💫 6.13 Event Recap: How Agents Will Revolutionize the Future of Data
Insights straight from unicorn founders & C-suite leaders at Neon, Gen, Datastrato, and Aisera.
On June 13, we brought together some of the most forward-thinking minds in AI and infrastructure for EntreConnect's latest event co-hosted with Data for AI and TechStars Startup Weekend SF: How Agents Will Revolutionize the Future of Data. From keynote deep dives to founder war stories and a powerhouse panel, we explored what happens when agents move beyond chat — and into orchestration, workflows, and production systems.
With a packed house of 700+ founders, engineers, and investors, we dug into the real challenges and opportunities shaping the next generation of intelligent infrastructure.
Featured Speakers:
Nikita Shamgunov, Founder & CEO of Neon ($130M raised, acquired by Databricks), Partner at Khosla Ventures, and former CEO/Founder/CTO of MemSQL (unicorn, $582M raised)
Howie Xu, Chief AI & Innovation Officer at Gen (NASDAQ: GEN); former SVP AI/ML at Palo Alto Networks, VP AI/ML at Zscaler, and Co-founder/CEO of TrustPath; Stanford Guest Lecturer
Junping Du, Founder & CEO of Datastrato; Apache Hadoop Committer & PMC member; ex-Lead SWE at Hortonworks and VMware
Ravi Karra, former Vice President - AI/Engineering at Aisera (Series D+, $180M+ raised), ex-VP Technology at Amex, GE Digital
Suqiang Song, Engineering Director, Uber – ex-Engineering Director Data at Airbnb (AI/ML Infra), ex-VP Data at Mastercard
Jerry (Saisai) Shao, Co-founder & CTO, Datastrato, ex-Tencent
TL;DR 📝
Agent-native infra is being shaped by real patterns: clean APIs, snapshotability, and orchestration-first design
Engineers are no longer just builders—they’re becoming loop supervisors
Metadata isn’t just a catalog—it’s the reasoning layer powering agents
Business logic is the next frontier of automation—not just analytics
🔎 Agent Systems Need More Than LLMs — They Need Real Infra
Nikita Shamgunov kicked off the panel with hard-won insight from Neon's recent surge in adoption. Over 85% of Neon's databases are now spun up by agents—most via co-generation and app-generation platforms like Replit.
His team uncovered key patterns shaping agent-native infrastructure. First, clean, minimal APIs drive faster adoption—because it’s still humans writing and testing agent wrappers. Second, snapshot and rollback support is critical to enabling experimentation at scale. Third, infrastructure must support writes, not just reads—raising new challenges in identity, permissioning, and rollback.
For builders supporting agent workflows, Nikita’s guidance is clear: make infra safe to fail, simple to test, and easy to integrate.
🧰 Engineers Are Shifting from Operators to Overseers
Howie Xu brought a sharp systems-level view of how AI is changing roles inside teams. Engineers aren’t being replaced—they’re becoming loop validators, supervising agents, and tuning workflows.
"It’s not about writing functions anymore," Howie said. "It’s about watching the loop."
In this shift, one-person teams are starting to ship what used to take 200 people. With the right orchestration, founders can now build ambitious systems solo. Howie also joked that if agents keep evolving, "I might need one to wash my dishes next" — a light comment that underscores the radical scope of where automation might head.
But it’s not all fun and jokes. The shift also creates new responsibilities: understanding edge cases, validating agent behavior, and redesigning workflows around autonomy. And in doing so, it redefines how we think about scale.
🔹 The Data Catalog Is the Agent’s Soul
For Junping Du, building infrastructure for agentic systems starts at a different layer: the catalog. In his words: "The data catalog is the agent’s soul."
That’s because agents don’t just consume tables—they reason about meaning, lineage, and structure. Junping argued that traditional BI pipelines weren’t built for this. What agents need is real-time orchestration, semantic context, and machine-readable metadata.
He also described the rise of multi-agent workflows, where different agents specialize in ingestion, transformation, or governance. In that world, the catalog becomes the semantic interface where agents coordinate.
As founders explore data infra in the age of agents, Junping’s takeaway is simple but powerful: if your catalog can’t reason, your agents won’t either.
🚀 Agents Won’t Just Analyze — They’ll Rewire Business Logic
Ravi Karra brought the enterprise lens, sharing what it looks like to actually deploy agents in high-stakes environments.
His core point: agents aren’t just here to surface insights. They’re already beginning to replace brittle business workflows. From GPU usage forecasting to field ops planning, Ravi shared how teams are experimenting with multiple business logic paths—all powered by agents.
"It’s not about query generation anymore. That’s largely solved," he said. "The hard part is upstream — cleaning data, enforcing policy, managing access."
Ravi also broke down the divide between front-office and back-office agents, stressing that each requires different infrastructure and trust systems. Chat interfaces aren’t the same as automated coordination across SaaS tools.
