Name: Build a Prototype from a Customer Insight in 30 Minutes
Uploaded: 2026-04-08T18:16:06.087Z
Duration: 40 min 36 s
Description: Build a Prototype from a Customer Insight in 30 Minutes

Transcript for "Build a Prototype from a Customer Insight in 30 Minutes": Okay. Hello, folks. How are y'all doing? Love it. In The UK, Minneapolis, Chicagoland, Nashville. Damn. All over the place. Allison Cole. Hello hello, to Denver. I'm gonna give it a minute because I think a few more people are streaming in here. Carly, you keep me honest. I super excited about today. I'll be honest, I'm a little bit nervous because, one, there's a ton of people here. There's, there's almost 50 folks here and I think a 100 plus RSVP, which is awesome. Two, I'm doing a bunch of stuff live, and that's always a recipe for a good time. And particularly when we're dealing with things like, MCP, which can give you different results, and we're gonna be doing it on break question data as well, so there's a little bit of sensitivity to that as well. So, bear with me. A little bit of grace will be helpful. But let me share my screen. And here we go. Awesome. So today, my goal is in the next half an hour is go from how do we get a customer insight out of an existing system, in this case, we most likely, great question, and some other tools to help us build a prototype in thirty minutes, something tangible we can actually get in front of a customer or a user or a stakeholder to help to shape what this product might be with the customer inside of the center of it. For those that don't know, my name is Ned. I'm cofounder and CEO of Great Question. Great Question is an all in one UX research platform that makes it easy to recruit customers, either from your own panel or from we'll help you recruit via our 6,000,000 candidates out in the wild with all the research methods, interviews, surveys, prototype tests, card sorting tree tests, you name it, prototype testing, and a research repository. The whole vision is that by having a single place for people to go for research, it makes it so much easier for you for anyone to run research, whether you're a researcher, a designer, a marketer, a product manager, you name it, we make it super easy. And so that's why we work with folks like Amazon, Asana Brex, Wealthsimple, all sorts of wonderful folks. So, I'm gonna talk a little bit about the problem, about why you even need to bring these things in together, what MCP actually is. I'm gonna show you how to set that up. I've already got mine set up, but I can show you roughly how to go and do it. I'm gonna give you plenty of docs after this, so there'll be no shortage of that. Oh, hi, Jen. Jennifer, good to see you. And that but I wanna spend most of the time actually showing you how I'm actually gonna prompt the system to pull in the data to go and generate the prototype because we don't have a lot of time. Thankfully, I've also got some prebuilt prototypes just in case, but I'm hoping that I need to get to that. Let's see. And then we'll leave a bit of time for Q and A in the back end. I've already introduced myself. So the problem is that customer insights are trapped in all these different silos, and this isn't a new problem. When we started Great Question, the pain point I was experiencing was that, you know, any insight or any any research that had been run was stuck in people's email inboxes. But I'd also have to go and talk to the support team, the data analysis team, and all these other teams to go and bring this data together. And that that problem hasn't really gone away, and, but I think MCP is largely starting to solve this. Also sales calls, we use Fathom, you could use Gong. Project management tools, we use Linear, you could use Asana, Jira, and then customer feedback, which comes in so many different places, whether it's coming in via support or Twitter or in Slack channels, you name it, it's all over the place. Each one of these pieces holds a part of the picture, holds some context. None of them talk to each other, but when we can bring them all together, we get a really well rounded view about what the problem space is, and then we can start doing interesting things like building PRDs, which we built a little bit of today, or, Hey, just show me what enterprise customers care about pricing, or what people have been complaining about around this area of the product that may be just individual contributors. I don't wanna hear it from admins, and it starts getting super fun. The technology that we're gonna be using today is MCP, model context protocol. It's basically the simplest way to think about it, it's just an API. It's just an API that the AI can go and communicate with. They connect to these tools that have got MCP, like Great Question has, but a lot of tools, most tools, not certainly not every research tool, but most tools have a way to go and connect it with your Claude or your OpenAI or, Perplexity. You can then query it. You can ask questions across this data source. And every one of these tools has has what we call tool calls. So they've got little programs that they can go and run to say, hey. Find me all customers that have complained about this feature, or, tell me give me a summary of all research that was run-in the last month. And then from there, we can use that data to go and take that context to help us build something, whether that's building a PRD, a prototype, a business case, finding user evidence, whatever