Vibe Coding with Noca AI: From Prompt to Production

0:00I see AI as a bridge between the business and the production. Everything like the requirements come from the business side of the business. And they're like asking, hey, I need you to do something. And then they go to the IT, IT go to developers, and then it's like a process. Now you can say like, this is what I need. Type to the chat. Boom, it got something read. Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Page, and we're here to demystify AI for you so you can get some value from it. Let's talk some AI.

0:35Most AI automation tools give you seemingly infinite power and at the same time, zero guidance. You land in Zapier, make, in-ate-in, name your tool. You start at this blank canvas and think, now what? But with generative AI, we're entering this new phase where automation can be dynamic, conversational, even intuitive in a sense. And today I'm joined by Avishay and Benny, co-founders of Nook AI. They're building something really cool. at the intersection of automation and AI coding, which has gotten this infamous name, infamous or cool, depending on your take, vibe coding, essentially stating what you want in natural language and you kind of let AI do the rest.

1:16We're talking about what's broken with automation today, how to fix the blank canvas problem and why the future of building might feel more like a conversation than an endless coding and configuration session. Plus we got a couple of cool live demos we're going to get into. Welcome to the show, guys. Thanks for having us, Mass. Happy to join. Yes, this is going to be good. And you two are building Nook AI, which we'll get into some short demos later. But why do you think we're seeing such a surge at the intersection of automation with Gen AI hitting the scene? Because automation's been around for a while, right?

1:52You've had tools like Zapier and things like that. But it feels different ever since like AI, generative AI has hit the scene. And if we go back, I'd say five or seven years, then came the good no code tools. Your makers and tools, you mentioned the MUX. You'd still need to have some technical or a lot of technical knowledge of how to build the automation. Gen. AI bridges a lot of this gap. You can just speak to the system and the system will build the integration and the automation for you. yeah and i've seen people doing this where they'll like it's like they screenshot an automation and then they that they've seen from somebody else like an n8n or whatever and they're like hey ai build this for me and it can do it so it's like the two sides of it in my mind there's the one you mentioned where it can help you in the building process but it can also be used in the actual workflow and that's the big thing in my mind i talk a lot about going from deterministic to probabilistic and then what that unlocks because you don't have to define every edge case of every workflow which can be painful and limiting in a sense i think that's been the big gap of why automation has not gone as far up until this point as it has and now it's exploding in a sense sure yeah i'd be curious benny your take on that like the yeah nature and things but basically I see AI as a bridge, basically.

3:22Bridge between the business to the production. Usually, everything, the requirements come from the business side of the business. And they're asking, hey, I need you to do something. And then they go to the IT. IT go to developers. And then it's like a process. Now, you can say a business guy, I say, okay, this is what I need. Type to the chat. Boom. It got something ready. This is number one. Number two, like if something is failing in the process or something is maybe not built correctly, the AI can say, look, you're trying to create an accounting Salesforce, but there is a mandatory field which you forgot.

3:59Just like a simple example. You want me to help you fill it in? Oh, good. So it needs to be like a guy. I think the main thing with the AI is to shorten the delivery quite a lot. You don't need a specialist as much as it was like before. and there's this element to evishay you're hitting on the democratization element because now it's not just the engineers that can build things it's anybody like and i feel like this is the most polarizing thing because some people think that's like the best thing since sliced bread and then you have folks on the other side that are like oh my god there's ai slop everywhere this is horrible it's going to lead to the downfall of everything but evishay what's what do you think?

4:42Do you think it's a good thing, bad thing, somewhere in between? What's the, your take on the democracy? I think it's an excellent thing, but it's no excuse to be sloppy. You can, you can use AI to build automation and you should, but you also need to review the work that's being done. One of the things, you also build AI agents, we can touch on that also later, but one of the things we strongly believe in is the importance of having human in the loop. Whenever you're using AI agent to take action. There are certain points where you should introduce human in the loop. A person, review the data, make sure it's what you want.

