We’re excited to announce the Preview launch of two new AI capabilities for Gemini in AppSheet Solutions: Extract Rows and Summarize!
About two months ago, we announced that Gemini in AppSheet Solutions was launching to General Availability with 2 capabilities: Extract and Categorize. The two new capabilities, Extract Rows and Summarize, will further enhance your ability to automate processes and build intelligent applications.
Why It Matters: New Ways to Supercharge Your Processes & Build with Confidence
With the new AI tasks, you’ll be able to:
Extract Rows: This capability allows you to extract multiple rows of information that follow the same data structure. Examples of use cases are:
Process uploaded invoices (PDFs) to extract multiple line items, including product descriptions, quantities, and prices, directly into your data source.
Extract key details from complex reports that contain repetitive data structures.
Summarize: This feature enables you to create concise summaries of long, verbose descriptions, as well as from different fields associated with a single record. Examples of use cases are:
Generate quick summaries of lengthy customer feedback forms or survey responses.
Create concise overviews of detailed project descriptions or incident reports by summarizing multiple associated fields.
Just like with its previous AI capabilities, Gemini in AppSheet Solutions helps you:
Automate Smarter
Keep Humans in the Loop
Easy to Use (No Machine Learning Degree Required)
Build with Confidence
Deploy AI Safely & Scalably
How to Get Access & Start Building
You need to have an AppSheet Enterprise Plus plan.
Learn More & Share Your Feedback
This is now in Preview and we’re looking forward to hearing your feedback! Feel free to let us know what you think here in general. Please also feel free to share specific examples of what works and doesn’t work for your use cases. You can do that by rating results when using the inline testing feature for AI Task!
Thank for sharing news and update with the community.
The typical use case for the extract rows should be reading documents such as invoice along with its details, normally in the form of the table.
To extract all the data needed to be captured, such as invoice date, company, total amount etc together with the details (tables/rows), currently we need to chain the two AI tasks. First to employ the Extract task, followed by Extract rows. As far as the performance and results are concerned, it works as expected and needed. However, in terms of the quotas being consumed, can we understands it would consume the quota “twice (or double) “ against one single shot to trigger the single BOT? It could be, but worth checking before it is GA-ed in the near future.
Any additional features are always welcome, but it is depending on how frequently those are used in real world. “Summarize” task is simply works just “summarize something” as its name suggests.
While the AI task was in private preview phase, I emphasized that AI Task should have “Free Prompt” task.
For instance with such a flexible AI task, we input the values out of the columns of row(s) and give a prompt to AI “Please summarize!” . Then we simply save the results to the target column.
With such a flexible AI Task, the app creators should be able to come up with unlimited ideas how to utilize the power of AI (Gemini) onto AppSheet to integrate.
Once again, I personally welcome the addition of “Summarize” task to AppSheet, but it is possible the usage is going to be pretty much limited. Not a day to day tool.
The AI task ( with BOT) to give a free (arbitrary & dynamic) prompt to the AI should be broaden the area of the usage of such a task (not yet materialized) with AppSheet.
I give you texts in English, then please return that in Japanese.
I give you USD amount, please return the JPY based on the latest exchange rate.
or whatsoever.
If such a AI task is available, then “Summarizing” task (you pushed it to preview) should be covered, by giving a free prompt, saying “please summarize the input and return it” as a rough prompt.
AI should be flex, but I m not sure why Google is trying to limit the capability of the unlimited power of it, by introducing the “Fixed” features. Is it because of money (fees)?
Personally speaking with almost 10 years experiences with AppSheet, I wont come up with the much of “usage” and “cases” to use Summarize AI task with AppSheet…..
Then I will keep continuing use Gemini API with AppSheet to feed the power of AI to our own AppSheet app, unless more usefull/flexible AI Task is introduced.
Thank you for the preview release of the AI feature “Summarize”!
After trying it out in various ways, I found that depending on the “Additional instructions,” it can produce quite flexible outputs beyond just simple summaries.
I have one question: When I tried this feature several times, the output from the AI seemed to be around 1000 characters.
By the way, this character count is for my native language, Japanese.
Considering the following conditions, the 1000-character output seems to have some leeway compared to other limitations. Is this a restriction on the AI Task side?
【Other Limitations】
・The Output column is set to LongText type, and the data source uses a spreadsheet.
・The spreadsheet cell limit is “approximately 50,000”
・LongText character limit is 5,000 characters (only in AppSheet DB?)
The pdf file I uploaded might have been in excess of the limit you have placed, but this was not mentioned anywhere else. Is this limitation only given while the feature is under the preview and will be lifted after GA?
The file size only max 2.1 MB acceptable is quite small in light of the real world file management. Hope this limitation will be removed or softened before the GA.
I tried reading multiple QR codes using “Extract Rows.” It correctly created a record for each QR code, which is good.
However, the “QRコード” for every record was just a repetition of the first URL.
I tried adding an instruction in the “Additional instructions” field to scan them individually, but the result was the same.
The ability to batch scan barcodes could be a powerful feature for AppSheet, so it would be great if you could improve this.
Giving the exactly same input to the AI task for the repeated jobs, then the AI task returns the different result sometime. Presumably, this is because of the behavior of ”Artificial General Intelligence” Normally, we do adjust this by changing the “temperature” parameter to control the randomness of the response back from the AI. We really need the same control over this parameter with AppSheet AI task as well.
Just today, we increased the max size for images. We’ll see about increasing the max size of PDF files as well, possibly after we let users leverage as a GA feature.
Thanks for sharing, but please provide that kind of feedback (about results provided by AI task) directly in AppSheet. See the “Learn More & Share Your Feedback” section of the announcement above.
We saw a drop in accuracy in a number of use cases when the number of columns increased. We made the call to limit the number of columns to avoid users trying out configurations that were less likely to succeed. We may raise that limit in the future.
AppSheet creators who want to do more than what is currently supported may prefer to use Apps Script for instance to connect to their own preferred Generative AI model.
It is understandable as developers you need to place a certain limit to ensure AI operation result in success without errors, but still max 12 column is too small when it come to the case where we use the AI task in general.
If Google stay claiming that AppSheet is “No code” platform instead of “coding/low-coding” platform, then this native AI tasks should be enhanced so that the creator can add AI functionality without coding. Technically, yes, it is better to user API along with Apps Script, but that is just for skilled developper who knows how to code.
I sincerely hope google teams will recall what the AppSheet platform is (Service for no-code app development for citizen developer.) Or Google change the terms of “citizen developer” who are eligible to do some coding?
To expect the better result out of calling to AI, it is natural we expect that we could get better outcome by passing the high resolution files such as image/pdf….
However, once we increase the resolutions, then AppSheet bans to upload as it is in excess of the max file size. This does not make sense to us.