it is. Oh my gosh. Can anyone else not see slides? Okay. Michael can see it. Okay. Thank thank, Josh. Carly, can you maybe help? Jason is, like, not seeing the slides right now. It might be a bit of a problem. Thank god. Okay. Now, it's super easy to go and connect this, and I'm going to show you my Claude, which is let me reshare my screen. Hopefully, this will fix it for, for you as well, Jason. So here's my cloud here. Everyone can see that. I come over here and I go to my settings. I can then come down here to connectors, and connectors is where you need you need to add it. Now, if you've got a personal account, this is pretty straightforward. If you're in an organization, different organizations have different requirements, and sometimes the organization needs to go and add the MCPP first before you're allowed to add it. Your mileage may vary. Thankfully, there's tons of support docs out there. Two, you can go and ask Claude, hey, how do I set this thing up? And it will guide you through that setup. And so here I can come in and I can go and add, browse all the available connectors, and there's tons there. There are just so many. And you can also go and add custom connectors when you want. And you can see that I've got a bunch here that are connected, a bunch that are disconnected. I can configure them. There's a bunch of Great Question ones, all sorts of things that are connected here, which is super easy. And so what you do, basically, when you're connecting Great Question, is you would come here, you come to settings, you would, add a custom connector. Oh, sorry. I need to enter this settings. You go to the custom connector. You'll get this URL, from Great Question. Add your API key and connect your research repository and auth it, and now you can go and add all your other tools. Now forgive me because I've got so many windows open here. I've gotta find the right one. But all this information is available on our website. And I'm gonna share my screen again. You know what? I'm probably just gonna share full screen and YOLO this, because it's gonna be the easiest way to do it. Cool. And so here we can see, the Great Question support docs. Okay. Now, let's go and start, doing our live build, and this is the very nerve wracking part for me, where I'm gonna go to new session. And I'm just gonna go and find my, dog screen that is over here. Okay. And so the first thing we wanna do is, I'm trying to identify I'm connected to the Great Question MCP already, and I'll we'll send the support docs that you can go and get connected there as well. And, I've got a prompt here that I'm gonna put in, and I could kinda type in anything. In this case, what I what I plan to do today is identify an area of our product that we might want to go and improve. And first, pull in data from our different data sources, like Great Question, a research repository, like Mixpanel, Alex tool, like Linear, like Pylon we use for support, pull this data together, tell me figure out what should we even improve about this? What do people have a problem with? What maybe the top one, two, or three features we might go and pull from? So I can just talk to it to tell it that, or I can type a prompt. Now, I've gone and prebuilt a prompt here that I know is gonna work, and I'm gonna paste it in here. Analyze our highlight reel feature using our MCP tools. Great question. Pylon, Slack. I'll pull out Slack for now. Mixpanel. Deliver findings in a webinar ready format. Frame issues as user friction opportunities, not bugs or broken features, use the voice of a product leader sharing insights with an audience of researchers and product builders, keep it honest but constructive, emphasize what the data tells about our user needs and where the biggest leverage points are. Now, this is a pretty big prompt. You don't need to have written something this big. It could have just been like, hey, tell us what do we know about highlight reels, and we'll probably figure it out. But I ran this a couple of times, and I felt like this was a more constructive prompt for me to go and run. Now, note, I'm doing this in clogged code. I could have done this in chat. I probably would have a similar sort of result. I personally just prefer clogged code. It's got some, like, edge cases, which I find to be easier, and it's now gonna go and run this query. It's exploring, the GQ repo reels and highlights using our agent here. It's got a bunch of different tool calls that it's pulling, different ways that it's looking at the data. It's probably looking at what studies have been run. It's probably looking at highlight reels that have been created, full calls. It's now analyzing pylons. Some of the thing, it's only got two two calls here, which is probably fine. And each of these calls is probably like a query, and then it's analyzing mixed panel as well. So this is probably gonna take about a minute, which is relatively straightforward. And I I could have laid laid on also maybe to start with the Great Question MCP and then add in pile on a mixed panel, but I did it all as this one sort of mega prompt. Often when I'm doing this, I will layer the prompts because it's a bit like a house of cards. If you stack too many things, you're like, go and write this PRD and produce a prototype and create linear tickets and do all this other stuff. It's gonna get bogged down. It's gonna take a really long time. And if it gets any one of those steps wrong, going back is gonna