5:21Don't just let it fly and hope for the best because AI is excellent, but you need to use it smartly and cautiously. Quick break in the pod. If you're listening to this podcast, chances are you've been thinking about how to actually use AI inside your business. And that's exactly why we built the AI Opportunity Finder. It's a free tool that helps you uncover high impact, tailored AI use cases based on your business, your goals, your pain points, and your industry. No fluff, no generic use cases, just real ideas that fit your business and the ranked by ROI potential. It takes about three minutes to run and it's like having your own personal AI strategist for free.

6:00If you want to try it for free, check out the link in the show notes or go to hatchworks.com backslash AI dash opportunity dash finder. And then there's almost this, you hit it right on the head. There's like this approach people take where it's almost like lazy AI where they're like, their prompts consist of fix this. Why do you keep, why do you suck at your job? It's like, you got to give the context, the detail. And that's the part that I think a lot of non-developers, they struggle with because A lot of them don't understand, you know, how databases work, how systems work. There's a system thinking approach to it.

6:36And I would argue that it's an easier skill to learn than people realize because you don't have to get down to the syntax level of things. Like just having a basic understanding of how your database is structured. And there's great tools for this. I'm sure Noka has some stuff in there too, but like Supabase where you can visualize your database. That's hugely impactful. And you just having a screenshot of that and adding that context to your conversation is powerful. And then I don't know if y 'all done some of this, but like when I'm like vibe coding with something, I run into a bug. Sometimes I'll pull in a different tool.

7:12If I'm in lovable, I may go ask ChatGPT or use Codex or QuadCode and say, hey, what's your take on this? And they'll take it back and say, hey, this is what ChatGPT thought. And they'll be like, ChatGPT is exactly right. So it's, oh my God, I got like these multiple like AI engineers, in a sense, working on this problem for me. Yeah, I think I completely agree with that approach. And I think we need to have somewhat different systems than we have had in the past for going live in production with AI automations. So I'd like to give you an example of a customer of ours who wanted to automate a process where they process 10 ,000 invoices per month.

7:51So they, the process was that they would upload the invoice. Someone would type the information. A second person would check the information, look at each invoice and then move it, approve it for payment. Completely manual. Yeah. All humans. And then they introduced NoCal. They started using our platform for document processing. and essentially the second human would check nine items. So it's a data main voice, a mouth on the invoice, who's the customer, who's the supplier, nine items. And this was the first big process, like core process they introduced AI to. And the way they went live, I think, is very smart.

8:36So the first thing they did was build the process and then they spent about a month checking every invoice and checking whether they approved the invoice and the AI approved the invoice, then it was a good outcome. And if they approved the invoice and the AI didn't, each false positive and false negative, they checked and they looked into. And during that month, they made iterations to the process to make sure the AI gets it right. And then they went live when they felt comfortable with the answer that they were getting from the AI. But even then, they didn't take their hands off. the process, they move to checking the AI every five invoices, every 10 invoices, every 15 invoices.

9:19You need to gain this call today before you can let go completely. Yeah, I think that's a critical point. There's still these first principle things around testing in QA and whether it's a human or AI doing it, it's important to do that before you just push something out into the wild. Like that, that heuristic is still true in a sense. But you hit on something though. I think that's a big gap. When people look to automate a process, I'd be curious, y 'all's thoughts, but I feel like a lot of times people jump to the tool too quickly. Yeah. What is the approach to automating a process? What should, let's pretend like I want to automate something in my business.

9:58Where do I even start? How do I identify what to automate? You should start with manual approach. Do it manually. Try like a few different options. And then you can go to an AI tool and ask like, this is what I'm doing. Like, give me a feedback. Maybe you can find like a better idea. And then you can try and do like A, B test. Maybe you should do multiple approaches, see what's working better. And then like slowly make it work like fully automated. Yeah. I like that you too, essentially in the process, you're talking to AI to see what were opportunities there are. I think that's half the battle too, is people don't really understand what they do on a daily basis.

10:35And it's like shocking as that may seem, but like we just work. we work and we're like little mini robots right but if you take the time to actually document with somebody i think it may have even been alex hermosi or somebody but they said take a week and just document every 15 minutes what did you do and then the manual version of that is okay let me look through all my notes and see what patterns there are what you just said that benny i'm thinking just shoot that stuff in the chat gpt and say hey where are the patterns what should i automate, like, I think that's a good starting point as well.

