Description Transcript
Gen AI has turned every employee into a potential builder, creating a massive opportunity for new teams to build the internal tools they need to move faster. But to turn this momentum into a strategic advantage requires more than experimentation, it takes a unified, secure environment where builders can ship confidently at scale.
In this webinar, we explore the Internal Builder Era and share a practical AI Maturity model for scaling from ad hoc automation to governed, production-grade application. You'll learn to empower more builders across the business without creating tool sprawl, brittle workflows, or security tradeoffs.
We cover:
How to enable the next wave of internal builders with the right foundations, patterns, and guardrails.
A practical enterprise AI maturity model for moving from experimentation to scalable automation and real business impact.
How Retool helps teams build production-grade apps with AI on your data, in your cloud, with inherited enterprise controls.
Read more 0:00 All right, welcome everyone. Just 0:01 getting started here for our webinar. 0:03 Let me share my screen. 0:14 Great. Thank you all for joining. Um 0:17 we're here for the retool webinar to 0:18 talk about new internal builders and 0:21 specifically how we can empower them 0:23 without sacrificing some of the core 0:25 tenants like security. Um, some 0:28 administrative things before we get 0:29 started. This is going to be a recorded 0:32 webinar. So, we'll be able to share out 0:33 the link to the recording for anyone 0:34 else on your team who might be want to 0:36 view it. And we are going to have some 0:38 time for Q&A at the end. So, if you have 0:40 Q&A, feel free to add it to the chat and 0:42 we will address those after the 0:44 presentation. 0:47 Um, but I really wanted to thank you all 0:49 for coming. Uh my name is David Swan and 0:51 I am one of our enterprise solutions 0:53 engineers here at Retool which basically 0:55 means that I work with our customers and 0:58 prospects to help understand where does 1:00 retool fit inside of their organization 1:02 and how we can help solve problems 1:06 and that kind of leads into sort of the 1:08 conversation because it's not you know 1:10 how we can help solve problems but who 1:12 is best poised to use retool to help 1:14 build those solutions out. 1:17 So if I go in and check out our agenda 1:20 today, first step is we're going to talk 1:22 about sort of the state of software 1:23 development. What are our current 1:25 challenges? What is retools approach to 1:29 AIdriven development and deployment? And 1:32 that is really something I want to 1:34 reference a lot is sort of it's not just 1:37 development but it's also that 1:39 deployment that last mile problem. And 1:42 we don't want to spend too much time in 1:43 slides. So we are going to get into a 1:45 demo where we're going to highlight a 1:47 couple of things. First and foremost, 1:50 the new development paradigms within 1:52 reach. How can we go from prompt to app? 1:54 How can we edit with a prompt? How can 1:56 we still edit with a you know a cursor 1:59 click for lack of a better term? 2:01 And then we'll again we'll save some 2:03 time for Q&A at the end. 2:06 I want to thank you all for joining and 2:08 with without any further ado jump into 2:11 the content here. So 2:14 what is like the whole premise for why 2:16 you're here today? It's not just to 2:18 apply security on top of software 2:21 development, but it's addressing the new 2:23 reality of software development. And 2:25 it's that it has a lower barrier than 2:28 ever before, meaning that new people can 2:31 come in and build, meaning that we're 2:32 quicker to get things sort of off the 2:35 shelf. And that has been really enabled 2:39 by some technology advancements as well 2:41 as some frankly some fits between how 2:45 people approach the problems and sort of 2:47 psychological components. And that 2:49 technology side is where AI native tools 2:53 have really come in to bridge that 2:55 imagination and skill gap. When chat 2:58 came out in 2022, 3:00 the idea of having sort of the world 3:02 exposed to any question was completely 3:05 revolutionary. 