What does this mean for founders? Agent adoption won’t come from fancy chatbots—it will come from cost-saving process automation with clear business value.
🚧 Orchestration Is the Hard Part — Not Just Automation
Jack (Suqiang) Song, drawing from his time leading data platform engineering at Uber, offered a sobering reminder: automation isn’t enough.
While most demos focus on what the agent does, real-world use cases break when agents need to coordinate compute, permissions, and timing across multiple systems. Jack shared examples where agents failed, not due to model performance, but because they couldn’t discover available capacity or resolve access.
This is where orchestration comes in. One agent may handle the job, but others must find the resources, book them, and queue execution—all within system constraints.
The next generation of infra must solve for this orchestration layer—otherwise, your smartest agents will still get stuck.
🔗 Metadata Is the Bridge Between LLMs and Action
Jerry (Saisai) Shao zoomed in on one of the most overlooked ingredients in agent architecture: metadata.
As Co-founder & CTO of Datastrato and a driving force behind Apache Gravitino, he shared how his team treats metadata not just as governance, but as the knowledge graph that powers LLM reasoning.
In Gravitino’s architecture, metadata enables agents to:
Understand what a table actually means
Discover lineage and version history
Enforce governance and trigger actions
Jerry described this layer as a "data brain"—where agents reason, plan, and execute based on metadata context. For any founder building agent workflows: if your system lacks metadata awareness, your agents will operate blind.
📊 Agents Are Reshaping the Nature of Work
Several panelists emphasized that agents are not simply automating jobs—they’re reshaping what work looks like.
Engineers are evolving into loop supervisors. Analysts are becoming prompt composers. And previously siloed roles are merging into orchestration-driven hybrids.
From Howie’s point of view, this is a new kind of leverage. From Junping’s, it requires a rethinking of infra and semantics. And from Ravi’s, it creates trust gaps that must be solved before agents hit production.
This theme reflects a foundational shift in human-machine collaboration: from "operator of tools" to overseer of autonomous systems.
🔎 Context Is King — But Most Systems Still Don’t Understand It
Context came up again and again—not just in theory, but in painful practice.
Junping shared why agents fail without semantic context. Ravi gave examples where HR bots misfired due to missing version control or location-specific policy metadata. And Howie pointed to the real blocker: tribal knowledge and meaning trapped in human minds.
Across these stories, the message was clear: data access isn’t enough. Data understanding is what unlocks scalable, safe agents.
For any startup building tooling, this means investing in lineage, policy, and context—not just interfaces or queries.
💪 Final Takeaway: Stay Close to Reality
When asked for closing advice to founders, the answer was clear: focus on signal, not spin.
Nikita urged teams to read more, talk to users, and observe patterns. Ravi reminded everyone that what works in dev often breaks in prod. Junping called for simplicity in infra design. And Howie told us to stay grounded in the loop—because that’s where all the learning happens.
🌟 A huge thank you to everyone who attended our event! Whether you came for the deep dives or the late-night boba debates, we’re so glad to have had you there.
👉 Follow us on LinkedIn and Luma for future event invites and exclusive takeaways
🗓️ Join us at our next gathering—we’d love to see you there.
Stay tuned for our upcoming LinkedIn posts spotlighting photos 📸 of the incredible audience who joined us at the event!
🗓️ Upcoming Events
Follow us on Linkedin and Luma to stay updated on future events!
July 18, New York, NY | 2025 AI and Sustainability Global Summit. EntreConnect is thrilled to co-host with Columbia University, NYU & Gemi.AI — advancing AI for Good with clear purpose, strong guardrails, and real-world impact.
July 23, Mountain View, CA | Autonomous Driving and Robotics Meetup. Join us for an evening with leading robotics trailblazers — featuring visionary founders and top executives from Anyware Robotics ($17M raised), RoboForce ($15M raised), Whale Dynamic (Series A+ ), Bot Auto ($20M raised), FORT Robotics (Series B+, $42M raised) and more.
July 27, San Fransicso , CA | PMs & Founders | Exploring Manus in Research & Productivity. Join us for an afternoon of working with Manus AI.
July 28, San Jose, CA | Startup Pitch Salon | Investor Feedback + Audience Vote. Join us with top early-stage founders, seasoned investors, and engaged audiences for an event of insight, feedback, and community.
July 31, San Francisco, CA | Infrastructure & Agentic Workflows with C-suite from MongoDB, Databricks, AppsFlyer, Alluxio, @AWS GenAI Loft
💨 In Case You Missed It
June 24 – Super AI Agents on the Rise with Genspark & Dell Capital. We brought together 200+ founders, builders, and investors for a high-energy evening hosted by EntreConnect, diving into the evolving frontier of real-world AI agents.
