Qrew Meetup: Quickbase Intelligence Package

Key Takeaways

All right, everybody. So, today I'm going to show you a few different ways that you can use the the AI agent to update formulas as also to create brand new formulas. So, I'm just going to pop in and get started here. So, I have just an example project management app for us to to use as a kind of a backstop here. So, the first thing I'm going to do is in this app, I installed the progress bar that you might have seen within the exchange. So, just a very simple, you know, full green progress bar. Nothing fancy. So, what we're going to do is take this formula and start to iterate on it and and, you know, make it a little bit more visually pleasing, some maybe add some features to it. So, what you can do now is another thing if you all haven't played around with the Intel Pack too much is the agent is going to contextually know where you are. So, depending on if you're on, you know, the My Apps page, a formula, a report, what have you, it's going to know where you are. So, you can get a little bit more specific or a little less specific with the way you I like ask a question out of it. So, for this one, I'm just going to go ahead and say update my progress bar formula on the projects table so that it changes from red to orange to yellow to green based on 25% increments of the percent complete fields. And I'm a little OCD, so I like to correct my spelling mistakes, but the agent should be good enough to know that, hey, you were trying to say orange there. So, I'll go ahead and send that. All right, well, so it's starting to do its thing. Something interesting to take a look at is these little thinking blobs. So, this will give you a good idea of kind of how the agent's going about the question. In this scenario, I gave it the like the scale that I wanted. In other instances, I've given it just, you know, go from red to green and then it'll kind of make assumptions for you based off that. And so, if it's ever going to do like some assumption building, you'll see it within these these expandable thinking blobs. All right, cool. So, there it is. This is that new formula that it's going to go ahead and create for me. Um pretty complex here, right? And then it also gives you, you know, the ability to have that human interaction so you can say yes or no if I want to actually make this change. So, I'll go ahead and click yes. All right, so it should be ready. I'll go ahead and refresh this page. So, now that's showing me an actual bar. So, just with one quick, you know, iteration on that, I was able to get it back to the working bar that I wanted it to look like. So, that's all great. Um the next thing I'm going to do is just kind of take this a step further. So, I'm going to take that same formula and actually had a prompt that I was using earlier testing this. And I'm just going to copy it and paste this time. So, this time I'm going to say now update the formula so that it uses a gradient texture, includes additional details about the project, and performs like an enterprise-grade UI feature. So, let's see what that comes up with now. And it's thinking, it's thinking, it's thinking. Okay, there it goes. We got that nice big formula now. All right, looks like it finished. I'll go ahead and refresh this and let's see what that looks like now. A little big. So, I might go back and just iterate on this. And then again, you can always just keep on going back and forth with this until you get the result you're really looking for. All right, let's see what that scaled-down version looks like now. A little bit better. Next thing I want to show is now that we have this big formula that we can obviously will continue to iterate on that. It's a little hard to understand what's going on here. So, what I can then do is come back to this and I can say, "Now, update this formula by adding comments explaining what each elements of this formula is doing and why using the double slash." Pretty big formula, so it's taking for a while. Okay, there it goes. Looks like it should have added the comments. Looks like it did. So, now I'll refresh this formula here. And now this is all got comments in it explaining each element of that formula, what it's doing, why it's doing that. So, I'm sure many of you on the call are builders. Um I know myself in the past, always adding these comments wasn't something that I I typically did, but would be super helpful for anyone else that comes back into the application might need to make an update in the future. So, always good to have that that in there for you. And you can use the AI just to to do it for you instead of uh you know, having to go through line by line. The next thing I'm going to move on to is just a few other types of rich text formulas you can develop. Uh so, on here uh this is just a basic uh form for my project. Um what I want to do is instead of just displaying all of this in this window right here, I want to create like what I'm want to call like a banner. So, it's going to be a rich text version of all the details here, but in a much cleaner HTML kind of rich text output. So, what I'll come here and do now I'm going to go ahead and create a new chat. All right. So, now