23:40Brian Casel · Tutorial
Claude Fable: Build me an app
Brian Casel skips the toy demos and hands Claude Fable a real production feature — then shares the two things it changed about how he thinks about AI-assisted building.
June 11thThe argument in one line.
The most valuable app you can build is the one that fixes your own industry's broken workflow, and a rigorous PRD-and-milestone process is what turns that fix into software worth selling.
Read if. Skip if.
READ IF YOU ARE…
- You work inside an industry with a workflow held together by spreadsheets, sticky notes, and a CRM that doesn't fit how the work actually moves.
- You're building, or considering building, software for your own field and want a repeatable process for turning an idea into a real, tested application.
- You're using or curious about Claude Code and want to see how a structured planning phase changes the quality of what the AI builds.
SKIP IF…
- You're looking for a line-by-line coding tutorial — this walks through planning and process, not syntax or implementation detail.
- You already have a mature product development process and are past the 'how do I even start' stage.
The full version, fast.
A software builder argues that the most valuable app you can make is the one that fixes your own industry's broken workflow, since everyone else in that field has the same unsolved problem. He demonstrates the process end to end by building a real estate CRM: a structured product-requirements interview locks in what the tool does and doesn't do, a data model defines the core entities, and the build is broken into four dependency-ordered milestones. Each milestone gets its own technical plan, a self-verifying build pass, and a short log that carries decisions into the next session. The video closes by naming the two things that turn a working internal tool into a sellable product: multi-user access and billing.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:32
01 · Build it, then sell it
The thesis: the tool that fixes your own industry's workflow is the most valuable thing you can build, because everyone in that field shares the problem. Brian introduces himself and the channel, plugs his free workshop, and starts a fresh application for a real estate CRM.
01:32 – 09:38
02 · Planning the CRM
Using a PRD creator skill, Brian is interviewed by Claude to define the app's features (contacts, pipeline, checklists, follow-ups, photos, mobile-friendly), what's explicitly out of scope for v1, and the core data model (contacts, deals, tasks, checklist templates). The PRD is generated as a browsable HTML doc.
09:38 – 16:35
03 · Plan mode and milestone 1
The build is broken into four dependency-ordered milestones, each with its own prompt.md. Brian explains the milestone-log hand-off pattern, then runs Claude Code's plan mode to turn the PRD into a technical implementation plan for milestone one, answering a few final UX questions along the way.
16:35 – 22:33
04 · Contacts and deal pipeline
Claude builds milestone one (contacts), self-verifying its work with a headless-browser skill and its own tests. Brian tests the live app, finds three small bugs (missing autofocus, a field that defocuses while typing, search not covering email), and fixes all three in one rapid-fire pass.
22:33 – 29:37
05 · Final build and demo
With all four milestones complete, Brian renames the app, tests the deals pipeline, hits and fixes a bug where attaching a photo before saving a new deal throws an error, tests follow-ups and checklist templates, and has Claude generate and apply a real-estate-appropriate color palette.
29:37 – 30:52
06 · Turning it into a SaaS
Brian closes by naming what would turn this internal tool into a sellable SaaS: inviting other agents and adding billing, priced either flat or by the number of properties managed. He points to his free workshop and training for the deployment and business side.
Lines worth screenshotting.
- The tool you build to fix your own workday can be the most valuable thing you make, because everyone else in your industry has the same problem and no purpose-built fix.
- A detailed planning document takes more time than writing the code, and skipping it is what turns AI-built software into unusable output.
- Deciding what a tool will NOT do in its first version is as important as deciding what it will do.
- Mapping the core entities — the people, the deals, the tasks — before writing any code clarifies what the whole application actually needs to support.
- A product requirements document and a technical implementation plan are two different documents: one defines what to build, the other defines how to build it in code.
- Breaking a build into milestones ordered by dependency keeps a large project from becoming one unmanageable request.
- A short log of decisions and learnings, written at the end of each work session, lets the next session start fresh without losing prior context.
- An AI coding agent can verify its own work — screenshotting and testing in a browser — before a human ever looks at it.
- Batching several small bug fixes into one rapid-fire pass at the end of a work session is more efficient than fixing issues one at a time.
- An edge case as simple as attaching a photo before a record is saved can break a workflow that looks fine on paper.
- A checklist template that auto-applies to every new record only works if the underlying process really is the same every time.
- Two things turn a personal tool into a sellable product: the ability to invite other users, and a way to charge them.
- Pricing a niche tool by the unit of the work itself — per property, per deal — can fit better than a flat generic subscription.
Build the tool your industry is missing, then sell it back
WHAT TO LEARN
The most valuable software often starts as a fix for your own workflow, and a disciplined planning-then-milestone process is what turns that fix into a tool other people would actually pay for.
