Unaligned #6: Glide, codeless software development with AI

Transcription for the video titled "Unaligned #6: Glide, codeless software development with AI".

1970-01-01T01:01:21.000Z

Note: This transcription is split and grouped by topics and subtopics. You can navigate through the Table of Contents on the left. It's interactive. All paragraphs are timed to the original video. Click on the time (e.g., 01:53) to jump to the specific portion of the video.


Introduction

Intro (00:00)


In-Depth Discussion On Glide

What are you doing? (00:13)

People don't know that I used to work at a programming magazine and I got real interested in new methodologies for bringing more people and more businesses into the modern world without doing a lot of coding and I'm pleased to have Glide's founder today on with us. David start out with who are you? I'm David. I'm the founder of Glide. As you mentioned, Glide's a tool that allows anyone to build beautiful custom software for their business based on spreadsheets. And we have that same interest that you mentioned is like getting more people, especially businesses to be able to create their own software, create their own technology. And we're seeing some interesting stuff with businesses trying to make sense of AI and figure out how it can help them. So happy to be here to talk about it with you. Yeah, I have on x.com, I've found 4,000 companies using AI and building AI tools. I think I have 400 companies that are building developer tools. Yeah. You're a developer tool, but you're not a really hardcore, you're not building C++ code or anything like that. You're really trying to help people who are running businesses, probably smaller businesses who have a lot of processes or data stuck in spreadsheets or other things and get them into building new kinds of apps. I tell me what you're doing. Sure. So Glide is a, it's a living contradiction. It's a developer tool for non-developers. We call people who use Glide developers. But my background is in designing and building tools for traditional software engineers. And when we set out to build Glide, we were really inspired by the success and the universality of spreadsheets as a tool for businesses to organize their data and run their processes. And we wanted to know, what's next? What's the next incarnation of spreadsheets, this democratic tool that anyone can just pick up and build with? So yeah, we are a software development tool, but we're a high level software development tool. And we're building LIDE for people who do not identify as software developers. If you have a casual familiarity with spreadsheets, you can use LIDE to build powerful custom software for your business. And we say spreadsheets a lot, but you don't have to be a spreadsheet wizard. There's no formulas. So everyone has a crazy uncle who's always trying to get them to learn spreadsheets a lot, but you don't have to be a spreadsheet wizard. There's no formulas. So everyone has the crazy uncle who's always trying to get them to learn spreadsheets. We're not that uncle. We're a very, it's just a generally approachable tool, productivity software for anyone in any company, regardless of their role to create custom software. Even back in the 1990s, when I worked for Visual Basic Programmers Journal, which is where I really started understanding corporate software development, right, which is very different than the building a video game or something like that. I would see tools that were aimed at people like this. No code kind of tools, but the systems just weren't good enough back then to The systems just weren't good enough back then to flexibly build applications that would be useful. It just wasn't good enough back then. Now with large language models with these, uh, the technology that's underneath chat GPT or a grok or Google's bard, Now we can see a new way of building. So new way of doing software. Is that something you agree with here? Yeah, totally. So I would call this category of software development tools for everyone. Usually it's referred to as like low code or no code. And these are pretty dirty words. Uh, they're associated with products that are like not very inspiring. And if you've used these tools in a big company, like clunky and limited, they don't tend to make better software than you could make if you hired a talented team of software developers. And this was actually why we started this company. Because we thought that these tools were really promising, but we hadn't seen anyone build a great product in this category yet. And we were inspired by tools like Figma, Google Docs even, these contemporary browser-based collaborative tools that just expand the audience for design or spreadsheets.


Low-code tools (04:20)

