Tech Industry Session: How AI is Redefining the Way Apps are Built
Overview
| Experience | In Person |
|---|---|
| Track | Artificial Intelligence & Agents |
| Industry | Enterprise Technology |
| Technologies | Unity Catalog, Databricks Apps, Databricks Agents |
| Skill Level | Intermediate |
| DOWNLOAD SESSION SLIDES | |
The traditional blueprint for building applications is more than 20 years old and optimized for a fragmented, ETL-heavy architectures of the early cloud era. The rise of agentic AI has broken this model, forcing us to reimagine how we — or perhaps agents — will build applications for the next 20 years. Today's AI native applications are increasingly written by agents, require easy access to real-time context and AI generated insights from other agents that legacy cloud platforms cannot provide without an expensive "Architectural Tax."
Join Akhil Gupta and engineering leaders from the leading tech companies as they discuss the shift from legacy stacks to an interoperable foundation. We’ll explore the "Lakebase" paradigm, where Lakebase, Databricks Apps and Agent Bricks provide the modern way to build AI-native applications. This session will move beyond the "vibe check," offering a candid look at how to eliminate silos and ship governed, production-grade AI applications at scale.
Key topics include:
- Identifying and eliminating the "Architectural Tax" imposed by legacy cloud silos.
- Understanding why AI-native applications require an interoperable app foundation: Lakebase.
- How do you provide easy access to AI to these applications with Agent Bricks and Databricks Apps
- Learn how a unified governance layer for both your data and AI helps you move from experimental prototypes to production agents faster.
Session Speakers
Akhil Gupta
/VP of Engineering, Databricks AI
Databricks
Jaya Upadhyay
/Sr Director, Enterprise Data Platform
Zillow
Phoenix U Majumder
/Senior Director, AI Engg. & Platforms
Workday
Full Summary
Building for the agentic era: how Workday and Zillow are rethinking the AI stack
AI-native development is changing where software starts, how it scales, and what the data platform must provide. The conversation centers on two pressures that now define enterprise delivery, move faster while establishing trust. Winning teams are redesigning experimentation, collapsing data silos, and treating governance as a product, much like the approaches explored in Databricks guidance on scaling data and AI transformation through people, process, and platform. Enterprises seeking a broader view of what this shift demands can also examine how companies winning at AI built the data layer first.
FAQ
Vibe coding is when non-engineers use AI tools to assemble working prototypes and MVPs in days. It matters because the origin of apps shifts to the business, so engineering focuses more on productionizing, securing, and scaling what already exists rather than building from scratch.
Keep the top of the funnel open but impose sponsorship and expiry. Experiments require an executive backer, workspaces are reviewed on 90-day cycles, and FinOps compares AI cost to ROI. Many MVPs are intentionally culled before engineering assumes production ownership.
Agents require fresh online state and consistent offline data for evals. Maintaining separate systems forces copying, which introduces staleness, quality risk, and audit overhead. A single source of truth, supported by open formats and shared storage, reduces complexity and accelerates iteration.
Agent behavior depends on probabilistic decisions and interactions across tools and agents. Teams need reproducible snapshots of state to replay decisions and continuous evals as models change. Most organizations are still building the state management and tooling to support dynamic, online evaluation.
Invest early in a single source of truth that spans operational and analytical needs, and design governed short-term and long-term agent memory from the outset. Both are painful to retrofit once applications and agent networks are in production.