The Real Work of Coding with AI: Claude Code & Product Building with Ben Novak

What if coding is no longer about writing code, but about thinking, prompting, and building with AI? Today, we're talking about how tools like Quad Code are changing who can build. Super excited today to have Ben Novak. He is an AI product builder and search expert specializing in helping organizations transition from traditional enterprise search systems to modern AI-driven solutions powered by large language models. His work focuses on rethinking how information is structured, tagged, and retrieved, moving beyond keyword-based search towards more intelligent systems using embeddings, retrieval, augmented generation, RAG, and context-aware AI experiences. He brings a deep understanding of enterprise search architecture, including taxonomy design, tagging strategies, and content organization, combined with hands-on experience building AI-powered applications using tools like Quad Code. Ben takes a practical, execution-focused approach to AI, designing and testing real-world workflows that improve how teams access knowledge, make decisions, and operate more efficiently. He is particularly focused on helping enterprises adopt AI in structured, low-risk ways, bridging the gap between legacy systems and next-generation AI capabilities. He represents a new generation of builders who understand both the complexity of enterprise systems and the opportunity to transform them through LLM's scalable search infrastructure and AI-assisted development. Join me in welcoming Ben. Hi, Ben. How are you today? Good, how are you doing? I am good. So my big question for you is like AI is so new and you are kind of an expert in this space. So kind of walk me through your resume. How did you get to this point? How did you discover AI and Quad Code? Sure. I wanted to, so my background is I worked as a product manager, product owner at JPMorgan Chase for about 15 years. I also worked as an engineering manager in that time as well. So I kind of spanned the product and engineering area. So in that time, I worked on enterprise search primarily. And one of the pieces that I wanted to do recently was to create a site, basically a demo site, showing people how the next generation of search could work, how RAG search could work, which is retrieval augmented generation. And what I haven't done recently is code deeply. So it's been a number of years since I've really been into the deep code. So I looked at Quad Code and read all the information about it and said, wow, this is something that really interests me because I understand the concepts, I understand what I want to do, but I don't want to sit down and crank out all the code, learn Python, learn all the latest languages. So what Quad Code enabled me to do was really just talk to it and help to start generating a site, start generating what I wanted to do and generating the outcomes that I wanted to achieve without having to sit down and relearn all the latest tools, all the latest code languages. Okay, so talk to me a little bit about what exactly is Quad Code versus Quad? Like, are they the same? Are they different? They're different. And so Quad is more of your general, you're going to chat interface with it. It's going to just give you strictly answers to your questions and discuss things with you, come up with your results based on whatever you prompt in there. What Quad Code actually does is it sits on your desktop or sits in your computer and it actually physically manipulates the files, creates code, creates content on your machine. Now you have to allow all of these things, but it won't automatically do these things. It'll ask you for permission, but it will actually go out and write the code, write the files, actually deploy where you want it to. So it's beyond just talking to it, which it still has the capability of doing. So it has all the capabilities of Quad where you can still talk to it, ask it questions, ask it information, but it will actually go out and write code, write things on your file system and make it happen. So tell me a little bit about, I hear a lot about vibe coding. Like, can I do that? Or is that something that like, you still need to have like this background in coding? Because I don't code. So like, what is that? So vibe coding, I think it's more of just kind of the, they call it, because it's your general feel of what you're doing. You kind of know what you want to do. And the vibe is, you know, just the talking, the, you know, just doing it on feel based on, you don't have a plan. You're just, you're kind of just kind of rolling with it based on what you feel and what you're doing. So you don't have to have a big plan and walk out, you know, have every single step figured out. It's more of just kind of iterative working and just, you know, doing it based on feel. What I think the interesting part is that you can see the results really quickly. I'm like, so does, you know, you have an idea, you can see it come to life very quickly, as opposed to having a full plan, exactly architecting what you want to do and everything else. You can just kind of talk to it. And what it's amazing about, I think about cloud code is it really fills in the gaps and it'll actually questions along the way. If, you know, in turn, if it needs, you