And yeah, I think LOMs are, if anything can make software development more accessible, it's AI. And we've done some prototypes to this effect. And every day, like you mentioned on X, we're seeing new people building new ways to build software that are more broadly accessible. Some are more playful. Today, I don't know, did you see, there's this project called TL Draw. Yes. They have a whiteboard and you just doodle and then you say, make it and people are making games and making forms. So it's just a wild amount of experimentation. I think everyone is prepared to rethink what does it mean to create software? And even what is software 10, 15 years ago, it was something, an app in a mobile app store. Now, is it a conversational spin on ChatGPT? Everything's up for grabs now. It was taking a moment to reinvent everything. You're aimed at people who are in businesses who have some sort of business process wrapped up in a spreadsheet, right? Do you only work with spreadsheets or do you ingest other kinds of like databases or other kinds of business data? Tell me a little bit about who's the perfect customer for Glide. That's the question I would ask. Yeah. So we started with spreadsheets because spreadsheets are where non-technical people have their data already. And we got traction when we added Google Sheets to Glide. And I'll give you a demo. You pick a Google Sheet, Glide slurps in the information, and it just constructs a starting point for you. It makes some software. And as Glide has matured and we've gotten access to larger customers with more demanding workloads, their data sets have gone beyond spreadsheet scale. Spreadsheets peter out at about 50,000 rows of data, where databases scale to millions of rows. So we're adding connectors to larger scale data sources. But at the end of the day, we bring all your data into Glide and we present it in the Google Sheet Airtable Excel type table interface. That's very easy to scroll around. It has smooth scrolling. When you can add interesting capabilities on top of your data, whether it's in a spreadsheet or a database in the form of computed columns, which are like spreadsheet formulas, but they're very high level and you just click and drag and drop them and also in the same vein, adding AI capabilities on top of this data. So yeah, we are, our best customers are who, they have some process in the company. There's a lot of data in a spreadsheet. Maybe it's paired with some sort of automation tool that's gathering input from a form and sending out emails to make reports.


The demo (06:57)

And they want to consolidate that into something a little bit more, a tighter package, more contemporary, better design, easier for their employees to use, and harder for them to break. What, technically, when we see the demo, what's going on? Is there a, are you sending all the data to chat GPT to do a whole bunch of work? What's, how are you using AI and what's going on underneath the covers? Yeah. Do you want me to just jump into it and show and talk about it? And then you can explain what's going on underneath the covers? Yeah. Do you want me to just jump into it and show and talk about it? And then you can explain what's going on. Okay. Yes. This should not be a surprise to anyone in the audience. This is a spreadsheet and this is where a lot of our customers start. We see a lot of people managing inventory and warehouse information. That's a, it's a frequent use case for us. So here we just have a plain Google Sheet. It's got products, orders, people, warehouses. And usually spreadsheets start to break if you have people on mobile devices that need to access them. Because spreadsheets are great when you have a large screen, but they're terrible on mobile. Or when you have just a lot of people who need to access them at once. I would say one of the best ways to ruin your spreadsheet is to add a hundred people to it. And that's when you create an interface on top of software, it puts some constraints on what your users can access when people use your spreadsheet. They can change the program. They can delete the formulas. You can put some guardrails on it, but it's wily. So I'll switch over to Glide. This is the dashboard in Glide where companies can create an account and make their apps. So I've made one for your podcast, and you've got some apps in progress. You're trying out some of our AI templates to learn these features, and you have some folders to organize the different pieces of software for your business. Our successful customers will have 5, 10, 15 apps for their customers, their partners, their employees, their administrators, their executives for different stakeholders in the business. But before I blabber on too long, I'll tie back to that spreadsheet I just showed. So we have this warehouse spreadsheet and we need to make an interface. We need to up-level our workflow to be more reliable, more distributable. Basically the first thing we ask you in Glide is give us your data. So I'm going to pick Google sheets because that warehouse data was stored just in Google drive. And here's the exciting part. Glide is going to ingest all the data and apply some very basic high level rules about software design. It interprets every tab in the spreadsheet as a tab in a mobile application. It looks at the data and we can view the data in Glide. This is just a mirrored version of the spreadsheet. It recognizes numbers. It recognizes images, URLs with that JPEG at the end, for example. It applies a bunch of simple heuristics over the data to try to get a picture of the data model, and then it designs a basic application that you might want to use. So this was the one click app from data generation step that we took. This is your first experience in Glide. And the app is fully responsive and cross-platform. So it's a PWA. This is one common confusion that we create at Glide is we use this term app. A lot of people associate that with the app store. These are web applications that are responsive. You can add them to your home screen on iOS or Android. You can install them to your taskbar or your dock. The most recent macOS release added to Safari, you can install web apps right to your dock. It's a really great experience. I know a lot of people are doing that with ChatGPT. We do that at Glide with a lot of our internal apps. But this is syncing in real time with the spreadsheet. And so if I go to this people tab where it's identified people in the company, and I want to add you Robert, but Robert, I'll grab a photo. I'm going to use my photo. That's fine. Don't have your photo on my laptop. I'm sorry. And that photo is going to be instantly synced to everyone using the application. And if we switch back over to the Google side and we go to the people tab, you can see that you've been added right here, thinking back and forth. So the people in your company who are accustomed to accessing the spreadsheet, they don't change their behavior at all. So these are like the people in the office, the administrators can still work on the data and it's syncing in real time and then you can customize this. You might want the people list to look like this. You get think a lot of stuff for free. So you get search, navigation, and like I said before, responsive across platform. And it's instantly publishable and shareable. It doesn't go to an app store. It just publishes a web application. You can make it private to your company. You can send it to a thousand people. They can all use it at the same time. They won't break the spreadsheet. They won't see what you don't want them to see, and they'll to a thousand people. They can all use it at the same time. They won't break the spreadsheet. They won't see what you don't want them to see. And they'll have a great experience. So that's Glide at a high level. A tool to build something for your intranet, for your employees, to use some sort of ordering process or inventory process or stuff like that. What's going on with AI here? Because we're an AI show. I'd like to know what's going on with AI here? Cause we're an AI show. So let me like to know what's going on with you got a large language model doing all this what's going on. There was no AI there yet, but I'll show you where the AI is coming in. All right. At the start of this year, we got really excited about AI along with the rest of the world and we did a bunch of prototyping bunch of prototyping and we said, AI can do roughly one of two things for us. They can make it easier to use our product. It is after our goal to make software development more accessible. That seems like a good thing to do. Yeah. Or it can make our product more powerful and it can make the software that our customers build have remarkable properties of AI.


