AI in Data | Notion

How to effectively use AI in data: it's a requirement at companies now, so do not fall behind.

Coding assistants and extensions

Tool	What it's for
Cursor	A full IDE, forked from VS Code, built around AI-first, multi-file editing
GitHub Copilot	The most integrated option, works inside VS Code and JetBrains, strong for day-to-day completions
Claude Code	A terminal-first, agentic tool built for larger reasoning and architecture-level work
Sourcegraph Cody	Indexes your whole repository, strongest for understanding a large existing codebase

<aside> 💡

Most engineers by 2026 use more than one tool: something fast for completions, and something more deliberate for architecture-level reasoning. Neither replaces the other.

</aside>

Principles for using AI without losing your judgement

Use it to go faster on the mechanical part, never to skip deciding whether the output is right
Understand code before you accept it. If you can't explain what a suggestion does, don't merge it
The judgement part (is this the right query to ask, is this dataset trustworthy) is exactly what stays valuable as the mechanical part gets automated
Treat an AI's confidence as unrelated to its correctness. It states a wrong answer exactly as fluently as a right one

AI for PR review and code QA

Tool	What it's good at
CodeRabbit	Free for public repos, catches the most issues but with more noise
GitHub Copilot code review	Fewer false positives, more conservative
Qodo Merge	Adds automatic test generation for changed logic, not just review comments

Pick based on what you actually need: coverage (CodeRabbit), precision (Copilot), or test generation itself (Qodo).

AI talking to your database

Text-to-SQL tools let you ask a question in plain English and get a query back. Accuracy depends far more on the context you give it, table descriptions, defined metrics, join paths, than on which model you use.

Snowflake Cortex Analyst and BigQuery's Gemini integration are the two most mature warehouse-native options, but both lock you into that specific warehouse.

This is also where Data & Tech Glossary matters most for data work: it's becoming the standard way an AI agent actually connects to a real database or tool at runtime, rather than working from stale, hardcoded context.