SQL Debugging on Steroids With Claude Code

Description:
Debugging SQL usually means guessing, running partial queries, and hoping your data makes sense.
This time, I let Claude Code lead the investigation.

In this video, I show a real-world debugging session from Atherio, where we recently finished a feature called DEE a data-heavy capability that relies on:

Large-scale data ingestion from Microsoft 365

Signal analysis powered by OpenAI

Historical backfills and rolling score calculations (last 30 days)

After onboarding a new customer, we needed high confidence that:

Historical imports were correct

DEE scores were calculated accurately

No subtle data inconsistencies were hiding beneath “working” results

Instead of manually writing dozens of SQL queries, I prompted Claude Code to help validate the data.

What happens next is where things get interesting:

Claude generates targeted SQL scripts to inspect the imported data

It spots questionable states in the dataset and proactively suggests deeper checks

It iteratively narrows down the investigation, just like a senior engineer would

In the end, it confirms the calculations are correct

Finally, it generates a single, reusable SQL validation script that:

Verifies the entire import

Recomputes expectations

Produces an initial validation report in one run

This isn’t a toy example or a contrived demo.
It’s how AI is starting to change how we reason about data, not just write code.

If you work with:

SQL-heavy systems

Data pipelines

AI-powered products

Or complex onboarding & backfills

…this workflow will likely change how you debug forever.