Where Glide can be used (12:39)

The only AI can lend to your software. And we said, that second one sounds a little bit more exciting. We think our customers would be more wowed by AI running custom in their software in their company than they would by a new onboarding that helps them get started 10 minutes back. I'll, I'll give you a demo of the thing we built. So I'm going to create an issue tracking feature for this application. I'm going to make a new table of data called issues, and I'm going to put one issue in here, leaky faucet, actually let's do a couple, let's see, huddle and all, we'll do crack pavement. There's a little bit of data. And back in the application, and one remarkable feature of Glide is we're programming against the live running app. I'm sorry, I went really quickly right there. You can just add new screens based on data, and Glide will do that simple design analysis and put the screen together for you. So this is our issue tracker. We'll report another issue. Stranger, strange odor in a break room. So this is, this feature is already live. It's published. Maybe this app is on 500 phones for the people in your company. Anheuser-Busch is one of our customers and they have apps in 16 different warehouses. And most of the people are using the software are using Android and they use it to get access to HR information and report issues like this. Okay. This is still isn't AI. It's not interesting yet. The way we added AI into our software is through a series of these computed AI columns that add AI capabilities to your spreadsheet. And let me just take a sip of water. I'm fighting a little bit of a cold Robert on my throat. It's getting scrappy. No problem.


For fun demo (14:31)

I'm jet lag came back from Spain a few days, a couple of days ago. I'm probably also going too fast. No, this is awesome. I'm enjoying this demo. This is like really cool stuff for people who have a lot of data in spreadsheets. Yeah. So I'll show you the the first action we're going to do is generate text. And this is the core LLM capability of just text completion, giving you more tokens. And we're going to do we'll do something simple. We're going to say we need an icon. This is just for fun. Given an issue reported by someone in our warehouse, output a single emoji that represents the issue. So I've given it instructions, and then I fed it as input whatever's in this name column, and the LOM is going to output an icon. So if someone reports there's mold in bathroom, the LOM is just going to follow those instructions, and it's going to give us this very nice representation of mold. It looks like a green little tree. And then back in our app, we can just, we can bind the user interface to this data. So if someone reports fire in parking lot, LLM is just going to follow those instructions and output an emoji. It's just a nice little accent. But what you really want to do is get the AI to help you make decisions in your business, categorize things for you. So we've developed a set of other columns that build on top of text generation to do pretty interesting high-level AI use cases for data, and we've developed a set of other columns that build on top of text generation to do pretty interesting high level AI use cases for data. The first one I'll show you is called text to choice.


