· 8 min read
Agents have made it easier than ever to write code, explore data, and build software, but getting from a question to something useful still involves a surprising amount of work.
We've spent the past year making MotherDuck easier for agents to use. Now we're making it just as easy for people to work with those agents directly in MotherDuck. Say you ask an agent to analyze some data. It writes SQL, produces a chart, and explains what it found. While that’s nice, if the analysis is valuable, you probably want to run it again tomorrow, share it with your team, or turn it into something others can use, whether it's your business stakeholders or your customers. Suddenly, you're back to managing pipelines, moving data between tools, and figuring out where the outputs should live so that everyone who needs access can find them and use them to support their team’s daily work.
At MotherDuck, we believe agents should do more than answer questions. Agents can do analysis accurately and effectively, which begs us to ask an important question: What if analytics infrastructure were as easy to set up and maintain as a consumer app or a personal agent?
Agent Chat is a new way to work with your data using natural language. It brings AI-powered analysis and tools to build and run data workflows directly into the MotherDuck UI, with access to all the tools needed to go from raw data ingest to working pipelines, insights grounded in your data and business context, and interactive data applications in one prompt and a couple of clicks.
By telling the native agent what you want to accomplish, it can help you get there by writing SQL and Python on your behalf. A separate AI client configuration or copy-pasting code from one window to another isn’t required anymore.
From asking questions to getting work done #
Most AI chat experiences for analytics focus on answering questions: generating SQL, explaining results, or creating visualizations. While those are useful activities, they're only part of what happens between a business question, a trusted answer, and a resulting decision to take action.
Sometimes, the right data isn't in the warehouse yet. Other times, it still needs to be cleaned or transformed. More frequently, the answer raises more questions than answers due to a lack of sufficient business context, or perhaps the analysis needs to run on a recurring cadence. While it’s easy to default to creating a dashboard, or a simple spreadsheet, to monitor key metrics, these artifacts aren’t able to keep up with the accelerating pace of business and industry changes.
Agent Chat lets users work through these steps in one place so teams can derive value from data and use it to make well-informed decisions and deliver business outcomes instead of static artifacts. There’s no need to pick a model because we handle model optimization for you: Ask it to explore a dataset, write and execute SQL, create a data pipeline using our native Python runtime, Flights, or build an interactive Dive app to go deeper and share the results. If you need to refine the analysis, you can keep going in the same conversation and star your favorites to refer back to later or pick up where you left off.
The work doesn't have to end when the conversation does. Your data lives in MotherDuck, while Flights can be run again or scheduled, and Dives can be revisited, updated, and shared with your team. Instead of starting from scratch with another prompt each time, you can build on what you've already created and focus the rest of your time on high-value work. Because everything is in one place, each handoff builds on what came before instead of starting from scratch.
Taming dashboard sprawl, from raw data to working application #
Let’s dive deeper into an example. Say you want to understand how average temperatures have changed across European countries over the last decade. The data is publicly available, but you haven't ingested it, built a pipeline, or set up a dashboard.
With Agent Chat, starting a request is easy:
How have average temperatures changed across EU countries over the last 10 complete years? Use the NOAA GHCN-Daily dataset on S3 to create and run a Flight, show me the logs, then build a Dive where I can explore the trend by country and year.
That said, answering a question often requires more than running a SQL query. First, the agent explores the data. It can inspect the files and schema, understand how temperature measurements are represented, and determine what needs to happen to prepare the analysis.
Next, it creates and runs a Flight. Flights are Python programs that run on MotherDuck's managed infrastructure. The native agent can write ingestion and transformation logic, execute the Flight, and inspect its logs to understand what happened.
Finally, it builds a Dive app to help you explore high-fidelity insights. Rather than leaving you with a static chart in a chat transcript, the agent can create an interactive application backed by data stored and queried in MotherDuck. This enables you to explore temperature trends across countries and years, ask follow-up questions, and share the app with others so they can do the same.
AI makes it easier than ever to create dashboards, but harder to keep track of what's current, trustworthy, and useful. MotherDuck tames dashboard sprawl by giving your data, pipelines, and Dives a centralized home base. Operators can turn their specialized knowledge into business results, while engineers and data teams can build and maintain workflows using agents.
Analytics, context, and data apps belong together #
As agents take on work that people previously did manually, analytics is shifting from something people do periodically to something software can do continuously. That means more analytical queries, more automated workflows, and growing demand for fast, efficient execution. We believe MotherDuck is well positioned to be the ducking fast, single-node execution layer for those agents.
At the same time, anyone can prompt an agent to write code and build applications. As the bottleneck shifts from writing software to giving it the right data, business context, and infrastructure, it becomes even more important that these elements live together in one place.
Agent Chat builds on three core, AI-native MotherDuck capabilities:
Guides give agents the business context they need to interpret your data, from metric definitions to join relationships and more, so they don't have to rediscover it from scratch each time you ask a question. #
Flights let agents create, execute, and schedule Python workflows to ingest and transform data. #
Dives turn analysis into interactive, shareable visualizations and data applications that teams can revisit and build on.
Together, these building blocks give agents more than a database to run SQL. And while AI makes it easier than ever to generate analyses, dashboards, and applications, that doesn't automatically make them trustworthy or useful. Agents can still make incorrect assumptions, and outputs scattered across chat sessions can be difficult to find, verify, or reuse.
By keeping data, business context, workflows, and applications together, MotherDuck makes it easier to inspect how results were produced, build on previous work, and share outputs that teams can actually use.
You bring the expertise while agents do the heavy lifting #
The people who understand a business problem aren't always the same people who know how to build the infrastructure to solve it. Agent Chat helps close the gap.
For operators, that means turning business questions into analysis and interactive applications without needing to write SQL or Python. If you’re an engineer or a lean data team,you can use Agent Chat to spend less time on routine pipeline development, troubleshooting, and infrastructure setup, and more time solving higher-value problems.
Lastly, for teams already using third-party agents through MotherDuck's MCP server, Agent Chat allows you to use the same underlying platform through a native, conversational experience. While the subject matter expertise remains with you, Agent Chat helps turn it into working software and data applications.
From data to action #
Agents make it possible to turn questions into working software without the usual handoffs between tools, teams, and workflows.
Today, that might mean exploring a public dataset, building a pipeline, and turning the results into an interactive Dive. Tomorrow, it could mean analyzing customer activity, identifying accounts that need attention, and building Dive apps that help teams take the next step.
We're working toward a future where agents don't just answer questions, but also help build workflows that run continuously, surface new insights, and evolve with your business. Furthermore, we also envision teams collaborating directly in Agent Chat, picking up where colleagues left off to build and refine Dives, Flights, and Guides together.
Agent Chat makes it easier to go from asking questions to building something useful without requiring everyone to become an infrastructure expert.
The point isn't to generate more answers: It's to do more with them.
Get started with Agent Chat #
MotherDuck is the action stack for delivering trusted, data-backed outcomes. To get started, just open the Agent Chat in the UI to start a conversation, explore a dataset, create a pipeline or scheduled job with Flights, or build a trusted data app backed by Dives and our warehouse-native context layer, Guides.
With availability across six AWS regions, support for bring-your-own-compute (BYOC), and Azure support on the way, it's easier than ever to move faster from questions to insights.
Table of contents