know, specifics or, you know, if there's a branching decisions, like, you know, where do you want to go with this? So it's, you don't have to know a lot. You just have to think, you need to know what you want to do. Right. And then it will really help you along the way. Okay. So I'm kind of surprised. So cloud is like a partner then with you in this cloud code. Cause like it actually asks you questions. Yes. Okay. So that is really kind of interesting. So did it take you a while to kind of get to figure out the program or was it pretty just like, as soon as you started using it, it was easy. It was iterative. So I think he said, you kind of want to start with, you know, your general idea. So mine was, I said, I wanted to create a personal website and I wanted to do it with these specific search engines. And, you know, I wanted to host it, make it, you know, a specific domain by Novak.com. And I gave it all of this information. Now I had already procured by Novak.com like as a domain, but what was amazing I feel like about the cloud code piece was it walked you through the steps. So I'm like, okay, I need to know how to, I want to be able to deploy this. I want to have all this work, you know, working through and automatically deploy. I want it to automatically go out there and push, push updates when I ask it to. And what it came up with, it was, you know, build plan, it came up with its own build plan. It came out with the things I needed to do that it couldn't do. So it actually, it really filled in a lot of the gaps. And what's, I think that's what's pretty amazing is a lot of these services and sites have built themselves to be very accessible for automation. So once you have, once you've been set up, you can, you know, it will, you can tell cloud what you want it to do and it will go out there and perform it for you. So like I said, an example is like, you'll have to sometimes go out to a browser and get an account for something. And it's like, okay, he's like, get me an account for this site, but give me your API key so that I can act on your behalf to do all these things for you. So like push all the code out, but do this. So that's, then once you do that, it's doing all that for you in the background. Have you ever made websites in the past? Not, not in a, not really. I mean, I used to work on Accenture.com and pieces like that. So, I mean, I worked on backend code. But I didn't really go out and build my own personal websites and not, not definitely not recent with all the recent frameworks and everything. I haven't been building, you know, my own personal websites. Okay. That's really incredible. How long did this take you to build your own website? The first iteration only took me a few hours. And I said, and I think the, now I also did a number of things with search engines in the background, which was a lot more complex, I think, than more, you know, a lot of people would do. But what I felt was amazing about it was once again, if you talk to Cloud Code and say you want to come up with a plan and you give it as parameters and it depends on how specific you want to be. I said, look, I don't want to spend a lot of money. I want to, these are the things I have available to me. This is my budget. This is where, you know, and these are the things I want to accomplish. Help me figure out a plan to make that happen. So there's a thing called plan mode in Cloud Code and it will come up with a specific plan of what you, what you can do. And then, you know, so you can talk back and forth. I said, this is like an advisor when there's no stupid questions. It's not going to judge you based on anything you ask. It's not going to, you know, you're like, I don't understand what that means. Or you go, you know what? I don't like that. Can we do something else? Or, you know, I said like, okay, this is going to be like $300. Like, no, I'd really like it to be like, you know, $15. Can you figure that out for me? So those types of things. Okay. That's so interesting. I think what I think is so interesting is I have so many people who say to me, like, I don't know how to use AI. I don't really want to use it. But what you're saying is, it is like the easiest thing to use is there's absolutely no judgment. Exactly. Okay. So then that's like kind of builds on everything that we've been talking about over the past couple of weeks is just use AI and start using it because that's how you kind of learn it. How has all of this change with AI and everything changed the way you think about coding? I think in the long-term, I think we'll still need developers who understand the code, who, like I said, at the fullest extent. I don't think they'll have to do some of the things that I said in the past when I ran engineering teams, you know, setting up simple database connections, setting up specific, the simpler things, this will be all automated. You can hand, so one of the other things that Claude ships with or allows you to do is there's this thing called Claude MD, which means it's called, MD means Markdown. What that means is it's under, it's the under overarching set of instructions for your project. So if you have something that you do every time, you have like specific area, like a specific way of doing things. So like you said, for developers, you know, like when you