Text to Choice (16:11)

Text to choice is just like standard LLM prompt completion, but the AI has a contract to only output one of a fixed set of options that you've asked for. So I'm gonna make these choices one low, medium, three high. I'm gonna tell the LLM you work in our warehouse and you're on the maintenance team and people are reporting issues to you. You need to determine how urgent each issue is. Anything involving fire or mold is highly urgent. So we've told the AI exactly what I just said, and we've given it instructions, and we gave it a set of choices. Let me name this urgency. One low, two medium, three high. And based on the issue, it's going to pick one of those categories. So back in our user interface, we can, one, we can show this information. Even better, we can sort our data by sorting as old school calculations. The LLM is not doing the orders that the regular CPU is on your iPhone. We can sort our data by sorting as old school calculations. The LLM is not doing the orders that the regular CPU is on your iPhone. And to improve the user interface a little bit, let's group. So now we have an AI online with custom instructions about its task, making decisions based on the data that our employees are reporting here. So if we get a lot more data, we give this to 100 employees, and they're busily reporting issues on a day-to-day basis, I'm just gonna import 25 of them, the AI is going to follow those instructions and do what we ask, faster than any person could edit the spreadsheet. If we go back to our layout, we have a lot more issues, they have their little cute emojis, but more importantly, they're being categorized for urgency. And this is this, the step I'm going to show you right here, I think is what's most important as we think about how AI is going to help us in our business and how it changes software development. Let's say our warehouse is storing volatile chemicals. This is a specific detail about our business that we want to encode into these AI instructions.


React with Raw MaterialsSome Devices Explode (18:29)

We store volatile chemicals, including aluminum and magnesium that react with water in an explosive fashion. Any issues involving moisture or water are highly urgent. So I'm programming this app. I'll just click save. The AI is going to reconsider its assignment of all the issues in our company. And I'm programming, but I'm not writing formulas. I'm not writing code and I'm not even typing. I'm just talking at an LLM. I'm using just my built-in Mac OS dictation feature, which is really good. It got a lot better recently just to save on type on your podcast. But this is I'm an engineer and I don't even know how I would code this. This is a remarkable new capability. It's very easy to use and easy to customize. That's only enabled by LLMs. So here leak in the pipes near the restroom on the first floor has been designated high, but it doesn't contain the words water or moisture. The LLM was able to make that association for us.


How Is This One Different (19:34)

So that's a very basic example. In the last few minutes, we created a totally custom piece of software that works on every device you can share with anyone in your company. And it has a custom AI running inside of it to help you make urgent decisions and flat things for you. Now, when I talk to business people about chat GPT, they're worried about intellectual property, going to an LLM up in the cloud and having that LLM use that intellectual property for maybe building a future model or do leaking that data somehow somewhere. That's a big fear. This LLM, are you running your LLM locally or using chat GPT API to do that? Yeah. So we select the best LLM for the job, whether you're working with text or images or audio and none of those LLMs is ours. A lot of the text is being handled by open AI, but yeah, the, the AI model providers are very sensitive to this concern from businesses and they're being good citizens so far. We have an enterprise agreement with open AI, which includes a lot of clear language about what they're doing. They're not training on this information. Yeah. If you're using the API, they're very adamant that they're not using that. They're not storing that data or using it for a model building. Yeah, that's right. But it's great that people are concerned about that. And what we see in companies, it's very early days, obviously. What I showed you is even for people who are familiar, they're deeply embedded in the AI space and you watch the open AI keynote. I think what I just showed you is like a novel approach to this stuff. Like we're just inventing this as we go and we're seeing where it sticks. But there's a big difference between what we call the state of nature, which is where you told your colleagues or your employees, you can expense chat GPT pro, and you can just put whatever you want in there. And then something that's structured, where you have control over the output, you can audit it, you've designed it. Someone in your company has, it's not every person on their own is asking the AI, tell me if this issue is urgent or not. It's someone with knowledge about the business, the facility, the business goals, the priorities, the details. knowledge about the business, the facility, the business goals, the priorities, the details, programming it and building it and then distributing it for other people to use as an application, not as everyone just fending for themselves with chat GPT.