create a formula rich text field that includes all the details of the project's info section on the project form. Make it use advanced HTML and make it as beautiful as possible. Okay, this one might take a little bit, so let's see how it does. See, so it's finding all the different fields from that that section there. It's just taking some time to to think through it. And obviously, you could get a little bit more detailed with this. I'm being really vague with these right now, where if you knew the exact fields that you wanted, I would list them. Um but for just the purpose of now, I'm keeping it kind of short and simple. Pretty big formula that it came up with. Pretty cool. Imagine how long that would take you to write. What I'm going to do here is just remove this. Let's add that new field. I named it something, the banner. Project info display, looks like it. Save that. Let's see what it did. Pretty nifty, right? So, again, you could probably ask it to reduce the size a little bit if this is a little too large. We can kind of play around with exactly the the color scheme that we're looking for, fonts, what have you. Um but I find this far easier nicer to look at than your typical I don't just quick base form. So, I found those to be pretty cool. Um with the same thing that I did here, um in another application that I've been working on, I had kind of vibe with it a little bit longer, but same general idea. And this is the output that I ended up getting with that one, which I found pretty cool. Actually was able to find these little icons out on the internet for me to display different risk areas, which this app was showing. And then the the numbers here Uh these are actually summary fields, and if I click them, they'll actually pull up the child records. So, I can click right there, it'll pull up the 24 child issues. So, if there's little things like that that you'd want to include, you could definitely advise against that that agent to get to that result. Cool. So, now that we have that, the last type of formula that I want to show is taking something like this, but squeezing it down into a form that would look really good on a Kanban chart. So, now that I have this, this is pretty easy cuz I can just come back and say, "Now, duplicate this field so that it fits perfectly within a card for a Kanban report." Okay, so came up with that. I'll create it. So, it's called Kanban card display. Oh, we got 3 minutes, so I think I got this almost perfectly on time. Perfect. So, right here, we'll just quickly create a new Kanban report. Let's see what that new output looks like. Right. And a nice thing about this is you might know that I think Kanban only has a a limit of Yeah, three fields that you can display. So, creating a rich text field like this takes away that limitation. You can really add as much detail as you'd like. And then there we go. I need to ask it to do a little bit on the you know, the percentage rounding there, um but I think that looks pretty pretty good. >> What I'm going to do today is just going to show you how to use a large language model. I mean, Claude is what I'm using just because it's acclimated to skills, but you can use other large language models to just help you prepare your Quickbase intelligent prompts before you put the prompts in. So, what we're going to do today is I'm going to show you a couple of things. I'm going to show you a skill that I've created. So, we're going to bounce back and forth between screens here. Here's the large language model. Um if you look at customizing and you look at skills, I have a skill called QBI Prompt Engineering. And what this does is it actually helps me to paste my prompts into Quickbase Intelligence um based on requirements gathering that the large language model does um as I kind of start my session. So, just float me an idea, nothing crazy from an application perspective cuz we only have 15 minutes, but like, you know, a closet organizing application or a grocery shopping application. I mean, something simple, three or four tables. Um just float it my way and then I'll um I'll build I'll build it as a demo on this call. I'm not going to do this on the call, but I have it actually interview me. So, I'll say something like um I need to output a BRD, which is a business requirements document. Um so, ask me a few questions and then prepare prompts >> [snorts] >> um you know, prepare prompts uh for Quickbase Intelligence. And then there's that. So, then I'll have it actually ask me, you know, just a handful of questions just like you would for any application. Um so, what types of work? So, contractor vendor? Yeah, we're doing all that. Um budget and cost tracking. Let's see. Drag order of most importance. So, let's say something like this. Um spouse and family, whatever. Keep it simple for this one cuz yes, we only have >> [snorts] >> a handful of minutes to do this. So, great answers. One more quick round. So, it it could go super thorough, right? With me. I'm not going to like so let's say by room. Um Yeah, invoices good. Sure. So, now it's I mean that's a pretty decent application, but what it's doing is it's building my business requirements document. Um so, this is