really suck. Now if you have any questions, please throw them in the chat as we go. If you have a question, someone else will have a question. I know this is probably or if the screen's not clear enough or any of that sort of stuff, please let me know. Could you share about any context in the local folder that Cloud Code is working with? Like, is the code base or analytic data in that products folder? Yeah. So I actually don't have, I do have a bunch of stuff in my local folder that it could consume from, but I'm not asking it to do that at this stage. Really, right now, I'm saying, go and find everything in the MCP, And then I do have a skill set up which will automatically save this stuff to my local folder, and I ask it to regularly save. Like, hey. Don't forget. Save this stuff to local folder, because it's kinda like a safe point in a game. If it doesn't do that, it'll just store in its local memory, and then if I leave it or close it or restart it, it'll lose that. And then it has to go and generate it again, and sometimes it'll generate a different result, and that's not great. And so I have to go back and start it again. So, yeah, that's kinda how I think about that. Jen, I haven't actually run this in in Copilot before. I haven't used a ton of Microsoft tooling. I'll be honest. I should go and try it. But, I'm mostly using Claude these days. I've actually I barely even use, OpenAI anymore. I mostly use it for image generation, to be honest. I just find it's I don't know. It's not great. Okay. And so now it's gone and said, hey. Look. It's gone and written a plan before it starts coding, but most of the most of the information is contained here. Right? It's found three and a half thousand sessions across the Great Question MCP. That's just our our our own repository. Heavy internal test content, so there's a ton of test data, obviously. Primary use cases, this is a lot for our own internal stuff. It's got a ton of stuff with Mixpanel, how many people have created reels, got a ton of support documents, and then it's got a deliverable that it's gonna create. I'm gonna say, great. Let's go and produce this. So it's gonna do a little bit more than just present that plan or go into a little bit more detail in it, but let's go and see what it generates. And I've just lost that, but I'll come back to it. Now, I probably also wanna go and give it some context on the current experience. So let me do that while we wait, and I'm gonna go to my I saved some screenshots from earlier. I'm gonna say, here's some screenshots from the current experience as inspiration. Now, I could ask it to go and, like, go and take screenshots of the current thing. It would spin up a web browser and do that, but that's gonna be annoying, and so this is kind of the fastest way to do it. It's only letting me do five images at a time, so that's why I'm doing this. So now it's adding this in. So I can keep typing as I go, and give it more context, it'll tell it to the new things, and it will just queue those up and get to them when it's ready. That's asking me a question. Allow Claude to write hard real, write this, markdown file, and it wants to write it on my local machine, so that's why it's asking me. I could allow it once. I could deny it. I could always allow for the session. I'm gonna always allow it. I trust it. It's not through anything crazy. It's not like it's just writing a file. It's not trying to read my inbox or read some sensitive folders or things like that. So you do need to keep an eye on those things, but generally, I've found it to be pretty respectful. Okay. I think I saw another question on the chat. Let me come back to this. You enabled connectors, but did you also disable some things like web search, or did you rely on your prompt to nudge the agent away from searching the open web, asking is your prompt wants the agent to only draw from user feedback? Yeah, so, I didn't I didn't necessarily tell it not to search the web. It would ask me. It would prompt me and say, hey. Is this okay to to prompt to to do that? And I've said no at this stage. But you bring up an interesting point, which is we could probably do some competitive intelligence at this point. Right? And so I might say, great start. Let's learn some competitive intel. Tell me who out of research tools has a great highlight reel creator. While you're at it, are there any good libraries for generating a highlight reel which could, fast track development? Once we're ready. I heard about a tool called, I think it's called Remotion or Remotive. So I'll do that now. So one from Michael Hunt. Do you prioritize inputs like customer feedback, usage data, hypotheses when building a prototype, and how do you decide it's ready to test? Look. I I personally don't have a good rule of thumb of, like, okay. How much to weight this versus that? And so I haven't done any of that, but you could definitely do that. How do I know when it's ready to build a prototype? I'll review the markdown file, and we can do that now. And let's go and see what it says because it's produced it for us, while it's doing this competitor analysis in the background. Okay. Here it is. And let me make this a little bit easier to read. Behind this as well. Cool. The hot air real. What our data tells us about research visualization. And so this is this is where I say warts and all because, like, this is an area of the product that we know we need to work on. In fact, we're actively