bring developers on, you say, this is how we do things in this company. This is how we, this is where we put our code. This is where we put our, this is how we do things. This is when we run our tests, where we run our tests, all these things. You put this in the Claude MD file and it's like the overarching instructions for Claude to bring into every project. So it's kind of like, you don't have to train the developer. It's gonna come in and go, okay, I know how to do things your way. And it will always bring that into the project. Okay. That's shocking to me that it's like that good that you can basically train it on all of your parameters and then it's just going to do that every single time. Before I move on to my other questions, I actually have a question about like people who aren't coders like myself, like I've never used Claude. What do you like about just the regular, straight up Claude for like normal people, the one you would just put into your browser? What's good about that compared to like a chat GPT or a Gemini? I've used those. So I've used all three of those and I haven't been using a chat GPT in the recent iterations. I have been using, I've kind of been leaning towards Claude and Gemini recently. One of the things Claude I think really does well, it's incredible with text, it's incredible with taking you down paths, asking clarifying questions. When you said it's maybe it's not clear exactly what you want. It'll give you a little list of things to click from, saying, okay, which path do you want to go down? And you'd like more looking for A, B, it'll literally give you a list, A, B, C, D, or something else. Like what path do you want to go down? It gives you like little diagrams on what it's thinking and potentially your answer or response. And one of the things that has been amazing is how much it changes every iteration. I mean, it gets, Opus I think just rolled out 4.7. It's a little bit stronger, which is all the capabilities, every single release become better with each all these systems. It's pretty amazing. What's Opus? So in Claude, you have a number of different models available to you. And this is kind of important to know. And it's like I said, it's actually all these different systems all have different modes or models to go through. With Claude, it's an anthropic set of models. There's Haiku, and that's like the fast, quick answers. And like something that doesn't need to think deeply about, that's the cheapest one to use. So like I said, when you buy a subscription, I think it's, the other piece you want to know is, you are buying a number of interactions or like token interactions, they're called tokens. Ah, okay, okay. Easiest way to think of tokens is your chunks of words. It's not exactly a word, but it's probably the simplest way to think about it to begin with. So it's a little chunk of information. So each model, like each interaction, there is like faster thinking, but maybe it's faster response, but cheaper. So like Haiku, you think of it like for quick answers, it's still smart. I mean, all these are incredibly smart, but it doesn't have to think deeply. And it's like an answer that can come back with quickly. There's a second one, which is called Sonnet. It's more of your general use. And that's the kind of their median model. And that's one that they mostly suggest for most of your work. And then Opus is like their top of the line, their current top of the line model. And that's the heaviest thinking. It'll take a while for its response, but it'll think more deeply. And I believe it also has the highest of what's called context window. And what that means, so a context window is your, is how much information the AI can take in, in your conversation. So every time you run your, you have your conversation, right? You click new chat and it's basically fresh data. It does not really reference your older conversations. Oh, okay. I didn't know that. So now, I mean, things have changed a little bit. There is a memory has been starting to be baked in some of these things, but there's still, it doesn't know everything about every single conversation you've ever had. It'll know general pieces that, and this is a recent development, but it said your context windows, everything about your conversation, everything and all the information you've given it. So all the information it can take into account. So whether it be, you're uploading files to read, you're uploading documents, that's part of your context. And it said, that's the part that it can take into account when it's generating its answer. Wow. So the longer these things go. That's kind of like when we're having a conversation, I can reference something from like when we started. Right. But if it was like too long ago, it said if you reference it from like a couple of days ago, you're like, I'm not exactly sure. I don't remember what we were talking about. So that's kind of what that really is. Okay. And these are all paid. The Haiku, Sonic and Opus. So those, I think there's a free version. I'm not exactly sure how much you get per. Okay. But for most people, you can just start with a free version. Yeah, you can start out with a free version and you get, you hit a limit. There's a specific limit, time limits. So if you need to use it more, that's where you start going