3:06 And now the advent of LLM assisted 3:10 development 3:12 has continued that sort of imagination 3:14 of where I can take sort of any idea, 3:16 not only ask a question, but I can 3:18 create take any idea and sort of build a 3:20 solution. You describe it, AI builds it. 3:25 that has dramatically changed who would 3:28 we consider to be a natural builder of 3:30 software solutions 3:32 and it's why we have seen this speed of 3:35 prototyping be truly unmatched 3:40 but that prototype is what I want to 3:41 interact a lot because we sometimes see 3:45 people are able to build really cool 3:47 things but not get it out of a local 3:49 host application or maybe not get it 3:51 connected to a real resource and these 3:54 Ad hoc autom automation 3:57 is where a lot of companies try to 3:59 address this. Try and build in either 4:02 tooling or backlog triaging in order to 4:06 address you know sort of how can we take 4:07 it from a local host application to 4:09 something that is widely sharable with 4:12 your entire organization on a real 4:14 production grade solution. 4:18 But that ad hoc automation sometimes has 4:20 ad hoc challenges. And while it meets 4:22 this like overarching need from a 4:25 technology side and it meets this sort 4:28 of psychology need right where people 4:30 want to be able to solve their own 4:31 problems right and AI is doing that it's 4:33 giving them this ability to have 4:35 self-sufficiency in their solutions 4:38 and that connection across the journey 4:41 has been incredibly impactful from going 4:45 to this entire life cycle of through 4:48 this entire life cycle of development 4:54 And that kind of was the original 4:56 promise, right? We were going to do PC's 4:59 on top of AI, introduce our 5:01 organizations, and see it deployed, see 5:03 all of these solutions out there in 5:05 reality. 5:07 And I don't know about anyone else on 5:09 this call, but I talk to a lot of people 5:11 and use AI every single day myself. You 5:14 know, retool is AI first. We have this 5:16 deployed across our organization, 5:17 different capacities. 5:19 But this curve is rarely the reality, 5:23 right? And that's because 5:25 we have this initial inspiration phase 5:28 where we're going from a PC or some sort 5:31 of initial pilot and introducing it 5:33 maybe in a department, but then when it 5:35 comes out to sort of get that hockey 5:37 stick level growth, that you know 5:40 enterprise level adoption, 5:43 that's where we're seeing all these 5:44 challenges being met. 5:49 these roadblocks being hit where we're 5:53 going in and struggling to actually 5:57 deploy them out, scale these solutions 5:58 out and connect them to real resources. 6:01 And that's where I don't think we have 6:02 this natural curve. In reality, we kind 6:04 of have an an S-curve and most 6:06 organizations are stuck at that plateau 6:09 where AI's introduced that PC maybe had 6:11 some departmental adoption, but it's 6:14 stagnated 6:17 and that's where we want to try and 6:18 address the problem and build bring a 6:21 solution that gets that true 6:25 prototype to production pipeline 6:27 realized. 6:30 And this is happening, you know, this 6:32 sort of plateau and people getting stuck 6:35 in the mud here because of a couple of 6:37 fundamental truths. 6:40 While that bar to prototype has 6:42 disappeared, the bar to ship hasn't 6:45 moved. 6:47 We still have a high bar a high bar of 6:51 security development standards and 6:53 testing even that are required to get 6:57 things out the door. And when we look at 7:01 studies, whether it be from McKenzie or 7:03 from other internal evaluations or other 7:05 studies, we're seeing sort of 1% of 7:09 companies actually implement their AI 7:12 solutions in production. 7:14 And 95% of the pilots that don't make it 7:18 anywhere, they stay stuck at that 7:21 plateau. And it's not because these 7:23 aren't amazing ideas and local host 7:26 applications that have been built out, 7:28 but it's because there's a backlog in 7:30 order to meet the enterprise needs of 7:32 development. 7:34 And while AI driven development has been 7:37 dramatic, 7:39 it does not mean the same thing as 7:41 AIdriven deployment. 7:44 And there are certain components that we 7:46 need to make sure we include to get that 7:48 out the door. And that's kind of this 7:51 lurking cost of infinite development 7:53 because the more applications that we 7:54 build 7:57 then we have this you know UIX 7:59 prototypes and that's that that initial 8:01 promise like I keep mentioning but for 8:03 each one of these prototypes 8:06 there's systems that need to be 8:08 connected there's authentication like 8:10 SSO there's governance whether it be 8:12 from data or it be from actual 8:16 functionality of the application there's 8:18 things like deployment 8:20 where are we going to put the 8:21 application right how are we going to 8:22 have build into like a true you know 8:24 CI/CD process build out native testing 8:27 or connect to your existing solutions 8:30 that work is still manual and with all 8:34 of these prototypes 8:36 it has exploded out 8:40 so much so that there's a I don't know 8:43 like may two organizations but heads 8:47 where everyone is now empowered to go 8:48 build prototyp types and explore the 8:50 auto possible. They have this new status 8:52 as builders and they're truly empowered, 8:56 but they get frustrated that last mile 8:58 problem. 