11:10Cause you gotta identify what's right for automation. I think there's another point worth mentioning is we're seeing a lot of AI agencies, a lot of automation agencies. And sometimes it helps having someone from the outside, look at the process with the people who are actually doing it because sometimes the people, the business people will introduce the process, discuss what they're doing. and then you can ask them, why aren't you doing this differently? And they'll have a short discussion and understand that there's a better way to do this. So I think people advising companies and agencies of automation and AI agencies have a big part in making these forces better.

11:55Yeah, that's a good point. Just having that outside perspective, because you're not jaded by how you do things today. You're not protective of your current workflows and processes in a sense, because people will get protective of it because they, it's their job in a sense. With the data privacy, people doesn't know something on how to use it properly and they can like share or expose data. Sometimes what's happened, you want to, maybe you want to run your process on a bedrock, like AWS dedicated service. Sometimes you want to do it like publicly. There's like a lot of things that you need to be aware of.

12:31Yeah. And if you're just building willy nilly, that's another thing on the democratization side, it's great. But then a lot of times people are building things and they don't realize that, oh man, I got some very sensitive stuff that's very easily accessible. And I have no idea of that too. And there's other, the other automation side I see, I don't know. I'm curious if y 'all see this. Sometimes people automate things that may maybe don't need to be automated or maybe the automation actually takes longer than just doing the thing. I've seen that before too, where it's look at this cool thing I've done.

13:04And I'm like, it's actually more efficient sometimes just to do the thing or like what value is it adding? I think that's the other component too, is make sure the automation that you're doing has some type of value in automating the workflow. Yeah, I think that's a great point. I think when you, in many automation, many customers come to us and they don't know where to start. They've seen them at all. They know they want to introduce automation, but they don't know what to build first. And we like to offer them, look at your processes, find the pain, find something that is a relatively simple process.

13:42But it happened many times. And then we can get a quick win. Once an organization understands they can in an hour or two automate the process that takes them a few hours a day and they can automate it once in a few hours and get rid of that process. I think those are the best place to start when you want to introduce automation into your business. Yeah, that's the flip side of it too. Like small win stack, I guess, in a sense too. If you're automating things to save you 30 minutes here, an hour there, that adds up over time across the organization. So there's still value there. So I'm curious, I got one more question for you.

14:19Then I want to jump into some of these demos and actually get a look at, but I feel like there's going to be this shift that happens around UX and UI because everything to date has been built for humans. Yeah. But as AI is now in the mix, like what happens to UX principles, UI? Do we need all of these interfaces that we have today that are very tailored for us as human being the conductor in a sense of clicking buttons and doing all these things or is it something completely different like where do you see that shifting to and is there a future where we just you know i don't know i know this may be a little extreme but ui just disappears completely and you're interacting with ai in a sense okay i don't think ui will disappear like completely maybe like 20 25 years i don't think now but i think like most of it would be like a chat So you will ask questions and then the AI will also pop like insights for you.

15:17Say, Hey, I see some based on your previous question, based on your previous interaction, I understand that this is what you need to do. At the end of the day, you will need reporting. You will need like charts. You need to sign documents. You will need like even to free some forms or like other like information that need to be structures in a way. You can always just do like everything in a free chat because then you can lose some of the data, but for sure it's going to reduce the Y works a lot. Yeah. But I guess AI in a sense could be the one filling out some of those forms and things like that where the user may not need it.

15:54But then again, like, you know, I'm still going to want to watch and you want to see it as well. Yeah. So the human in the loop aspect, and I'm still going to want to watch YouTube videos, so there's still some value for screens in the future, unless that just starts beaming into our head at some point down the road. Bishai, any perspective there before we jump into the demo? Yes, sure. A, I think things are changing quickly, very rapidly in the last two years, and more so in the last year. So it seems like it's accelerating. And what we are seeing currently is not that people don't want UI, don't need UI.

16:31They're looking for good UI and UX to introduce AI into the processes. requests we see often and i'm working in salesforce and i'm having trouble reaching the data i need i would like to introduce a tool that could help me reach the data so salesforce came out with the tool i believe many of most of the big platforms will have their own tool but users are looking for a chat like approach or feel in order to access different systems and so i think the The question currently is not, well, we need a UI for humans in the future. I think it's what sort of UI we need to offer humans in the present. Yeah.

