4 Lightning Talks on Practical AI Workflows from Notion, 1Password, MotherDuck & Evidence

How Data Teams Are Using AI to Transform Their Workflows

Four data professionals from MotherDuck, Notion, 1Password, and Evidence shared practical approaches to integrating AI into their daily workflows, demonstrating how artificial intelligence is reshaping the modern data stack.

Using Cursor IDE for Rapid BI Development

Archie from Evidence demonstrated how Cursor, an AI-powered IDE built on VS Code, dramatically accelerates the development of data applications. Unlike traditional chat-based interfaces, Cursor provides comprehensive context about your entire codebase, enabling more accurate code generation.

The key advantages include:

• Automatic awareness of all files and dependencies in your project
• Integration with documentation (like Evidence docs) for enhanced context
• Real-time code generation with diff-style visualization
• Natural language commands for complex tasks

During the demonstration, Cursor successfully generated a complete deep-dive analytics page with multiple components on the first attempt, showcasing its ability to understand both the codebase structure and the specific requirements of BI tools.

Enriching CRM Data with LLMs in Snowflake

Nate from 1Password tackled a common go-to-market challenge: incomplete CRM data. Historically, teams would manually update Salesforce records, with team members updating 20 accounts each morning—a time-consuming and error-prone process.

Using Snowflake's LLM integration, Nate developed an automated approach to classify companies by industry:

Automating Data Catalog Documentation at Notion

Evelyn from Notion addressed the perpetual challenge of maintaining data catalog documentation. Despite significant investments in data catalog tools, many organizations struggle with incomplete metadata, rendering these tools less effective.

Streamlining Data Pipeline Development with MCP

Mehdi from MotherDuck showcased how the Model Context Protocol (MCP) revolutionizes data pipeline development. Traditional data engineering involves slow feedback loops between writing, testing, and debugging code against actual data sources.

MCP enables LLMs to:

• Execute queries directly against data sources
• Validate schemas and data types in real-time
• Generate and test DBT models automatically
• Provision data directly in cloud warehouses

The demonstration showed an LLM independently:

1. Querying S3 files to understand data structure
2. Handling errors (like type mismatches) through iterative testing
3. Creating validated DBT staging models
4. Loading processed data into MotherDuck

This approach significantly reduces the traditional back-and-forth between code generation and testing, though it requires guidance to follow senior engineering best practices rather than brute-force solutions.

Common Challenges and Best Practices

Key Takeaways for Implementation

1. Start with Sandboxed Environments: Test AI workflows in controlled settings before production deployment
2. Provide Rich Context: The quality of AI outputs directly correlates with the metadata and context provided
3. Maintain Human Oversight: AI accelerates workflows but doesn't replace the need for expert validation
4. Document AI Boundaries: Clearly define what AI should and shouldn't do in your workflows
5. Iterate on Prompts: Invest time in crafting effective prompts rather than accepting first results

The consensus among panelists: AI tools are transforming data workflows by eliminating routine tasks and accelerating development cycles. However, success requires thoughtful implementation, continuous validation, and a clear understanding that these tools augment rather than replace human expertise. As data teams adopt these technologies, the focus shifts from manual execution to strategic thinking and quality assurance—ultimately enabling teams to deliver more value in less time.