9:00 And whether that is a frustration 9:02 because the AI solution isn't getting it 9:05 across that last mile or it's causing a 9:07 sort of butdding heads between the 9:11 nontechnical org and the technical org. 9:14 It's resulting in frustration. it's 9:18 resulting in a backlog that's growing 9:20 and sometimes code that doesn't meet the 9:23 standards of the organization 9:26 and where that backlog happens as has 9:29 been true for as long as software 9:30 development has been around shadow IT 9:34 comes around or people go around 9:36 engineering go around IT maybe third 9:39 party solutions are bought or maybe uh 9:42 security steps are sort of skipped and 9:45 that expanding backload uh has caused 9:48 those impossible trade-offs 9:50 where people need to in order to get 9:52 their solution out quickly maybe 9:54 introduced new vulnerabilities or new 9:57 weak points or failure methods. There 9:59 was a study that came out last week that 10:01 said basically that 4x the velocity 10:05 resulted in 10x the amount of 10:06 vulnerabilities. 10:08 You know, sort of that that rush to 10:10 production motion was causing problems. 10:13 And we're seeing this across the entire 10:17 SAS space, right? There's been scenarios 10:21 where databases have been dropped. 10:22 There's been data breaches or just 10:25 solutions that maybe are uh don't have 10:29 the consistency and the reliability that 10:30 we would expect. 10:33 And that's where I think we have a ton 10:35 of sort of ability to help balance the 10:39 scales to help connect the 10:43 problem at hand to a situation where we 10:46 can get the actual 10:49 solutions into production. Because if 10:51 you look at this as like a scales 10:53 perspective, right? Like we have all of 10:55 these applications could be forms, 10:58 portals, cred apps, uh you know, sort of 11:01 all these different systems that 11:02 integrate or are ideally going to 11:05 integrate to your production systems. 11:07 But in order to do that, 11:10 those governance concerns need to be on 11:12 the other side of the scale. And shadow 11:15 it is sort of pushing that imbalance, 11:20 making it hard to ever truly make time 11:24 and again forcing those trade-offs. 11:28 And that's our goal here at Retool is to 11:31 bring in a solution that kind of 11:32 balances those scales. Whereas if you're 11:35 building out whether again it be the 11:36 forms, portals, different systems, we 11:39 have a platform that applies these 11:43 enterprise standards of security so that 11:47 shadow IT or these impossible 11:50 trade-offs, right, don't even have to be 11:52 concerns. Whether you're building out 11:54 three applications, 30 applications, 300 11:56 applications, 11:58 we're applying this once 12:00 and we're applying these enterprise 12:02 concerns or considerations 12:05 for everyone across the or every 12:07 application across the or 12:11 kind of the promise of retool. We want 12:13 to improve that development so we can 12:15 get 10x the applications. We can 12:17 introduce new builders and we can still 12:18 meet that promise of AI where we're 12:21 giving sort of a new found uh user base 12:24 to create these solutions without having 12:27 to sacrifice any of the security 12:28 concerns. 12:31 And to do that I'm going to talk about 12:32 sort of the retool approach to 12:34 development 12:36 which is three steps and then I'm going 12:37 to showcase it. 12:40 So what are those three steps? Step one, 12:42 we have to build an application, 12:44 right? Uh this is 2026, so we want to 12:47 build with a prompt, right? We want to 12:49 be able to use natural language to 12:51 trigger an LLM to go build an build a 12:54 solution 12:56 again with security first. So being able 12:58 to bring your own key, you organizations 13:01 largely have sort of a existing 13:04 relationships and existing security 13:07 agreements with the major LLM web 13:09 providers. 13:10 We don't train on your data. We don't 13:12 ask you to sort of ignore those existing 13:15 rules. said bring your own key to be 13:17 able to do it securely and with inside 13:18 of your existing solutions 13:22 was eventually how do we build and 13:24 retool again natural language the prompt 13:26 being able to drag and drop in that same 13:28 IDE mixed modal development 13:31 and then being able to write code still 13:35 be able to write SQL write Python write 13:38 JavaScript use existing languages that 13:40 people already know and support and 13:42 being able to edit and inspect where 13:43 that is AI driven and AI generated. 