working on it, which is why I bring it up as an example, where it's brought in our product analytics, it's brought in customer support data, it's brought in data from our own research. It's saying, look, there's a lot of sessions being created, a lot of highlights being created, not a lot of reels being created. Some foreshadowing, there are some AI highlight reels that are being generated, but mostly it's manual, which is the traditional way you would do it. And so there's a bunch of momentum where it drops off. Edge ended reels have got an inflection point since March, where users hit friction. I made a reel, but it won't render. Okay. Some saving problems. I can't fix a clip without starting over. Yes. You can't edit a clip within the context of a highlight reel. I have to go back and edit the highlight. It's very annoying. Clip doesn't match with a highlight in a transcript. We we currently allow you to create a clip based on words, like you select words versus which makes it a problem if someone's not talking. It's hard to select the gap where they're not talking. Some download buttons. This is somewhat to do with permissions. And I can't search across interviews to find what I need. A new global AI, ALSKI sort of solved that. Cool. So it's like double down on ad generated reels that's clearly working for us, make the reel rendering pipeline bulletproof, make it easy to save, and things like that, solve the last mile, get clips into the tools stakeholders already use. So get into Slack previews, into slide decks, and email friendly formats. And so this is a fine PRD or like, you know, summary of what we've learned. I wouldn't say it's the best, but it's not a bad place to start. Now what I might do is say, how can I make this better? Yeah. Let's start with that. How can I make this PID better or make it more like a PID I might produce a prototype with? Now, you'll notice that we've got nine minutes to go, which is not a long time, but I'm confident that we're gonna get to a good result. Keep the questions coming. Is this useful? Am I going into too much depth? Am I going too fast? Is this all irrelevant? I welcome feedback in the comments as we go, and we can keep this interactive. Okay. So it's thinking about how we might make the PRD better. And I'm typing here. I'm actually kind of a lazy like, getting somewhat lazy with typing. I will often I've got a tool called Whisperflow that lets me just talk to it. And so maybe I'll use that for the next time I put a prompt in there. And so mostly, I'm just talking to it. I'm like, I'll even walk around the block sometimes with my phone and just be like, okay. I need to figure out how we're gonna rebuild highlight reels. How are we gonna do this? Okay. Let's go and look at our, you know, who are the customers that use it the most? Let's go and understand what they're doing and maybe help me put together a research plan and things like that, some more foreshadowing, with, with stuff that's coming up. So give it some patience while it's doing this. I probably gave it too much stuff to do in this call because it's doing 26 tool calls through some competitor research, and it's looking at these video libraries, which is definitely, like, something that we should do later versus now, if my goal is to go and generate a highlight reel. But it's gonna go and produce this plan, a new plan. Okay. I'm just gonna YOLO this because we're we're running a little bit short on time. Once you're finished with this PRD, can you please produce prototype? Example, prototype for me, in HTML. I don't need to tell it this, but I'm trying to avoid spending a lot of time building a React app or something more complicated. So my HTML CSS, and, make each, page tab at the top so I can click through to show the team what I've created. Okay. Now rewrite the main so it's reviewed all three docs. It's put together a bunch of additional docs. It's it's reviewed all three docs, now rewrite the main document as a PRD, you could hand to a designer engineer to produce that from pulling in the competitive intel and technical context. Great. So I'll put that in there. Now let's go and see what it went and produced for me. Great. So here's a competitive analysis that it's done. It's looked at Dovetail, it's got some AI features, some standout, where it falls short, some pricing stuff, and it's got some links to different docs I could go and look at. It's got some stuff around user testing, where it falls short, heavy enterprise pricing, obviously pricing related than the the thing. Redux, which I've heard really great things about. I'm glad we're pulling inspiration from there. Saying it's best text based editing, that's interesting. Is it best editing UX, and timeline based? Okay. But also strong for research. We want fine grained controls, condensed. So, anyways, it's got a bunch of interesting stuff here that it's gonna pull together, and then it's also gone and looked at adjacent tools like Grain and things like this. This would've taken me forever. Now, has it done the best job? Probably not. Has it done, like, 80 or 90% of it? Probably. And I could go and look at it in more detail if I had more time. It's even gone and produced a table to help me figure out where are we gonna compete and what are AI auto highlight reels of the new frontier, text based editing is UX Advantage, every notice of clip is an emerging pattern, Sharing is critical and often weak. Great. It's also gone and