01Build it, then sell it
- The most valuable software idea is usually the one already sitting in your own daily workflow, because everyone else in your field has the same unsolved problem.
- Being the builder and the domain expert is a genuine edge: a tool built by someone who lives the workflow beats a generic system retrofitted to fit it.
02Planning the CRM
- The planning phase deserves more time than the coding itself, because a detailed upfront plan is what keeps AI-built software from turning into unusable output.
- Explicitly writing down what a first version will NOT include is as important as listing what it will, since it keeps the initial build achievable.
- Mapping the core data entities before building anything clarifies what the underlying system actually needs to support.
03Plan mode and milestone 1
- Breaking a large build into a small number of ordered milestones, sequenced by dependency, keeps the project from becoming one unmanageable request.
- A written product plan and a technical implementation plan are two different documents — one describes what the product does, the other describes how the code will do it.
- Passing a short log of decisions and learnings from one work session into the next preserves context even after starting a completely fresh session.
04Contacts and deal pipeline
- Software built by an AI agent can verify its own work — taking screenshots and running tests — before ever being shown to a human, catching obvious errors early.
- A short, deliberate round of bug-fixing at the end of a work session, batching several small fixes into one pass, is more efficient than fixing issues one at a time.
- Testing a feature by actually trying to use it surfaces real usability gaps that a written spec won't catch.
05Final build and demo
- Errors during testing can reveal edge cases in a workflow's sequencing that are worth fixing before launch.
- A reusable checklist template that automatically attaches to every new record only pays off when the underlying process is genuinely repeatable.
- Small visual details — a default color palette, dark mode, sensible defaults — do more for a tool feeling finished than any single big feature.
06Turning it into a SaaS
- An internal tool becomes a sellable product once it adds a small number of specific things: the ability to invite other users, and a way to charge for access.
- Pricing a niche tool can follow the shape of the work itself instead of defaulting to a flat generic subscription fee.
Terms worth knowing.
- PRD (Product Requirements Document)
- A written plan describing what a piece of software does and doesn't do before any code is written; used as the source of truth for the whole build.
- Plan mode
- A mode in an AI coding tool where the agent reads a task and produces a technical implementation plan for review before writing any code.
- Milestone log
- A short written record of decisions and technical notes from one build phase, handed to the next phase so context isn't lost when a session restarts.
- Agent browser skill
- A tool that lets an AI coding agent open a headless browser, click through the app it just built, and screenshot the result to verify its own work.
- Design system
- A pre-built set of reusable colors, fonts, and components that a new application starts with, so visual styling doesn't need to be built from scratch each time.
Things they pointed at.
16:40toolClaude Opus
20:00toolAgent browser skill (Vercel)
Lines you could clip.
00:00
“The tool you build to fix your own workday might be the most valuable thing that you ever make because everyone else in your industry has the same problem.”
00:45
“If you know exactly what your industry is missing, then building the perfect tool for yourself and for your people is actually within reach.”
16:40
“If we had not gone through the detailed extensive PRD process and having it interview me and extract all of my key decisions, I would not feel very confident in this technical plan.”
13:00
“I like that pattern to be able to pass along the learnings from one milestone into the next.”
30:00
“I bet that if you're in real estate, you probably know exactly how to make this even more useful and maybe even sellable as a SaaS.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
00:00The tool you build to fix your own workday might be the most valuable thing that you ever make because everyone else in your industry has the same problem without the tool that's purpose built by someone who knows the work inside and out. So today, let's build a client management app for real estate agents. Now, if I were a real estate agent, which I'm not, this is what I would probably build.
00:21And you know what? I would probably also look to sell a tool like this to other real estate agents and turn it into a SaaS business. So today, I'm gonna walk you through that motion to show you that if you know exactly what your industry is missing, then building the perfect tool for yourself and for your people is actually within reach.
00:38You know, the real estate niche is interesting because the way that they manage deals can be a mess. I mean, clunky brokerage CRMs, a couple of spreadsheets, sticky notes, a chaotic inbox, none of it is shaped like an actual real estate deal.
00:52Most real estate agents have that same problem and so once someone who really knows the industry well is able to build a tool, well, that's . Now, even if you don't work in real estate, I bet you've got your own version of this. A workflow that your industry waiting for someone who really gets it to build the real thing.
01:13Real quick, I'm Brian Castle, and this channel is about one thing. The people who know an industry cold are the ones who should be building its software even if they've never coded. I made a new free workshop that walks the whole path, VibeCoder to Builder.
01:28That's at buildermethods.com/workshop. Alright. Let's build.
01:32Alright. So I started up a fresh application. We're just gonna call this one real estate manager.