really helpful when you're working on like you know with multiple um stakeholders or whatnot to just have some sort of sounding board for everybody to to look at, right? Or not sounding board to look at, but like some sort of agreeable document that we can start poking holes in, right? So, obviously this is going to take a second to build the BRD and this is way more complex than I did on the last call. So, bear with me. Um it's going to build out the business requirements document that I could you know share with people. They could you know we could put it in a Word document or something like that. Share it, you know, mark it mark it up so on and so forth. And then it's also going to make all the QBIs the the quick base intelligence prompt sets for me. And that's that's where the benefit of that skill comes in and I'll share that skill with you all um you know after after the call or give it to Aster to give to the group. The BRD thing was probably not a good decision in a 15-minute span, but I want to at least give you all context, right? Like 80% of an app build in my opinion is planning and we we don't put that time in in the beginning to think about the overall thought of an application, right? It just did that for me. Um so, now now that it did that, so let's see create app blah blah blah blah blah Uh let's do run the prompts in order. All right. So, want me to build a formula field, so not not for now. Um we will revisit we will revisit the formula field. Um go ahead and give me QBI. Go ahead and and give me QBI. It's a Quick Base Intelligence, right? That's that's what I I showed you all from um So, see, it already has all these prompts in batches. And oh my gosh, there's there's 18. There's 18 of them. It's pretty amazing. So, here's where the the cool magic happens here. So, we're going to call this uh Paul's Paul's app. And then I'll put delete. Sorry, Paul, afterward. Um and I would put a description if I was going to keep this app, I promise. You all know that. All right. So, now I'm over here in the little Quick Base AI. Click this. I've already I've already um kind of I want a new chat cuz I just did this. Yes, I want a new chat. All right. So, you saw how I clicked the copy. I paste it in. Bam. And I'm kind of doing this whole motion over and over again. And then I'm going to say yes to all here because it's, you know, needs my approval. So, just made all my tables and fields. Rooms table. Uh prompt three. So, I'm on prompt four. And yeah, it's crazy, right? Like I'm a glorified copy paster at this point, but it it really helps to um have a really healthy format here. Um just there's going to be less errors, right? And it's it's only giving me the payload in the amount that can be handled by the machine, right? So, it's not it's not going to like fail on us or do something goofy. I'm already on the sixth prompt. And you'll you'll see at the end of a finished result with relationships and tables and fields already all built out based off the business requirements document. So, here's that. And then when you see how fast these are generating, it's pretty sweet. Um part B And I'm just hitting alt tab and bouncing back and forth between these two. And there's that. I'm on 12. And you see it's I mean, it's keeping up with me. 13, three to go. And it's And the cool thing that's doing is it's actually returning all of that information to me in the in the in the uh in the pro in the chat itself. And at the end, I copy all of that information. And I I put it back in Claude to make a a code page on the dashboard. I'm I'm making relationships right now, by the way. There's my first relationship. There's my second. And there is Oh, I had 17. So, you see how this 17 18 Oh, man. 18 total prompts. Crazy. And then there's my 18th. All right. So, now all of this is made. I've got a physical app, but I want to fill it. Can you Can you fill all tables in this app with 20 records of dummy data? All right. There's that. So, now I'm actually going to take this whole conversation here. I'm going to copy this. I'm going to paste it back in. So, here is the output. And make me make me a mobile ready dashboard first for a QuickBase code page and use SimpAuth for viewing rights. And I know this just from so use XML API just cuz I ran into this issue. Um I know that, but you guys might not know that. And this is just extra icing on the cake, honestly. So, I'm going to have that thing do its uh little code page. So, I want it to continue. Yes, I'll create 20 dummy requests. So, it's going to make if we refresh their tables here, they're all it's going to 1 2 3 4 and 5, right? So, it's created 20 dummy records for five different tables. Um I built all my tables all you know, the my relationships based off the the um the business requirements document. And it's going to make it a correction that it found inspection date was a date field. And then, this is this is the piece while I was over here. Let's see. It's building blah blah blah blah blah. It's building this code page that I'm going to paste in here and then to show you what it could look like from a dashboard perspective. Now, obviously, caveat practice security all that good stuff. Like I usually have the large