17:17Yeah. And then also for agents as well, what is tailored for them in a sense? That's good. All right. If you should, let's jump into some demos here. What do you got for us in? Feel free to do any kind of setup in context for Noka. And because I think the way I understand it is you're this blending between your automation tools and your vibe coding and bringing this like nuance together. Yeah, I think we, and there are two, two parts of the systems. I'll show them both in the demo. And we come from the integration space that he's been working in integration and for the last 30 years. And then he's a domain expert.

17:56Yeah. I'm actually younger than, I'm actually younger than five years. He started when he was five. He's now 35. 30 years? He was a boy genius. 25 years. But we've come from the integration space and we have a somewhat different approach or an additional approach to Vibe Coding. We'll showcase that in a minute. And we'll read with a blend from between like enterprise automation tools and Vibe Coding tools. Okay. So let's see something. Or a few things. and I'll share my screen. Can you see my screen?

18:34Yes, we have. We got a note. It looks like, yeah, no cutout. Okay. And for those that are listening, we'll be very descriptive and talk through this. If you want to check this out, go pop over to YouTube and you can see this thing live and in action. Yeah, for sure. So, NOFA has two main sections, the Vive Coding, or Web App Builder, and the Slow Builder. So I'll start with the web app in the Vive coding space. And since it takes a few minutes to run, so I'll show you what I'm doing in the Vive code. And we'll go do some automation, like pro-based automation, and then we'll come back to see the outcome of what we built.

19:16Flip back and forth. And for those listening, what we got right now is a very similar kind of screen people are used to. It's this chat box like you'd have with chat GPT or a lovable or something like that to set the stage. So this becomes your entry point into building something in Noca. Exactly. So I'll give it a prompt. I wrote down, build me the attached landing page and add a form to it, which creates items in Monday on a board called Priority. And in the form, have the following fields, the company email and follow. And I'll then upload the screenshots I took from Monday earlier today. So if I can show you, basically what I want to do is I want to take a screenshot of a website and you can see it in Monday, like the task management tool, the popular task.

20:09But it can be any website. Can you see the picture I'm showing? No, we see like the browser. So maybe point it on a tab. And there's still. Maybe I can, I think you can drive it like as a tab. Let's see. Okay. Anyway, thanks for me. Yeah. So 25 years old. Yeah. You're very technical for a 25 year old. So I took a screenshot of a web page. It can be any web page. And I asked the Vibe project tool to replicate this, but I would also wanted to add a for that integrates with Monday. And I specified the next thing I do is... I think I want to say something about it like for a second. Basically what we do with Vibe coding today is pretty much nothing.

21:03Just generate an application or a code or maybe integrated with a very specific database. Like in real world, like enterprise, big size enterprises, they have a CRM, they have an ERP. They don't want just like the skeleton or like the layout. And then, okay, what's next? I need to take like an iPaaS product to connect it. I need to write the code to connect it. It's too much. So what we're trying to do like with Noca is say, okay, you don't need like multiple tools, just give me a prompt. We know how to build the integration for you. We know how to build the layout for you. And it's working. So this is like the main difference that we're trying to do.

21:43I think that's where vibe coding gets a bad name sometimes too. You see people building stuff and it's like, okay, you basically just built a pretty front end. What is it connecting to? And then to your point, in actual businesses, they want it to connect to the stuff they already got, which is where the data resides in a sense. Exactly. I completely agree. This one can take two to five minutes. Maybe we can go to the different demo. Okay. Um, so I'll move to the, um, I moved to the flow builder. So we're done with the web app. Now we're back into the flow builder, like, so the flow builder would, is it correct to say this is something you would analogize to an M8N or a make or something like that.

22:28Like the automation flow builder. We've never heard about those names. Can you repeat what the names made? Yeah, you've never heard of this, I'm sure. No, isn't something new, like fresh? I've never heard. Yeah, this is the automation integration automation section of the program. And so you can build with drag and drop like you do in other systems. But I think the interesting part is being able to build in English. So you can just type a prompt and the system will build integration for you. That's interesting. So you're basically connecting the vibe coding approach to building out automations in a sense, versus just manually building out these connections.