13:46 Once the application is built, there's 13:48 fundamental things that need to happen 13:49 to make it production grade. Step one 13:52 needs to be governed. That governance 13:54 steps in with connecting to real data. 13:57 So having an LLM use an existing 13:59 resource connection so that it's 14:01 connecting to Salesforce, Snowflake, you 14:04 know, whatever your API of choice may be 14:07 to actually get real data in there. but 14:10 do that in a way that security is going 14:12 to improve it because it's using already 14:14 existing resource connections and it's 14:16 leveraging things like role-based access 14:18 control in your SSL and then deployed in 14:21 your flavor of choice whether that be in 14:23 your cloud like your AWS account or that 14:26 being in sort of a retail SAS 14:28 environment 14:30 we've got a built application that's 14:32 connected to real data ready to be 14:34 deployed but to make sure it's reliable 14:36 and tested is that last mile right where 14:39 it needs to be fully integrated into 14:40 your SDLC and this can be connected to 14:45 your existing solutions right whether 14:47 you're connecting to existing testing 14:49 platform like playright or cypress or 14:52 using source control using moniing 14:54 monitoring inside of retool or porting 14:57 out that observability statistics out to 14:59 like data dog or splatter 15:03 how an application is built there times 15:05 it needs to be debugged and having a 15:07 clean interface to debug bug surface 15:10 things like what is our linting issues 15:12 like actually have a console to you know 15:15 go through the interactivity of the 15:18 application is that last mile that keeps 15:20 the applications reliable and usable 15:23 because the moment 15:25 a fails is where we lose the trust and 15:29 that is something that is a huge limiter 15:32 when it comes to getting AI generated 15:35 apps into production is when it hits 15:37 that first stumbling block so we build 15:39 in a secure manner that's governed and 15:40 ready to be launched. We can help 15:42 address that. And that's kind of the 15:43 retool approach, being able to connect 15:45 to your production systems and get that 15:47 value of having that mix development but 15:50 launched in this world that is approved 15:52 to get the actual applications in front 15:55 of the real users. 15:58 Now, I've walked through from a slide 16:00 perspective sort of why this is a 16:02 problem, sort of what is our approach to 16:05 help get the applications out there in 16:06 production. But I think it all comes 16:09 down to reality. So I'll walk through a 16:11 quick demo of how we can support those 16:14 uh sort of natural language style 16:15 development modes and then deploy it 16:18 within best practices. 16:21 So if I jump in here, first thing I want 16:24 to showcase is the retool that we 16:26 already know and love. But you'll notice 16:27 that we have what are we building today? 16:29 Whether you want to launch a new 16:31 application just from the homepage or go 16:33 and create a new app, 16:36 there is the 16:40 ability to quickly go in and make sure 16:44 that we have sort of this this 16:47 connection between all of our different 16:50 uh tools. 16:52 And I am just going to connect here to 16:56 my 16:57 new app. 17:00 And this is we are in live. So I'm going 17:03 to go to my solution. We're going to 17:05 work backwards where I can quickly go 17:07 and see that I have an application and I 17:11 want to showcase how I built it, right? 17:13 From being able to see all of the 17:16 different support tickets, right? like 17:18 the idea of, you know, having anyone be 17:21 able to build the right solution is 17:23 being able to support their personal 17:24 workflows. So, if I was on the support 17:26 team and I wanted to be able to triage 17:28 all of my different support tickets, be 17:30 able to go in and say, "What's the 17:32 priority? This one is high. I want to 17:35 update that. I want to change my assigne 17:38 and reassign this to a team member 17:40 because I'm going on vacation, right?" 17:42 and be able to update these steps or 17:44 even export it and send it off to a 17:46 downstream system and have this 17:48 connecting not only just linear but also 17:51 our employees uh registry to pull in 17:55 those different assignees and again 17:57 export to external action. But how do we 17:59 get here? If I edit my application, 18:10 we're going to see the same existing 18:12 development modalities of retool with 18:15 one new addition assist where I could go 18:17 in and I could like I mentioned still 18:20 drag and drop to expand or I could 18:22 prompt that. And that's exactly how I 18:24 built this application is I went in and 18:27 I was able to start purely from a 18:31 prompt. 