AI for better software (21:53)

Yeah. This is really an interesting idea. I I'm like, oh man, how would I use AI in all sorts of ways to make my software easier to use, better integrated, right. And add some new features that just wouldn't be possible in the visual basic world that I used to live in. Yeah. The best thing we can do, we're just learning from our customers. So today I saw this, literally I saw this today from one of our Glide experts. These are people who get certified in Glide to help businesses. They've built an app for a company that audits super yachts before today from one of our glide experts. These are people who get certified in glide to help businesses. They've built an app for a company that audits super yachts before they're delivered for purchase and they have these exhaustive checklists and the yachts are moored at all these different ports. And there are local people who come onto the yacht to work on these issues. And he built with our AI system, a two-step process based on the location of the ship and the ship can move based on the location. What's the predominant language now translate all of the issues into the language likely understood by the people who work in that port and he demo. And then they came up with this. The customer never expected that this could happen, but he just said, when you park the ship in the Italian dockyard, you're going to be able to do this. work in that port and he demo and then they came up with this, the customer never expected that this could happen. But he just said, when you park the ship in the Italian dockyard, the checklists automatically are in Italian. And when you move it back to Portugal, it's Portuguese. And I, I hadn't thought about this. We didn't build this system with translation in mind, but just the remarkable capabilities. I can show you one more example. Just, I think my point is like, get the juices flowing and we've built a tool. People can explore these possibilities, but I can show you one more. Yeah. I would love it. Okay, cool. Another thing that these LLMs open up, they have this broad language understanding. They were able to see leaky pipe and associate that with water and moisture for our, our facility.


Programming with documents (23:41)

But they can also help you program with things that you usually couldn't program with in, in a spreadsheet or in a traditional programming language very easily, which are images, documents, videos, and audio. Yeah. Oh, I'll, I'll give an example using the AI to program with documents in Glide. So this is a simple invoices app. These are all the PDFs. At Glide, we use a lot of Glide. And these are the invoices that Stripe sends us on our own account for the 15 apps or so that we have active. And all the data set contains is a list of these PDFs. And we have this very simple app where you can scroll around and you can look at them and you can add more but you have to squint your eyes and see these little thumbnails wouldn't it be great if the app if the application knew a little bit more about the contents of these documents so as a general rule when you're programming with an llm you want to get a lot of text we have these two text abilities audio to text document to text these two text abilities, audio to text, document to text, image to text. So I'm going to choose document to text, point of the invoice, and I'm going to call this column content and the AI goes through and extract the text of all the PDF files and this works for Excel files, documents, whatever. Now I have something I can feed to subsequent steps. So the first thing I'm going to do is I'm just going to get the date, the due date of each invoice. Read the invoice and tell me the due date. The input is going to be the total content. And before I can even finish configuring this, it's produced all the dates and all the documents. And I only want to see the dates. And this is a new column. This is text to date. I can even finish configuring this. It's produced all the dates and all the documents. And I only want to see the dates. And this is a new column. This is text to date. So it's coercing the LLM to output a valid date instead of saying, sure, you got it. David, here's the due date of your invoice. I hope you can pay it on time. It's just going to give us the date because sometimes the chatter can interrupt your application. So now back in our layout, we can at least show the due dates. So I don't have to squint at the thumbnails.


default editor vs (25:51)

And like before I can do things like sort, sort by due date. The most recent ones are first. And when you're programming in glide, you're sometimes you're switching between layout and the data editor where you're programming in glide you're sometimes you're switching between layout and the data editor where you're working with ai you have this thing at the bottom it's called the peekaboo data view which just brings the spreadsheet like interface from our data editor into the layout editor so i'm going to just continue here i'm going to add one more column text to number read the invoice and get the total amount due. I'll give it the content. I'll tell it to put a dollar sign at the beginning, and I'll name it amount due. Again, the LLM is going to go through every invoice, read all the content, and extract a valid number representing the amount due. So now I can chart this. I can show our spending over time. I can sum all of our spending. So I'll just throw this on there, amount due so now i can chart this i can show our spending over time i can sum all of our spending so i'll just throw this on there amount due and then i'll do one more generate text read this invoice and extract the invoice number and i hope it's going to actually get this one a little bit wrong so I can show you a technique.