done some research into tools I could go and use, including Remotion, and how this might work. Now I'm a product manager. I'm not super technical. This would be something I would share with my team, but, really, engineers are probably gonna make these decisions and not me. But I at least have done some initial research to inform them the things that they might not have been thinking about as they look to go and figure out how they're gonna technically implement implement this. Cool. So I feel like we've got some good stuff. What screen should the prototype cover? I'm thinking these moments are the PID. So it's giving me some options of what I might choose. Before and after, end to end journey, creation focus. Let's do a full flow. We've got four minutes. Let's see. Fingers crossed this comes together. If not, in about two minutes, I will, I'll cheat a little bit and show you what I what I prepared earlier. Can we take a step back? I'm a little confused on Great Question as a whole and how it plays into this MCP, invalidation put to combat hallucinations. Great. Yeah. Let's take a step back while it works on this. Okay? It's gone and put a plan together, for these things. I'm not gonna review it. Let's just see how it goes. Normally, I would review it, but I'm not right now. So how great question works. So we're an all in one tool. We make it easy to run all these different research methods. We have a research repository, and we can go and, recruit from a panel of your own customers. Here's a study that I set up earlier, which has all the things we would set up in a traditional study from the plan and the screener. In this case, it's for interviews, so set up a calendar. At the end, I have a repository to see what it is that I've gone and learned from the team. I can click into one of these interviews and, play it back. I can view it in full screen, to play the session back, and AI has gone and chapterized it. From here, I can go and clip play a playback a clip from this interview. I can also go and query this with AI to ask it something like, what sucks about expense management? This is a demo with an expense management tool within great question, and so it's gonna come back and give me a summary of these pain points. And it's gonna cite who the person was that experienced it, and, what they what they said about this particular thing. And if I click on this, it's gonna come take me to that moment in that interview, it's gonna play back that session live, and then I could I could create a clip or a highlight reel from this. The reason I show you is that this is the workflow that I think people, have been using in the product where I can then go and select this clip, and I could go and add it to a reel of credit earlier. I could then go and view that highlight reel, and here it is. It's re re rendering because it's, I've gone and added a new clip to it. Let me see it refresh. It should render. And you can see the different clips that are here. Now, I can then go and share this with my team, I can edit it, I can do a bunch of things. It's gonna it's gonna drive me crazy. Live demo with this saving. Now, how do we are there guardrails and validation in place to combat hallucinations? Within great question, yes. Not within all tools. I find that hallucination is somewhat overblown. It does happen. You do need to be reading your PRDs and documents and things like that to verify that it's not. But I don't think it's as big of an issue as it was maybe a year ago or two years ago, even three months ago. I feel like the technology's developed a lot. The great question what we do to help prevent hallucination is when we cite anything, when I do this, like, AI querying, we will give you this the themes we found. We'll then give you the quotes. We'll then cross reference that the quote actually exists in the transcript so that it's not making quotes up to reference and validate this theme. It's not perfect, but it's one way that we're we're we're we're found to to remove and eliminate, hallucinations, which I think and that's very rare that it comes up in our tool, but it but it can still happen. So hopefully that kind of answered your question. Well, it looks like there's a reply that someone replied to it. No. I can't see it. Okay. So now it's producing this prototype, and we're at thirty minutes. So I'm gonna, cut and cheat a little bit and show you one that I prepared earlier. So you've seen how we do it today. This is one that I prepared earlier, and this is probably roughly what this thing will produce now for us anyway. And so it's got the study set up as we as we have. It's then got an AI synthesis tab. I don't know if that's the right language, but that's fine. I can now go and edit it. I've got the ability to go and generate them in different ways. I can generate one based on, this theme of approval process frustrations on receipt management pain, or policy confusion, or tool fragmentation. I could go and enter a prompt if I wanted to go and say, hey, show me a highlight reel of only enterprise customers talking about this, and it would then go and generate a highlight reel from that. It then is showing me the interstitial, you know, generation animation that it's gone and created for me, which is kinda cool, as a way to, kind of buy time and tell you what it's actually doing. I've then got an improved reel builder. Similar sort of interface, I guess, it completely transformed it, but it's showing