18:33 We scroll back to the top. We'll see 18:37 our blank canvas was right here 18:42 where I was able to walk through create 18:44 that application. And because this is a 18:48 real system that I'm connected to an 18:50 actual linear resource, it's going to do 18:52 a full human in the loop development 18:54 cycle where I can walk through build out 18:58 a plan that's going to support those 18:59 users and then walk through each step of 19:02 that plan and I can integrate it 19:04 throughout the process. like tell best 19:05 practices integrate with more steps and 19:07 actually have it build out my 19:09 application connecting to real resources 19:12 writing real GraphQL in this experience 19:15 or real SQL real JavaScript making real 19:18 API calls 19:20 so much so that we get to an end state 19:23 of a working application. 19:27 I'm going to refresh my page here and 19:30 show again that same experience 19:33 where I was able to walk through and not 19:35 only 19:37 get our application built out with full 19:40 export functionality and seeing our 19:42 assignees but integrate a second 19:44 database connecting to retual DB to pull 19:48 in the employees 19:50 calling not only the resources but the 19:52 components to target my native uh prompt 19:56 developments. ment style. 20:01 Now, what I've shown is a pre-built 20:04 application, but I can go in and I can 20:06 continue to build this out. something 20:08 you know whether it's simple as or 20:10 complex as adding it in a database or 20:12 more simple like changing the styling 20:13 like 20:15 using modern design best practices 20:22 and drawing inspiration 20:25 from high design 20:30 fidelity apps 20:34 make this app feel more modern with a 20:40 dark theme. 20:43 And just like I used uh natural language 20:46 to build the application, I can use 20:49 natural language to edit the 20:51 application. And in this scenario, it's 20:53 going to do stylistic edits, right? You 20:55 know, I come from a data background and 20:59 sort of like building out those 21:00 solutions, but I don't have those 21:02 natural design systems natively built 21:05 out. And so I want to integrate those 21:08 into the development process 21:11 using those modern aesthetics, drawing 21:13 inspiration from other brands, 21:16 applying themes and the out of the box 21:19 retail components to make this happen, 21:20 too. But the quick prototyping and 21:23 iteration that I'm seeing has been 21:24 incredibly valuable. And it's going to 21:27 go in and add these steps. But if I ever 21:28 wanted to change something like remove 21:30 that target icon, I can quickly go in 21:33 and I'm going to then not only be able 21:36 to see these steps, but I can also 21:39 govern. And I'm going to let this run in 21:41 the background while we talk about those 21:43 security steps. Because while we're 21:45 talking a lot about these resources in 21:46 this scenario being linear and retool 21:48 DB, 21:50 there's so many different resources we 21:52 can connect to and out of the boxes 21:54 retool. This is what provides that first 21:57 layer of security. Being able to go in 21:59 and have an administrator 22:01 or anyone with the right role-based 22:04 access control define what resources we 22:07 want to connect to, set the 22:08 configuration so that each end user can 22:11 connect to BigQuery, can go and connect 22:14 to your GraphQL endpoint or whatever 22:16 source it may be, whether that be 22:18 dragging and dropping or using natural 22:20 language like I showcased with linear. 22:24 And these can be again connected to your 22:27 real resources. But then there's also 22:29 the need to be able to govern and sort 22:33 of secure this application for lack of a 22:35 better term. And one step of that is 22:37 source control, right? We have most 22:40 organizations have connections to things 22:43 like GitHub or Bitbucket that do that. 22:46 And that's what we're able to do here 22:48 with Retool as well. And as it continues 22:50 to build this out and continues to work 22:53 through the editing motion, we're going 22:55 to publish the application and actually 22:56 connect it directly 22:59 to my GitHub instance and show a real 23:01 poll request and show how if we were to 23:03 go in to your development environment, 23:07 we're going to see applications that are 23:08 protected that need to be able to 23:11 actually create branches so that they 23:14 follow that proper SDLC best practices. 