language model act like a security expert and check my code pages to make sure that they're secure. Um but there's a lot more detail that I would give on security of code pages in the future. There's my little dashboard. I'm going to go ahead and um, copy this. I'm going to paste it in here. I'm going to click save. I'm going to go back to my app home and just show you guys uh, what the the dashboard looks like. So, let's add widget. This is really just to get your all's creative juices flowing, right? Like I'm I'm not I'm not, you know, advising like from a security perspective. You guys got to do your own diligence. Use your judgment and use the large language models to actually proof test what you're building. But, here's um, so could not get temp token. So, then how I troubleshoot this? >> [snorts] >> I just click F12. I right click. Um, copy console and then I paste it back in into Claude and we just go back and forth and have a little debate. And then it fixes stuff. There's so much I think we're going to be able to do with uh, the YAML that comes out of the solution API. Um, like just building apps by just changing snippets of code within it's it's called QBL, right? CML is all it is. I I it's it's going to be mind-blowing, I think, what people are going to be able to do. I don't I don't even think we understand quite yet how crazy it could be. >> Yeah. >> Critical operational information is often buried within unstructured documents and emails, which necessitates manual review and data entry before it can be utilized in operational workflows and reporting. Take project invoices for instance. We need to ensure these invoices are paid promptly and we need to track expenses against projects to evaluate our current spending. When this information is logged manually, crucial deadlines might be missed and we lack timely visibility into project financials. AI actions can clear these bottlenecks by instantly and automatically processing our invoices or other key documents as they come in. Let's take a look at an example of invoice documents we might receive. You'll notice that while the formatting of these documents varies, the relevant details like dates and amounts remain consistent. AI actions enable us to extract the necessary information from these documents without needing any training on specific expected formats. Let's move on to the workflow that will automatically process our invoices. You'll see that the workflow automatically starts when a new email with an invoice is received. This means that the process can happen instantly as invoices arrive, eliminating any manual delays. Next, we'll log the details in QuickBooks and ultimately pass the invoice information to our custom AI action. Here, we've defined a structured output detailing the exact properties we want to extract, whether from the email message or the invoice itself. We're also introducing categorization and summarization, demonstrating how the AI action can interpret or generate related information about the invoice beyond just extracting details directly from the documents. Our instructions for the AI specify what inputs it will receive, the output it should produce using these inputs, emphasize that accuracy is paramount for this process, and describe how to apply categorization and summarization. This serves as the additional guidance for our workflow that will allow the AI to perform its task effectively and consistently. The email information and attached invoice are set up as inputs for the AI. Once processed, we'll upload the details into QuickBase. Let's see this process in action. We'll send a couple of example emails containing our invoices for processing. Once those emails are received, our AI action automatically starts, uploads the file to QuickBase, and then passes it to our AI action for processing. Now, it's extracting the relevant details from the invoice, and upon completion, we'll log that information on our invoice record back in QuickBase, allowing us to instantly evaluate and report on the invoice. Let's go back to our project invoices and refresh. We can now see our two new invoice records have been logged, and the relevant dates and amounts have been extracted. We also see a categorization applied based on what was included in the invoice, and a summary so we can know what products or services were procured without actually opening the invoice. We've transformed a manual workflow that required opening emails, downloading documents, and logging details in QuickBase into a fully automated touchless process. This not only improves efficiency and frees up time from focusing on administrative tasks, but ensures critical operational data is captured instantly and no longer remains trapped inside unstructured emails or documents. And the same approach can be applied anywhere important information arrives through files or messages. Where in your organization do these bottlenecks still exist? What could change in your operations if this information was captured automatically and made immediately available to your systems and teams? AI actions is here to help automate these processes, unlock instant visibility, and give valuable time back to you and your team.