What if your data analysis workflow lived in the same place as your code? That's the question developers and data analysts are increasingly answering with "yes" - using AI coding tools to analyze data directly inside the codebase rather than bouncing between Jupyter notebooks, BI dashboards, and SQL clients.
There's a useful progression worth mapping here. The baseline approach is using a general AI tool like ChatGPT to write pandas (a popular Python library for working with data tables) or SQL, then copying that output into your environment. Step up from there and you're using an AI coding assistant embedded in your editor - Cursor or Claude Code - that can see your actual data files, schemas, and existing code. At the top end, you're running queries, transformations, and exploratory analysis through the AI directly, with results versioned alongside the code that produced them.
The Notebook Problem
Jupyter notebooks are the dominant tool for data exploration, and they have a well-documented fragility problem. Notebooks encourage running cells out of order, make reproducibility hard, and create a sharp separation between "exploration" code and "production" code. The result: a lot of analysis lives in notebooks that never make it into the codebase in usable form, or gets rewritten from scratch when someone actually needs to use it.
AI coding tools change this. When your assistant can see your actual database schema, read your CSV files, and understand the existing codebase context, it can write analysis code that works the first time and integrates cleanly. The analysis doesn't have to be translated later - it's already in the right form.
What This Looks Like in Practice
In a typical session with Cursor or Claude Code, you'd point the tool at a data file or database connection, describe what you're looking for, and get back executable Python or SQL. The AI understands your data structure without you having to re-explain it. You can ask follow-up questions, request different aggregations, or ask it to explain an anomaly - all without leaving your editor.
This isn't perfect. AI coding tools still make wrong assumptions about data types, generate queries that work on a sample but break on edge cases, and sometimes confidently produce results that are technically valid but analytically wrong. The analyst still needs to verify outputs, not just run them.
But the workflow shift is real. For teams where developers are also doing data work - common at startups and small companies - keeping analysis inside the codebase means less context switching, better version control, and analysis that doesn't get siloed from the product.
The question of which approach is "best" in AI-assisted data analysis isn't really about which tool is fanciest. It's about how close the AI is to your actual data and code context. The closer it is, the less you spend re-explaining your setup - and the more useful the output.