to the more paid versions. So what surprised you most when you started using AI? Basically the quality of answers, the quality of just being able to bounce ideas. I think that kind of going back to what I said before, it's a judgment-free zone. And, you know, you're like, I don't know anything about this. You know, like explain it to me like I'm a child. You know, it doesn't care. You know, it just, you know, it gives you some simplistic, a simple answer to start your information. You can ask clarifying questions. And it said, it's now, once again, you have to take everything with a grain of salt, right? There is thing called hallucinations. It may not be a hundred percent right. I've heard about it. So what does that mean? So AI hallucinations are basically, you know, it will make things up. So the way all these things work, it's, I mean, it's not real. It's basically anticipating the next word. Sometimes the, you know, it will give you answers that don't exist. It will come up with an answer. Sometimes you have to keep it on the rails, asking for specifics, making sure it doesn't, you know, come up with a specific or like something made up. It's hard. I can't think of an example off the top of my head. Yeah, exactly. It's basically anticipating. Huh. I don't think that's ever happened to me. How do you know if that happened? I don't think you do. Instead, unless you really know the information, that's this, I think that's part of the challenge. And like you said, if you know the information that you can start seeing. So when you're looking at it more of an enterprise text or context, you know, you know, your policy, you know, it'll come up with like, it'll make up or derive some information about your policy that maybe doesn't exist. It's like, well, it's kind of figuring out based on the policy here and the policy here, this would be your policy, but it doesn't really specifically say that anywhere in your policies. Oh, so that's why it's so important to have like really good clean data when you're starting to use AI somewhere in a company. Correct. Okay. Cause it's going to make stuff up if it doesn't have the answer, but it's kind of like the best guess. Is that a great way to say it? Correct. Okay. But it's called hallucination. That's correct. Okay. Interesting. What do you think is most important now for like someone in your field? What's important in this new AI world? Cause things are changing so quickly. Like what, what do people need to think about? I think I look at, I try to look at AI as an accelerant and then I don't like the idea that it's replacing people necessarily, but I'm hoping that it's accelerating people's work and hopefully to get to past the mundane, you know, the things that would take us time, but it's still useful, but it would take us, so for example, like I would, you know, I would look at reports. I'd want to categorize, like in a search area, I'd want to categorize specific queries. I'd want to, you know, look and try to find patterns of why people, you know, are missing out on searches, you know, and it's more, that's usually a manual thing. Now you can pass this to AI and say, you know, what are some patterns? What patterns are you seeing? What, you know, where, what categories are you seeing? And if you have specifics, you can, you know, you can give it, so it's, there's different prompting techniques and you can give it specifics, give it examples and it'll come back with, you know, trying to fit into your examples or you can kind of just say, hey, you know, I don't care, you know, figure it out yourself. What are some categories? And, you know, it's interesting to see both patterns, right, I think there, because if you give it, if you're guiding it to the way you do things, you know, it's good, it's your, it's exactly what you were looking for, hopefully, but then sometimes it's interesting to see what is the AI thing? Maybe there's patterns that you're just totally not seeing or they're outside of the box that you wouldn't normally see. So I think there's different ways of using it and I think that's a big way to look at AI is like to be able to like think outside of the box, quickly look at data, quickly look at decisions and maybe different ways of looking at it. That's interesting. So it's kind of the more fun parts of work, like the more creative instead of the slog of going to work every day. Exactly. So many people are kind of afraid of AI, I have to say, I've kind of been, it's invigorated me as like, I mean, I'm getting so many ideas, like you said, it takes away that kind of dumb stuff that I don't wanna do, like go through this report and tell me blah, blah, blah. I mean, I'm only scratching the surface. I am not a Ben Novak, like writing my own code. It's just like, give me X, Y, and Z. So I think it's interesting that you're using it in such a complex way to even think for things that maybe you miss. That's actually very, very clever. What's your biggest challenge when you work with AI and cloud code? I think one of the areas, so I said, it's definitely a learning process. I think one of the things that was sort of the context was something you start to learn about very quickly. So once again, I said like the context, the entire area that it's taking into account, you want to, every time you answer, you give