To avoid AI hallucinations, Square's AI tools translate merchant queries into deterministic actions. For example, a query about sales on rainy days prompts the AI to write and execute real SQL code against a data warehouse, ensuring grounded, accurate results.

When setting up an AI data agent, don't invent example queries from scratch. Instead, bootstrap the process by analyzing your database logs (e.g., from Snowflake) to find the most popular, real-world queries already being run against your key tables. This ensures the AI learns from actual usage patterns.

AI data agents can misinterpret results from large tables due to context window limits. The solution is twofold: instruct the AI to use query limits (e.g., `LIMIT 1000`), and crucially, remind it in subsequent prompts that the data it is analyzing is only a sample, not the complete dataset.

Users get frustrated when AI doesn't meet expectations. The correct mental model is to treat AI as a junior teammate requiring explicit instructions, defined tools, and context provided incrementally. This approach, which Claude Skills facilitate, prevents overwhelm and leads to better outcomes.

To enable AI tools like Cursor to write accurate SQL queries with minimal prompting, data teams must build a "semantic layer." This file, often a structured JSON, acts as a translation layer defining business logic, tables, and metrics, dramatically improving the AI's zero-shot query generation ability.

The concept of "Skills" was born when the team found that telling Claude *how* to query a data source and follow design guidelines produced better, more flexible dashboards than building rigid, parameterized tools. This discovery highlighted the power of instruction over hard-coding.

To make an AI data analyst reliable, create a 'Master Claude Prompt' (MCP) with 3 example queries demonstrating key tables, joins, and analytical patterns. This provides guardrails so the AI consistently accesses data correctly and avoids starting from scratch with each request, improving reliability for all users.

Open-ended prompts overwhelm new users who don't know what's possible. A better approach is to productize AI into specific features. Use familiar UI like sliders and dropdowns to gather user intent, which then constructs a complex prompt behind the scenes, making powerful AI accessible without requiring prompt engineering skills.

Text-to-SQL has historically been unreliable. However, recent advancements in reasoning models, combined with AI-assisted semantic layer creation, have boosted quality enough for broad deployment to non-technical business users, democratizing data access.