Share your best unusual AI use cases! Here's mine

Ooh, now this is a good use for AI

For a little I tinkered with the idea of using Airtable’s barcode scanner to log ISBNs and scrape WorldCat, but lightly tinkering with the AI prompts didn’t like to actually access WorldCat.

Also, love using the Universe to share your build!

Just downloaded it to take a look… one suggestion to try out is instead of Regex formula fields (very traditional Airtable!) to dissect the AI output, try the Generate Structured Data automation. I’ve gotten a test base to, in a single automation, take a movie title, scrap IMDB to create a paragraph of information, and then use the structured data automation to key/value the output and update the record fields. The only hallucination/errors I’ve gotten come at the text generation level not the structured data assignment level. 

Love this idea! I started down a similar path to log my book inventory and create a ‘read roadmap’ for myself for the new year. I plan to include my audiobook/e-book library in the base but haven’t looked into how to load the info quickly and easily. 

​@nroshak , was there a trick you used to get your AI fields to analyze your images? Any time I’ve tried to set up an AI field to pull information from image, it tells me the file isn’t in a format it can read and the AI won’t run

Hi Erica! I’ve had similar issues when using AI in an automation, but it worked well in field agents for me. That said, the model makes a big difference. Here is a field agent prompt that works for me with GPT-5.2:  

Use OCR to extract all of the text from the following picture of a book cover. Use the extracted text to create a JSON object that includes the language, title, author, and other text, following this format: "{

 "Full extracted text": "War and Peace The famous Russian novel by Fyodor Dostoyevsky"

 "Title": "War and Peace",

 "Author":" Fyodor Dostoyevsky",

 "Other text" :"The famous Russian novel by",

 "Language": "English"

}".

Here is the picture of a book cover: {BookCoverScan} Here is the JSON object:

Incredible idea, ​@nroshak 😍