23:17 We especially in the era of sort of more 23:21 application development and more users 23:23 being onboarded. We want to make sure 23:24 that that is done in a way that supports 23:27 new users, new growth and keeps the 23:30 security top of mind 23:34 and jump back over where we should have 23:37 our stylistic changes all done 23:42 again. got that beautiful application 23:44 and I could publish it out. You see my 23:47 rounded edges, my dark theming still 23:49 with the ability to go do all of that 23:51 functionality. 23:53 I'm going to publish it to my publish 23:55 folder 24:02 and 24:04 we're going to go in and we are going to 24:05 quickly 24:09 then showcase how if I go back to home 24:12 our applications, our ticket 24:13 prioritization tool 24:17 can 24:21 be 24:24 connected directly to our uh source 24:28 control instance. So again 24:32 we walk through and where we have these 24:37 branches 24:40 jumping across from again protecting the 24:44 application and whenever a change needs 24:46 to be made quickly building a branch 24:49 walking through that merge process 24:51 creating a PR and using your existing 24:53 SDLC your existing SSO and your existing 24:56 security best practices. 24:59 And that's where retool takes the sort 25:01 of application best practices out of 25:04 purely a 25:07 um 25:09 sort of prototyping world and takes 25:12 those final steps 25:15 from launching 25:17 to securing and monitoring these these 25:20 solutions. 25:24 Now, we've walked through a lot here in 25:26 30 minutes or less as promised, but one 25:28 thing I do want to highlight is it's not 25:30 just sort of how we build the 25:33 applications, but it's how we support 25:34 and launch them indefinitely. And that 25:36 was the sort of theme of a recent study 25:41 that we did where sort of this build 25:43 versus buy shift where interviewing you 25:46 know almost thousand different customers 25:49 and pulling in sort of different data to 25:51 understand where are organizations 25:53 making the decision to buy off-the- 25:55 solutions and build customly and when 25:58 they make those decisions what are the 25:59 concerns top of mind and from this 26:02 research we're seeing a lot of sort of 26:03 top of mind considerations to get those 26:06 build solutions out the doors that last 26:07 mile like I've mentioned from the 26:11 launching to securing to integrating 26:14 with your STLC. 26:17 Really appreciate everyone's time and 26:19 with that I wanted to open up for some 26:21 Q&A. Please feel free to put your Q&A in 26:24 the chat and we'll go through them one 26:27 at a time. 26:36 Great. So, it looks like the first 26:40 question here as I'm let me read them 26:42 through as they're coming in. 26:46 Um, 26:49 so there was one question about will we 26:51 be able to add images to show the 26:52 designs? We want to drive the 26:54 inspiration, right? We are actually 26:57 working on that this year. being able to 27:00 pull in the whether it be like a a Figma 27:03 design or a 27:06 a PDF or an image file, 27:10 being able to use that as sort of 27:12 inspiration and as the original prompt 27:14 is something that's actively on the road 27:16 map and on track to come later this 27:18 year. 27:21 And then that also is going to connect 27:22 to you know the edit mode of these 27:24 development you know whether that is 27:28 like how does that interact with AI to 27:30 ensure performance right there's a 27:31 question about how do we make sure there 27:33 aren't delays across interactions 27:36 um 27:38 and the AI system running in the 27:40 background is running in the development 27:44 right so you noticed as I was going 27:46 through that process when I asked it to 27:49 redo do the design styling, right? It 27:54 took some time to work through those 27:56 steps to inspect every single component 27:58 to go and do research on the internet to 27:59 find brand inspiration to make it 28:01 modern, dark themed. That's when the LLM 28:04 is taking action and that's where maybe 28:05 you could see some, you know, uh, lag, 28:09 for lack of a better term. But outside 28:12 of sort of the active assist 28:16 working, 28:18 there shouldn't be the LLM uh causing 28:21 those latency issues. And so that if 28:24 you're seeing that, we'd love to learn 28:26 more. Um please reach out to your 28:28 account team or the community forum. We 28:30 can drill in a little bit deeper. 28:34 Um there was a question about UX and 28:37 design and that is really exciting too 28:40 because one of the things that retool 28:42 has always been known for is how can we 28:44 use your existing libraries and sort of 28:46 your existing best practices. And so if 28:49 I jump back into our 28:52 demo here and we go back into a 28:57 um ticket prioritization tool and edit 28:59 it, 29:01 you'll notice on the left here we have 29:04 preloaded JS and preloaded CSS. 29:09 As we change the building modalities to 29:11 be more LLM native, we're not changing 29:14 the core competencies of retool and sort 29:17 of what separates the platform from 29:20 building out prototypes to production. 