Building a Database for the Agentic Age

Session title: Beyond Text-to-SQL: What “Agentic-First” Really Means for a Database

Description

Software is increasingly being written by AI coding agents, not just people. These agents read documentation, set up infrastructure, build data pipelines, and troubleshoot systems on their own. That changes what makes a good data platform. Agents work best with software that is open source, so they can inspect the implementation instead of treating it as a black box; that runs anywhere, from a laptop to any cloud; that starts quickly instead of requiring heavyweight control planes; and that is easy to modify, extend, and operate.
Many of today’s data platforms were designed for a world where centralized teams provision clusters, manage complex infrastructure, and commit to a single vendor ecosystem. That model made sense when humans were the bottleneck. In an agent-driven world, complexity becomes friction. Agents prefer systems that are portable, transparent, and composable. In this talk, I’ll explore how AI coding agents are reshaping the data stack and why the next generation of data infrastructure will look less like a managed appliance and more like software: open, programmable, and able to run anywhere.

The important point is what “agentic-first” actually means. For a business audience, agentic usually means text-to-SQL - ask a question in plain English, get an answer. That’s useful, but it’s a small part of the picture. From an engineer’s point of view, an agentic-first database is really about extensibility: how easily an agent (or a person) can set things up and build around the database. For example, how easy it is to set up the database and add connectors around it to pull in data from other systems; how easy it is to add jobs that load and transform data; and how easy it is to migrate a query from another database. I’ll make this concrete using Firebolt, the open-source analytics database we built. It runs as a single binary on your laptop, like DuckDB, and also scales to many nodes in the cloud, like Snowflake. It reads open formats such as Iceberg, DuckLake, and Parquet, understands Postgres SQL, and is intentionally simple to deploy, with very few moving parts. Because it’s open and lightweight, an agent can install it, experiment safely in a sandbox, and move fast. I’ll finish with a live demo: streaming data from Postgres or a Kafka topic into Firebolt. If time allows, I’ll build the whole pipeline end to end using just an AI agent - from a live PG/Kafka into Firebolt, and out to a live dashboard - demonstrating how easy it is to work with a database built this way.

1–2 takeaways

1. Attendees will understand how AI coding agents are changing the requirements for modern data infrastructure, from deployment and operations to extensibility and performance.
2. Attendees will gain insight into how a modular, open architecture enables a single database to support diverse workloads without requiring a new specialized system for every use case.

Who is this session beneficial for?

Data engineers and infrastructure engineers who design or upgrade data systems; backend and application developers who build apps that need fast or real-time data; anyone using AI coding agents in their daily work who wants to understand what those agents need from their tools; and architects comparing open-source databases with managed cloud ones. The session is friendly to a wide range of experience, from intermediate practitioners to senior architects, and assumes no deep database background.

Speaker Bio

Anusha Rao is a Data Engineer at Firebolt, an open-source analytics database. She works at the intersection of customer engineering, databases, and AI, helping customers build and optimize modern data applications. Previously, she worked at e6data, where she contributed to the development of a lakehouse query engine. Her interests include distributed systems, developer tools, and agent-driven software engineering.

Reference Links

1. Website - https://www.firebolt.io/
2. More detailed architecture design - https://bauplanlabs.github.io/SAO-workshop/papers/23.pdf