What a.json can do (27:05)

Okay. Content. Click save. There. And let's go to the data editor to see in more detail. You can see it's put chatter in here. Generate text is general LLM completion. The other ones are very precise. Just give a number, just make a choice, just give a date. When you're using GenerateText, you can get extra information that you don't want. OpenAI's new JSON output mode can help a lot with this. Okay. Another thing that can help is just telling it more of what you want. So you can say, example, 1234567, four, five, six, seven, eight, nine dash one, two, three, four, five, six. You can see it's getting a lot closer, but it's still got this swirly one in here and just lopping off one example character. Got it on the right track. Yeah. And now finally, we can put as the title of each of these invoice number. Now we have this application. We can put as the title of each of these invoice number. Now we have this application. We can render it as a table. We can send it to everyone in the company. They can put it on their phone. Uh, when they add a new invoice, the AI will immediately parse it and add all the structured information and we can display it for them easily on these screens, our finance department is reading through these, they can see everything more easily. So let's say I work at a warehouse and I have one of these monster spreadsheets and that's right in the warehouse and I want to start using you guys. How do you charge for that? What should I be prepared to pay? Yeah. Glide has a free tier and then we have tiers for subsequently larger businesses and it's, it's frankly. Employees in my warehouse and they're all sharing the spreadsheet, right? Yeah. That plant probably starts at about $99 a month. We generally scale with the usage. 50 is not a very large distribution for Glide. We have customers with thousands of people using the app, but generally it's on a per user basis. And if you use a lot of AI, we have a usage based currency as like credits. And some is included with the plan. And if you use an inordinate amount, we have a couple apps on our platform that are like public facing AI, play with AI apps. And we found one the other day that has stored a terabyte of music, and is transcribing the music for people for their music community. And I think quite a bit on our AI system. Tell me a little bit about your company. How many people you have working on this and how did you get funded? Sure. We're X Microsoft, four co-founders who are working on developers. Yeah. And although there's many different Microsofts, I was, I'm a latter day, I'm a Satya Microsoft guy and I, as a designer and I remember I would have a meeting where I would talk about our design goals and how design is going to influence the product. And people were like, if you said that four years ago, you wouldn't have, you would've been fired on the spot.


Views On Challenges & Ai Moonshots

Microsofts Satya Nadella (29:58)

But yeah, remarkable company. I just watched their insight keynote. Satya is a all time great CEO, incredible. And he's just getting warmed up. It seems. So yeah, we've been, we'd spent about 10 years together. We were at a company called Xamarin where we designed developer tools for enterprises, building iOS and Android apps internally, which is where we saw this phenomenon that got us interested in custom mobile software, which is one aspect of Glide. We did Microsoft for two years. I led design of developer services there, building browser-based tools for traditional software developers. And we just got the startup itch. And we said, we're good at design and we're good at developer tools. And we want to make something with a broader audience, more ambitious spreadsheets, got to 2 billion users. How do we make something big like that? Is there room for a great product that can help more people connect to things like at the time it was just software development and user interfaces and databases. And now Glide's a great platform for us to help non-technical people build with AI. We're very excited. But yeah, we did Y Combinator. We did a series A last year with Benchmark. We're about 40 people, thousands of customers, great traction. And the AI stuff is still in beta right now. I think the demos went pretty well, but that's the status of our company. Coming soon. You're one of the first crop of companies that is using AI in a new way. It's not, you're not building a chat bot competitor to chat GPT. What are you learning about AI as you're trying to program with it? Your demo went pretty well, but what are you learning from using AI as a programming tool? When you use it in a serious way, the mistakes aren't cute and they make a lot of mistakes. So that's that, that the mistakes count. So we're trying to be careful and we're trying to learn together with our developers, you know, non-technical people building with AI, make them understand the capabilities of these things when we first put the AI stuff in place, people were asking the AI to do calculations, arithmetic. LMS are not good at that. No. And we had to say, we need to tell people don't ask this thing to do math for your business. But the subtle thing, the specific thing we're learning is that. Probably by the end of next year though, there'll be a math model, right? That's a large language model, right? It's good language, not for math. But there are other ways to do math. There's other models that you could pass that off to. You don't have that kind of integration yet of clouds or of one guy calls it orchestras of AIs where you have many models working together on a problem. We do. We, we actually I just, I did show that the cooperation between that document step and the analysis was two different clouds and two different models, but you could also bring your own and use API integrations to connect to them but yeah, we're learning that in a very there's very subtle applications for AI at this sort of classification ingesting document layer. It's not all chat bots and one prompt builds your entire application. There's a lot of utility in these subtle details. And that's pretty exciting to us to be at the frontier of that. Very cool. What else are you seeing? Let's just talk about the world of that. Very cool. What else are you seeing? Let's just talk about the world of AI. What are you seeing that you're paying attention to in the AI world? Are you thinking about how to use like a llama model? One of these smaller models, other than the chat GPT model inside your app to add more capabilities for your customers. What are you thinking about right now? Rag and a vector indexing of all of the customer data is something, one of our most requested features. Rag is you're using a term. Some people don't haven't heard before. He's evil augmented generation. I, I know I've implemented it.


