Don’t Let ChatGPT or Claude Boss You Around!

I leaned heavily on ChatGPT (“Chat”—we’re on a first-name basis now) for automation and scripting guidance. We started in Zapier. Juggling actions, triggers, and scripts led to nothing but QuickBooks errors.

So we switched to Make. After several rounds of modifications, Chat insisted I click an area to “Add another Line Item.” I couldn’t find it anywhere. I even uploaded a screenshot to show my options, and it doubled down—swore it was there. I searched every module, nook, and cranny. Nada.

With my patience gone, I typed:

“Is it possible that you are wrong? Please review the documentation and present citations on how to Add Item and where to do it.”

After a pause worthy of a soap opera, Chat admitted defeat. There was no way to do what it had been telling me to do.

Moral: Even the mighty ChatGPT—brilliant, omnipresent, and maddening—makes mistakes. Challenge it when your gut says it’s wrong.


Enter Claude 4.5: The Other Genius Who Gave Up

When Claude 4.5 Sonnet dropped, the internet was buzzing about how great it was for coding. Still simmering in Make, I gave Claude my automation dream.

At first, the advice mirrored Chat’s. Then, after hours of error messages and rewrites, Claude sighed (metaphorically) and told me to call QuickBooks support or hire a developer. Translation: give up.

I didn’t. I searched the exact error message, found hints online, fed them back to Claude—and behold, a breakthrough. Twice, it tried to throw in the towel. Twice, I refused.

Moral: When an LLM tells you it’s done thinking, that’s your cue to start. Human persistence beats machine fatigue.


The Simple Fix They Both Missed

After the chaos of Zapier and Make, the real solution turned out to be blissfully straightforward: a Google Apps Script that authenticates through the Intuit Developer API and posts directly into QuickBooks.

Clean, stable, no drama. I’m now testing in a QuickBooks Online sandbox and will soon pull live data to run the full process in production.


Takeaway

AI is powerful—but it’s not omniscient. The best results come when you stay skeptical, resourceful, and slightly stubborn. Automation isn’t about obeying the bot; it’s about collaborating with it, warts and all.