me that, you know, how I can view the different clips, and it's still got the reorder functionality, which, to be honest, I would probably add my own context, which I don't particularly like this reorder functionality, but they each their own. And then I've got a real viewer where I can go and see it's chapterized, it's telling me who's on screen. It's got some better summary points about this. Now has it gone and fixed every problem in this thing? No. And I'll wait for this to come and generate. But it's done something not too bad. Now here's a version that I did as well that was in, that it built in React. This probably took, you know, this took maybe three or four minutes to generate this HTML version. This maybe took five or six minutes to generate something in React, which is then a bit more interactive. I can click in here, I can type something, It can walk me through what that experience does. It's not actually gonna generate the clip, but I can get a sense of what it looks like. When I click on this, it's gonna show me, you know, a preview of what that might look like. It's got some drop downs. It's got the ability to reorder these things. I've got a highlight reel selector to go and add additional ones. I can search for past ones. I can save this. I can see what the edit and save states are. Hasn't built out everything, but it's a pretty good start. And I'll do one more check before we go because I I would love it for it to actually produce a version. Okay. It's asking me to approve this. Oh, it's trying to start a server. Okay. So we we can see it. It's actually gone and generated this. And so I'm gonna bring it up. Moment of truth. Can someone give me a drum roll? Boom. This is what it's generated. So a different interface, it's sometimes it's gonna do different things. What just happened? Oh, it's now spun up the server for me as well. And so I can say, generate reel from a theme. I can go and see what each of these pain points are. It's got it's showing me the clips on the right hand side. It's got a slightly different ordering mechanism, allows you to still have the clip while you're playing below the fold. It's got captions, which is great. It's automatically tagged it. It's got an instant preview generation. I then go and share it. I can see it's viewed and engaged on the right hand side. March. Nice for what we've already got. And then I've got a bunch of different share and export options, including downloads and embeds and things like that, and shows what a Slack preview is. This is pretty nice. I think if I show this to customers, and we probably will, then we would, we'd be in a not bad place. What do we think? Is that cool? Have I lost everyone? Cool. So what Claude did, it went and accessed oh, Sasha. Access Great Question research repository and a bunch of other tools that have got important context. Addison, Ethan, hey. Patia, it's got that it's brought in together to go and inform what it needs to go and build. And you saw how little effort I put in. Imagine if I put a smidge more effort, how much additional context I could give it, how much better an output I could produce on the back end of this. You saw I was rushing through this. We asked it to go and synthesize the results, cross reference a bunch of other data, do some ideas. I could have probably said, give me five variants of this. I'm not sure what I wanna do yet. Give me some and rank rank them by evidence strength. That would've been a great thing. And so we were able to then go and produce a working prototype in roughly thirty minutes. I think I downspanked a little bit at the start, so, we got there in the end. But I would say we barely scratched the surface because after this, what I could do is I could go and say, generate linear tickets or go and add this evidence I found on existing linear tickets so that people know that it's not crap. Draft a better PRD for me. Now that you've got the actual end prototype, actually go back and revise what the PRD might look like, taking some of this feedback in mind. I could have theoretically gone in some organizations, you can go and push this feature to Claude, submit a PR, and get an engineer to review. Your mileage may vary. I can validate it. I can go and say, hey. Use MCP. Go and find users that have used highlight reels in the last three months that have created a clip but not created a highlight reel and put them in my CRM so I can run research with them. I could set up an intercept to target them. When people have saved a highlight reel, so they've finished their their process, do an intercept, and then run a prototype test with this new prototype with them using great question. I could deepen the intelligence. So I did some competitor intel, but I could probably go much further and ask it to take screenshots or go and sign up and use these tools. I could get it to go and find people to run research with. I could cross reference sales calls. What are people asking about or complaining about alternate tools? I could do some job to be done analysis. I could socialize it. So I could get it to go and draft a Slack message to share out this prototype and the evidence in Slack, and it will actually render Slack and allow you to go and hit submit and push it into Slack for you. I could generate a better a highlight reel from this. Like, get all the evidence of people complaining about it from my research that's just gonna make it much more visceral to the powers