Curtis Fonger: How to Clean Up Your Data with AI Without Breaking It

Show Notes

"AI is really good at cleaning up data. Humans are not as good at it."

That is Curtis Fonger on the one job most teams still grind through by hand. It is also why he built a safety net before he built the cleanup tool.

Curtis went through Y Combinator in 2021 and built Whalesync, two-way data sync between SaaS apps. He is now building Scratch, a data studio that pulls your Webflow, HubSpot, Notion, and Airtable data down to local files so AI can work on it with a full visual diff before anything publishes. On this episode of Veza Talks he sits down with Collin Belt to talk about what happens when AI can rewrite your source of truth in seconds.

What’s in the episode:

Sync is a data problem, not a workflow problem. Curtis's first job out of college was on the Microsoft OneDrive team, and that lens is what turned a fragile three-day Zapier setup into Whalesync.

Data is messier than teams think. Step one of almost every sync project is cleanup: duplicates, empty records, broken formulas, two hundred images crammed into a single field.

The flying-blind problem. You used to write the script yourself, so you knew exactly what it would do. Now you do not write the script at all, you just hope it does the right thing.

Why undo may matter more than review. Scratch runs locally and keeps your SaaS data in Git on the back end, which gives you selective undo even for changes nobody reviewed. Curtis thinks the recovery angle has more pull than the pre-publish check.

The real fear is not your own mistakes. Eighty percent of drivers think they are above average. If you are the CRM admin, you are on the hook when one rogue agent after one bad prompt wipes the deal pipeline.

Claude as a brilliant but error-prone new hire. Output quality tracks the context you give it, which is why people who have managed teams have an edge. Prompting turns out to be a communication skill.

And the line that frames the hour: AI is the most sophisticated autocomplete we have ever designed. Brilliant at what it is best at, and a poor choice to run your business.

How much of your CRM could one bad prompt rewrite right now, and could you undo it?