it another instruction, it's gonna add that to the context. And that window, so when you're paying, with the cloud code, you're paying for what they call, what I said, the tokens, right? So the input, output. Every time you're answering or taking those, you're adding something new, it's sending all of that. And so your context starts getting huge. If you're working on the same project and you're just keeping it in the same window. So what you wanna do is really start understanding the concept and there's capabilities inside it where it's like, there's a thing called compact. It's like, okay, here's all the information. What it does is it compresses it down to the brass tacks of what really it needs to know out of that entire conversation. So it's not constantly sending all that information. So it's like, what I've learned about all of this is how to optimize how you're doing things because you learn it really quickly. So I was on the $20 plan and I quickly hit my limits and I couldn't figure out why. So I upgraded to the $100 plan. But I mean, so it's, I think, I also learned, I said, more of the capabilities. I said, that compact is a big piece and just really understanding all, you don't have to keep a giant conversation going. Every time you wanna do a new function, it's gonna be able to read the code that you're doing. So it knows that piece of it. Now, if you wanna keep the context of what you're doing in between, that's where I was talking about the Cloud MD. You can ask it to kind of update that. That Cloud MD is like the memory of the project. It's like, okay, these are the things we learned. I mean, once you do it, it's kind of interesting to have it write back and say, hey, can please update Cloud MD for me? And so you have memory of what we've been through. And what's really interesting, it's just on plain text. It has all of its learnings. It's like, okay, we tried this, it didn't work. We did this. The user likes to, I said, I learned, like I said, I learned some of the hard ways. You know, haiku is cheaper than sonnet, than opus. And those are the things it's like, this person wants to save money. Default to the cheaper, you know, unless it's specifically told. So, I mean, I think there's a lot of learnings there. It's more, I said, it's just kind of learning as you go. I mean, I don't think, the thing is you can't really do anything wrong. And it runs all locally first too, which is nice. So like I said, I wanted to publish it to the internet, but you don't have to do that to start. You can start it with saying, hey, I just want to build a website. And it can actually start everything up on your local machine and not actually push everything else. You can just look at everything on your own. Okay, that's so interesting. So I was going to ask you what happens if something is wrong and you already answered that. Now I'm curious, I have a new question that I didn't even have written down. Like, what have you learned about your thinking? Like, I find it interesting that you were saying like, you know, it's telling you like, you like the cheaper option. Like, have you learned anything else from the way you use it? I mean, I find myself asking a lot more questions, a lot of what ifs. And I mean, I think what's amazing about this is like, once again, there's no judgment. You're like, okay, you know, I'm thinking this, you know, help me figure out, does this make sense or not? What are the options? What are the impacts? And it was like, I was running some specific tests on search engines, but it was like, I want to try this test. Does he think it's going to move the dial on the metrics we're coming up with? And you know, it kind of came up with anticipated things and yes, no. And it said, but I think what's amazing about all of this is it usually comes back with a lot more than what you asked it. So it's like, you know, it usually doesn't say yes, no. It'll say, it'll give you a long description. It'll give you a lot of information of maybe you want to try this. And you know, I think that's what's amazing is it can help test your thinking, but it can actually teach you a lot along the way too. You know, if you want to really try to do new things, I think it's something to bounce ideas off of. I love that. So my other day job is I teach English and pronunciation and I love the idea of using AI to kind of test your thinking. If you kind of like put yourself out there and like ask some more questions. So I absolutely love that. Okay, let's use your crystal ball before we wrap up here. What do you think is gonna happen in like two years? I know that's a broad question. Maybe like, you know, for people, normal people like me and then like for people who are specialists in your world and the enterprise, like where do you see all of this AI stuff going? I think things are gonna get very accelerated. I think people are very excited about it, but what I haven't seen yet in practice and I'll be curious to see as I talk across different companies, how much production code are they rolling out? You know, and what levels of production code are they rolling out to different areas? So, I mean, so if this is your public facing website, is this your face or your company, how confident are you you're gonna roll out AI, AI generated code without any person looking at