that be to get them to prioritize building this feature. I could then continue to go and explore technology. I could look at things like Loom and Notion and Figma and see how they do this. I could estimate the impact. What's the revenue impact or the sales impact or likely impact gonna be when we actually make Hirel significantly better? And then I could make it repeatable. Maybe I could say this is a skill. Anytime I'm like, alright. I wanna explore x feature, it goes and runs that entire process and pulls a PRD for me, produces a prototype without me having to go and do that work, which is largely the same anytime for any feature that I'm gonna be working on. I could also get it to set up an alert for new data. Anytime someone else in the future complains about highlight reels, tells us some feedback, or a competitor launches a change to highlight reels, tell me about it because I wanna go and use that to go and shape it and go and ship it in the PRD. There's a lot you can do. So great question. Our MCP roadmap our MCP went live, we've started using October, but it went live in, about, this week. We launched it this week. The uptake has been incredible. We're gonna be adding more integrations. We're gonna be more proactively servicing insights, so that idea of having an alert around how that reels, we can do the same thing for you. We're gonna add in more team workflows, and we're gonna be adding a write MCP. The write MCP is the really exciting things, because not only does it let you write to your repo with other data you might find from other places, which can be useful for right reasons, but we're also gonna allow you to set up a study from within your cloud environment. So this would allow us to do the workflow that I described where you say, find me people that have used this feature from Mixpanel or whatever your analytics tool is, that could be a good candidate for this, or publish an intercept to this particular page when someone hits save, to recruit people and set up the prototype test so that I can go and get instant feedback on this particular area of the product or this prototype I've gone and built. So you can see how this whole stack is collapsing, to allow anyone to be a a product builder end to end or at least decide which areas of the stack that you wanna go and operate in. Super exciting. So everyone that's on this call, you're gonna get access to your MCP. If you're an enterprise customer, if you don't have it, please let me know. Email me, net@greatquestion.co. Happy to provide you access. We'll be sending out some of these prompts. They're not super complicated, you've seen, but I've got, some, like, bells and whistles in there that I'll send, a setup guide on how to set up great questions MCP or really any MCP, and a recording that you might wanna share this with your team if it's relevant. Anyway, that was a lot. Thanks, everyone, for sticking around. Feel free to drop if you've got something else. If you've got any questions, please bring them put them in the chat. Sydney, we have policies to prevent us from sharing customer data with agents. Is this a tool I could advocate for when our legal team questions it? How are we keeping the data separate? Great question. So so, the way that we do it, a Great Question, is we strip out any PII before it goes anywhere near the models. So last names, email addresses, social security numbers, you name it. It's all stripped out before we share it. We put a token in place, and so it gets shared with the model, and then when it comes back, we refresh it. That's one way we've done it done it. Second, we don't use the data for cross training, and all our agreements with our third party providers like Claude and like Anthropic and OpenAI such that they're not allowed to use the data for cross training purposes. So I think this is I think most legal and security teams are evolving in their thinking. I think even, you know, definitely two years ago, it was challenging to to get this stuff approved. It's significantly improved for most. They just wanna know that you've got the right guardrails in place. We have SOC two compliant, HIPAA compliant, GDPR, and we love having those conversations, and we've never had a case where we haven't been able to get it approved. Whether it's at all, I'm not happy to answer any questions we've had for that, but I'm confident you should be able to get AI tooling in place because otherwise your company's gonna go be behind, and no company wants that. No CEO wants that. How are we keeping the data separate? So I think I kinda answered that. Emily drop. Hey, Emily. Good to see you. Awesome. Awesome. Anyone else got a question? Cool. Alright. Well, I I think with that, unless there's any last minute questions, I'm gonna wrap it up. Just wanna say thanks to everyone for coming. I would love your feedback. We're gonna send out a brief survey after this. If you found this useful, reply there or send me an email, Great Question or co. If there's more stuff like this that you would like to see, let me know. I love talking about this stuff. If you've got something specific on yours, like, hey. This is not gonna work for us because x, y, and zed. Like, how do we think about this? Love, love, love it. We're going through this big AI transformation. Everyone has to go through it. And the the more I think we can share, the faster we're all gonna get through it to the role, working in AI speed, which is super important.,