Challenges (34:08)

I'm not exactly sure what the, I think is retrieval augmented generation. But yeah, you embed all of your content, for example, maybe all the rows in your spreadsheets in a vector space, which is like a traversed path through the neural network of the LLM, which produces a, a coordinate, a point in semantic space. And then your users can type a search query and you can do this sort of unstructured search, semantic search. And we have customers who have built this, load all of the instruction manual, you take all the PDF instruction manuals. One of our customers builds a field service software for people installing and maintaining emergency equipment in hospitals. And they have five gigabytes of PDFs with the manuals for the electric stuff. And last year they had a binder. Now they have an app and they index all that content by extracting the text. And then you can type, the technician can type in their native language. It doesn't have to be English. And it will find the right content. It will translate it if they asked a different language, just like the yacht maintenance boat. So that's the yacht maintenance app. That's one area we're really excited about. Next is like the clear next step is helping people index and ask questions about their data. And to that point of calculation, yeah, asking qualitative questions. Who in my company is the best person to help me with this problem? Or quantitative questions. What was our average sales last Black Friday? Those would go to different engines.


AI Moonshots (35:45)

So yeah, figuring out how to split those queries and index the data differently. That that's what we're thinking about a lot, but yeah, I think you meant more broadly in AI. We were, we watched the open AI keynote. We're watching the Microsoft keynote. It used to be this joke that the Apple keynotes were happy. Apple kills your startup day. So we're watching so excited about the new AI thing, but wondering, okay, did they just make us obsolete? I'm really interested in the new assistance API, and we're building the next generation of our onboarding to help people even think through what they're trying to do with our software. It's not that people come with a fully formed opinion of the prompt that's going to get them their goal. They have a vague idea of a problem. They don't even speak the language in our case of software development. They don't know that the language in our case of software development. They don't know that the little icons at the bottom of your iPhone app are tabs. They don't know to say that. So helping people even formulate and evince their own problem statements and the solution that can help them is really exciting to us like an AI PM that helps you. But yeah, I think it's, we are just as entertained and awestruck and inspired and excited and nervous about all these changes that are coming so quickly. Like everyone. Uh, I think we, we have some AI expertise, but it's everyone. It's like, everything's just turning over all the time and it's the most exciting time ever to be building a startup, to be in tech. I think everyone's just thrilled and really excited to see where it goes.


Concluding Remarks

Glide Fort (37:07)

I wanted to talk to you because you actually have real customers in real warehouses and doing things in the real world, a lot of AI companies are still about to go to the customers. What are you learning about how normal everyday people are looking at both software, but generally AI-based software? How are they thinking? Are they afraid of it? Are they excited about it? What are you noticing on the street? I would say people are curious. AI still requires way too much of a creative leap to apply to your business problems. We at the, we're thinking about AI all the time and I can make a demo. We have, we're approaching 10,000 apps this year connected to our AI services. So people are trying and they're really curious. I would say it's very early and these are new patterns. New patterns in the data and computation and logic side where a system could make an unstructured determination and classification for you. It's just a new capability that no one's wielded before. And on the user interface side, when you have an LLM that can interpret and structure information for your own users, we're discovering new paradigms there as well. One that we see often in our apps is if someone has a detail screen for their business with a business object, like a customer, maybe at the bottom, you have this sort of comments from people in the company, just using the LOM to summarize those comments as three bullet points above the fold at the top of the app is like a really interesting new UX move you can make. above the fold at the top of the app is like a really interesting new UX move you can make. So I'd say we're in a learning phase. Companies are interested. People are busy. I think that it's a great time to be someone with doing a startup or trying to translate this, this exciting new world to traditional businesses and help them adopt it. They need a lot of help. This one is what I would conclude. Totally agree. And that's why I wanted to talk to you and understand what's going on in the world of business. Thank you so much. This is awesome. Where do we learn more about glide and get started? Yeah, you can go to glideapps.com and we were, we're not able to get glide.com from a real estate company. There's a few glides out there, but glide apps.com is where we are today. And if you type glide.new into your browser, you'll get to the screen where I started my demo basically, where you can create an app. Very cool. Thank you so much.


Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to Wisdom In a Nutshell.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.