23:11Yeah. So when I ask people, how long does it take to create a customer sync between NextSuite and Salesforce? Usually I get answers between a few days before to a few weeks. And what we're seeing is that you can accomplish this in a few minutes. Oh, wow. So what I told those listening, yeah. Talk us through the prompt and what it's doing for those listening. This is pretty neat. Yeah. So I'll give a simple prompt. I told them when a new customer record is created in next weeks, check if there's an account in Salesforce with the same name. And if you find an account in Salesforce with the same name, send a Slack message to a Bichai already have in Salesforce.

23:58force if not i would like to create a new account in source force based on this account that was created in next week so the first thing the system does is creates the flow the structure of the flow but it doesn't stop there it goes even deeper and and it's it does the actual mapping and the the fields we can if we look step by step we can see that it understood the prompt it understood that this flow starts when our customer is trading in NetSuite. So we can see it shows the right record of customer, the right type of event. It then goes to Salesforce and reads to check if there is an account in Salesforce that has the same name as the company name that was created in NetSuite.

24:46It will then create a rule to check if it has results for the tested mail. If it does not find an account, it will send a message as requested to Avishai in Slack with a company in Netsuite. And if not, it will create a new account in Salesforce based on that account it was created in Netsuite. So as we see, the system took the prompt I gave it. It understood the structure and it also understood the mapping, the right object, the right fields. and next step is to connect the go ahead let me stop you right there and ask you a question too and just for those like listening it's cool to see he put in the prompt and you can see it visually creating the different nodes and things but are you already logged into your say net suite account or your slack account to where it has a sense for your instance and like slack for instance it has a sense for your channels and whatnot.

25:50Is that already, have you already connected the tools you want to connect? So when they, it goes through this flow, it already has a sense for your specific accounts that accurate? It's a good question. Basically like a fresh user, we know they have any connection because of which is like not new to the system. Everything is ready. But if it's a new user, you can say, Hey, I need to work with your, our next week. Please give me your authentication. Let's collect your next week. You would connect it. Once connected, then what we do, we take like all the metadata, like the objects and find the tables, blah, blah, blah.

26:25We take it into a vector database. So everything is like vectorized. So we can do all the LLM magics, like with it, like for the mapping, for the matching, we know which field it's, it's worked really nicely. And yes, if you will see the wizard, like from the right side, you will say, Hey, you need to connect to NetSuite, give me the connection. You need to connect to Slack, give me the connection. And then you go. Yeah. Yeah, thanks. Great question. So what we see users struggle with as well is going live with automation because you need to do some work in the backend. And we use AI for that as well.

27:01So once you've done this, you've created this slow with a prompt, all you need to do for this slow to go live is just to click a button. Once I click go live, this automation will be live, as I said. And if we go into NetSuite now and create a new customer, this is NetSuite. And let's create a new account. We'll call it Avishai. Avishai. And Met. And save it. Good old NetSuite. I feel like, Benny, this is back in your beginning of your career. Yeah. Mine for sure. I don't know about you guys. You guys young. Some of you young. And yeah, so next week takes a minute. It's still saving on the next week, but I fished that.

27:50Okay. So we've saved this one next week. We go back to now and you can see the information in here. And if we go into the actual flow, we'll be able to see that this is the account that was created, Avishai and Matt. It searched for a customer, an account in Salesforce, followed No, Avishai and Matt, and then created a new account in Salesforce. course, we had the account ID and we hired all the details. And basically what we did is create a, um, customer sync in a front in a minute. And we went live what used to take. And this, yeah, that's in that, that is insane. What's cool about this though, for those listening, like you almost have this, what you just went into is like a log of what happened, which I got to imagine is helpful if like you're trying to debug something or something goes awry this kind of gives you the sense of all the things that are happening and you can go deep into those exactly yeah let me take the word for a second let me go ahead let's run again like another record in in that street and then go fast like to noca before like it's finished saving let's see what's happening because you can see everything like in real time so you can see it's like actually working yeah That's great.