29:22 Still the ability to upload whether it 29:24 be pre-loaded JS or your custom 29:26 libraries at the application level, 29:29 JavaScript or at the workflow level. 29:31 Python 29:37 um token interaction because retool is 29:40 uh bring your own key. We do uh sort of 29:44 integrate with your LLM. So you should 29:46 see token you know implications of that 29:50 but it would you know sort of be 29:52 integrated to your uh existing LLM of 29:55 choice. If you're using retools provided 29:57 LLMs, uh that is a conversation we'd 30:00 love to understand sort of like and make 30:01 sure that we have the right uh account 30:03 plan and agreement. 30:07 Uh there can we have different branches 30:09 develop different features at the same 30:10 time? Yes, different users can go in and 30:14 create uh branched applications, right? 30:18 And that we see all the time with uh 30:21 sort of collaborative style building. 30:24 So there is a demo that I'm working on 30:28 with one of my peers, right? And so if I 30:31 pull it up right here, 30:34 we're going to see, 30:36 you know, healthcare example where I can 30:38 build a really beautiful UI that's 30:40 highly interactive, but we have it 30:42 connected to source control. So if I 30:44 were to go in and create a new branch, 30:46 my peers can also create branches at the 30:48 same time and then go through the same 30:50 exact PR process, commit and merge that. 30:56 And as long as it doesn't cause any 30:58 conflicts with the main or with any 31:00 other development, right, that we're 31:01 seeing across my peers that I'm 31:03 co-working this with, 31:05 we can have those branches interacting, 31:07 being built at the same time, keeping 31:10 that sort of best practices, but being 31:12 able to move fast and collaborate. 31:20 You guys are adding a bunch of great 31:22 questions here. We really appreciate it. 31:24 Um, sorry, I just was going through 31:26 these one at a time. So, yes, different 31:28 branches. We can have people break those 31:31 out to create different features at the 31:32 same time. Um, and as we just wait to 31:36 these next ones here. 31:44 Ah, this is a great question. Um so this 31:47 one is how does retool AI assist have 31:49 context to help you perform and select 31:52 and write queries to external data 31:55 sources like Postgress that we host in 31:57 Azure how does it understand the 31:58 database structure 32:00 that is the beauty of using a 32:03 pre-connected resource 32:06 is that just like when you are currently 32:10 building in retool right so if I were to 32:13 currently go into this application that 32:14 I was building I were go to, you know, 32:18 create a new resource, right? Go to my 32:20 employees query. I see I have the schema 32:23 not just for the employees table, but 32:26 for all the tables that I have inside of 32:30 retail database as I connect to linear, 32:34 right? Again, I can see the schema. I 32:37 can explore what's inside of here and I 32:39 can see the output for when that query 32:41 is ran. 32:43 that same level of exposure can be 32:45 accessed from. 32:48 So CIS is able to go and basically 32:51 understand the schema of the 32:52 pre-connected resource to understand how 32:55 to write this query 32:58 and you know that is that same level of 33:01 exposure that I'm seeing as the end user 33:03 is essentially what we're exposing out 33:04 to the LLM to be able to build out the 33:08 query. It's not the data, right? We're 33:10 not just like you are with retool, 33:11 right? We're not hosting or uh storing 33:13 caching your data, but we're using an 33:15 LLM to to write and kick off a query. 33:19 And that's where again that whole idea 33:21 of like how do you get an application in 33:23 production? Well, we do it by having 33:26 connected resources that adhere to your 33:29 security policies and meet your existing 33:32 standards. 33:36 So just for those who've joined late, 33:38 we're walking through the Q&A just 33:40 walked through a really great question 33:41 on how does um assist understand your 33:44 database to be able to write the right 33:46 query with inside the right structure. 33:47 Right? Like when I said connect to 33:49 linear, how did it write GraphQL versus 33:51 making API call? Because linear in my 33:54 system is a preconfigured GraphQL 33:56 endpoint. 