanced research and real-world impact. She is part of the AI/ML leadership team within J.P. Morgan Chase’s Consumer Banking division. She also serves on the faculty at Northeastern University’s Silicon Valley campus and has held senior AI/ML and research leadership roles at eBay, Amazon A9, Amazon Alexa, and Twitter. Her expertise spans Machine Learning, Natural Language Processing, Information Retrieval, and large-scale AI, with a track record of applying these technologies to transformative challenges in search, personalization, ranking, fraud detection, and customer experiences at global scale. Dr. Mukund is particularly passionate about responsible and trustworthy AI, including bias detection and the responsible adoption of generative AI in high-impact industries such as financial services.

Kundan Vyas Director, Product & Partnerships at StreamNative, owning the end-to-end cloud product portfolio across Serverless, Dedicated, and BYOC offerings for Kafka, Pulsar, Flink, and Agentic AI. Leads strategy and execution for lakehouse-native integrations with partners across Iceberg and Delta ecosystems, delivering AI-ready, real-time data platforms. Also owns global partnerships across cloud service providers, ISVs, and system integrators—driving co-build, co-sell, and go-to-market initiatives that accelerate customer adoption, expansion, and new logo growth.

Closing Panel: Data Platforms Were Built for Humans - What Do Agents Actually Need?

Session Overview

Key Takeaways

Chapters

Related Resources

StreamNative Tiered Storage v2: Higher Throughput for Pulsar Cold Data

New Destinations in the Stream Materialization Framework

StreamNative Brings External Data Lineage into Databricks

Make Your Data Ready and Safe for Agentic AI