it? And how many people in the middle before you have the trust that it will just go out and do something? And I said, you hear a few horror stories where the AI deleted somebody's database or something, you know, and it's learning. So I think that's, I think in the few years, I thought, I think we'll be figuring that out. I would think, I think things will get accelerated in terms of ideas to production. But I think everybody's still in the learning phase right now. How much do we trust exactly what's out there? And then how much do you trust that you don't know? I mean, I think one of the advantages of human written code is that you have a concept, you wrote it, you have a concept, once something breaks, you go, okay, I think I know where that is. I think that was the interesting part. I said, when you mentioned the bugs, when I rolled out my website for this, I wasn't writing the code. When it broke, I'm like, I have absolutely no idea why this is broken. The interesting part, though, I think it was the capability it had. It gave a pretty good error message. I copy pasted it and said, hey, this is broken. This is the error message. Can you go fix it? It said, to its credit, it would go out there. Now, it wasn't 100%. I think that was the other thing is, it did roll out broken code, not every once in a while, something that didn't work. And you're like, okay, well, you probably should have caught that. I know my setup isn't perfect yet in terms of asking it to test, validating, things like that, because I've been just kicking the tires on things. But I think with an enterprise, they probably have a little bit more rigor around some of the things they'd roll out. But getting back to the bug piece, I think what was amazing is, I did hand it the error message. I told it what was going on, and it was able to figure things out. And you can actually see the live thinking of how it's working. And it's interesting there, because you can almost hear it thinking, it's like, okay, I tried this. Oh, that didn't work. So I'm gonna try this now. Okay, well, that didn't work. It'll start thinking, okay, thinking of alternative approaches. So I mean, I think that's what's amazing. You can kind of watch it thinking. It'll tell you what's going on. That's interesting. So you're kind of learning as it kind of- Exactly. Okay, what recommendations do you have for people who are just kind of getting into this AI space? Do you have any tips? I would say definitely just start with, start with the free versions of whatever one you want. There's Gemini, there's Claw, there's the OpenAI, ChatGPT, whatever you feel comfortable with. They're all getting better every day. Start asking questions, but start asking deeper questions. I mean, realize that you can be really specific about what you want, and look up, I would say, look up prompting techniques. And it's not really, it sounds more difficult than it is, but it's more of like look up prompting techniques. Look up how to give examples to the AI. It can look out how to get the best responses from it possible beyond just asking a simple question. Have a look at something and maybe give it more specifics on how you want something back. And you'll be amazed at like how it returns things in the way you want. I think that's really the starting area is just try, like I said, start prompting, look up like prompting techniques. I said, once again, it sounds a lot more impressive than it is. It just kind of, it's very general. And then just try each way and see what you get out of it for your own personal life. I think that's the, I think that's the best way to do it is like something that you are interested in because some of the trainings you take, you're like, okay, they tell you to type something in and like, it means nothing to you. So it doesn't really resonate when you get the answer. So I think it's at least get the idea of what they have in there and then do something yourself. So interesting. Okay, is there a book for this or you just Google it or chat GPT it or something like that? Yeah, so there's chat, I think you can, I mean, I asked Claude how to use Claude, so much. That's brilliant. Yeah, I mean, in terms of like, I'm in Claude code, I'm like, how do I do this? What am I wanna do? So, I mean, I asked Claude how to use itself a lot. I think the other thing that is out there is called notebook LM. It's a great way to learn. That's amazing. Yeah, isn't it? So once again, it's, you can start the simplest way is just say, hey, you go to notebook LM and say, I wanna learn about this, you know, like specific, I wanna learn about how to use AI, how to prompt, you know, and then it'll pull in sources. You can, after that, you can actually generate, the abilities are amazing. You can generate a slide deck, generate videos, videos based with themes. So I mean, it's that if you wanna make a cartoon, you can do it. So I think that's a great way to learn it. That's how I've been kind of looking at new things to learn that way, because I just make myself a quick podcast or video to listen to while I'm like driving or something. That's amazing. Okay, well, Ben, this was incredible. I learned so much today. Thank you so much for being here, everyone. Thanks for joining us as always, and we'll see you again soon. Take care, everybody. Thank you.