34:00 All right, next question. How does one 34:01 secure against AI? learning and 34:02 capturing sensitive data result align 34:04 with internal AI policies. 34:07 Great question. There's two ways we 34:08 approach this. One, 34:11 bring your own key using LLM that 34:13 already have rules about training and 34:16 understanding. Um, and the one that is 34:18 supported and defined as your, you know, 34:21 internal agreements that that is a way 34:23 that a lot of companies, you know, use 34:25 reachable AI to already meet the 34:28 security rules. And then secondly, you 34:30 know, we are using an LLM to write and 34:34 to build out applications, not 34:36 necessarily to analyze the data itself, 34:39 right? So there is a um a different 34:43 level of data that we need to be able to 34:45 create than may would be to sort of 34:48 create that that end analysis. And those 34:50 are are different functions that we can 34:53 leverage inside of retool and really 34:54 choose what we expose and what we share 34:56 with the LLM. 35:01 um just working through these questions 35:04 and appreciate all of the great 35:06 feedback. 35:08 How does someone secure against AI 35:09 learning capturing sensor data? So, we 35:11 just covered that one with bring your 35:12 own key iss 35:19 quickest way to build your own CRM 35:20 project management tool. 35:22 Um, so step one is to 35:27 understand sort of what you would want 35:29 that serum project management tool like 35:31 being able to concretely describe it in 35:34 two or three sentences as a prompt at 35:36 step one. If you want to start with, you 35:40 know, sort of demo data, we can do that 35:42 and then in step two connect it to your 35:44 resources. 35:46 But once you have your prompt and once 35:48 you have maybe the UI built out with 35:50 assist, you can then connect it to your 35:52 resources to find those in retool like I 35:55 showed in that create new resource 35:57 screen and then tell the LLM to replace 36:00 any demo data or any pre-provided 36:03 examples with the actual production 36:05 data. So kind of building it out 36:07 iteratively in couple of different steps 36:09 is a way that I really like to approach 36:10 those problems. 36:14 Um 36:17 the there was a a question on MCP 36:19 servers. So we're spending a lot of time 36:22 on retool assist because that is the way 36:25 that we can support sort of the natural 36:28 language sort of prompt to appgen level 36:31 capability. 36:33 uh MCP servers are already supported in 36:35 the agents part of the platform which 36:37 allows for automated backend actions and 36:41 we can call external MCP servers to be 36:43 able to go do those actions and leverage 36:45 them as tools 36:47 and I know that we'll have future 36:49 webinars and other conversations that 36:51 drill deeply into agents. Um and if 36:53 there's any questions please reach out. 36:55 We'd love to walk through those that 36:57 product with you. 37:01 Um and then language settings besides 37:05 English available. Um so retools can be 37:07 fully internationalized. Um we have 37:10 customers all around the world. We 37:12 support different languages today. And 37:15 because we're leveraging LLMs which also 37:17 are multilingual those prompts today 37:20 largely do support, you know, if you 37:22 wanted to write it in Spanish or Italian 37:24 or or other languages. Um, and then the 37:27 front end application that it builds 37:29 out, as long as you have 37:30 internationalization set up, we should 37:33 be able to support you. 37:37 And then the last question I have here 37:40 is one of my favorites because it 37:43 leverages workflows, which is probably 37:45 my favorite part of the platform. Uh, 37:47 for those who are not familiar, 37:49 workflows are our back-end automation. 37:51 So it's that if this, then that style 37:53 logic. I want one block to happen which 37:56 executes another block and another 37:58 block. These are created with uh today 38:02 manually 38:04 but we are incorporating into our 38:06 roadmap being able to build these out 38:08 with assist and there is maybe a beta 38:13 version is the best way to put that that 38:14 out today where you can sort of at a 38:16 high level create workflows from a 38:18 prompt and actually drill into that and 38:20 that is actively part of our our roadmap 38:22 to improve that further. 38:28 give a you know a minute or two or two 38:30 for any other questions. Um I've really 38:34 appreciated your active participation 38:36 everyone and sort of walking through 38:38 this especially while we had technical 38:39 difficulty there. Um but I wanted to 38:42 again highlight we have this build 38:44 versus buy report uh QR code you can 38:48 find it here as well as we will share 38:50 out the link.