01:37I know, not very creative. Today, we're just gonna be focused on this Real Estate Manager app right here. Now, I am using my starter template for a new application that I use for all of my apps and actually you can use the same starter template that I use.
01:52That's available in my build new application template. Most of all, we're gonna be using my PRD creator skill, which is going to drive our whole process and workflow for creating a product requirements document and then break that out into buildable milestones and then we can build this thing out.
02:12So you can get the PRD creator skill for free by going here and following the installation instructions. But now, I already have all that stuff installed, so let's go ahead and kick this thing off. So I'm gonna invoke that PRD creator skill and get this start.
02:28Right. So to kick things off, we are gonna build a PRD. This is a product requirements document.
02:34And by the way, this part of the process of building any application, I think is easily the most important critical phase and actually you'll spend most of your time in this pre planning phase.
02:48I know that might seem boring, not very exciting, but you know what? The more you get into this, the more you realize how critical it is. And this is actually where your creativity and your input matter the most.
03:02Because if we can set up a really strong detailed plan upfront, then it makes the whole process play out much more predictably and we end up with a tool that is actually useful rather than or something that we need to, you know, circle back and have to fix things over and over again.
03:19Now, the good news is you don't need to have a high level technical understanding of what you want to build upfront. Giving it like what is my main idea for what I want to build here and what are the main features that I think that this type of app should have based on what my actual needs are. Now, of course, I'm not a real estate agent, but if I were or if you are or if you're building something for your own work or your industry, then you know this inside and out.
03:46You know exactly what your workflow is like a CRM for real estate agents. One that fits how deals actually move, not a generic sales CRM. Alright.
03:56So obviously, we're gonna need to manage contacts, leads like leads, buyers, sellers all in one place. Right? We're gonna wanna move these deals through a pipeline with stages that match an actual real estate transactions like new lead, showing, offer, under contract, closed.
04:14We'll want to drag deals along a visual pipeline board. We'll probably need the ability to have like a task checklist for each closing for you know, to get a property deal to its closing and since this is so industry specific, I would imagine that every real estate transaction follows a very common task list.
04:34So it would be good to have the ability to set a template task list to follow on every single deal. We should have follow-up reminders that no lead or deal goes cold. And since this is a an application for real estate and we're dealing with properties, we should probably have some photos of the properties, uh, to really make it visual and easy for me to remember like oh that's this property or that property.
04:56So I finished it off by saying you know the key here is that the pipeline is built around the transaction life cycle. I'd like it all to work well on mobile since agents are rarely at their desk. Now we won't actually be building a mobile app here but the web application that we build will be mobile friendly.
05:14That is all the information that we really need to start to enter into this detailed process. Now the PRD document itself, I'm gonna go with the HTML option on this instead of markdown.
05:25That'll be nice and easy to read and follow. It'll serve as like our source of truth for the whole project. And now here is the pattern that's going to repeat itself as we go through this planning process.
05:37Right? Uh, we have defined what we are building, but we also need to define what we are not building. Because a typical application in this space or this type of application might be good to have at some point.
05:51We don't really need it in v one. Same thing with having like the ability to have a team or multiple agents, email or SMS integrations. These are nice things that we could add on later, but it's always important to start with a minimum viable product like a like v one, get the core fundamental feature set in and that's a foundation that we can build upon later.
06:12Okay. Next, we're getting into the data model. So, you know, I know that this can sound a little intimidating, but the data model is really important to start to think through early in this stage here because these are like the things, the entities that make up the bones and the here.
06:31So obviously, we're gonna need to deal with contacts. These are like the people and there are going to be deals. There will be tasks attached to deals and there will be a checklist template.
06:42Remember, we talked about every deal following a similar template. So we'll need to define a template and then that generates tasks. It'll be these are gonna live in a database.
06:52So there are gonna there's gonna be a list of contacts, a list of deals, a list of tasks and so on. So this is not everything of course, these are just like the core building blocks that are gonna go into this application. But this is a really good time to notice things, to say like, oh, but what about this thing?
07:09Uh, okay. So now we're gonna start to get a little bit more granular and follow a similar process of deciding what's in scope and out of scope, except now we're gonna do that on each individual feature.
07:21So starting with the contact management feature, we're gonna kind of cycle through all the features like this, but now we are going to decide, you know, like line by line. Alright.
07:30So I'm just sort of like pushing through the features. Now I'm up to the deal detail view. So obviously, we're gonna be able to see each property and, uh, the address, a button to view that on Google Maps, some photos to be able to see what that property looks like, the checklist.
07:46Now, it says that a few things are out of scope, but I did notice that one thing is that it's not gonna be showing the price, but I think it should show the price. So that's a correction that I wanna make here. So I'm gonna go into the deal detail question and I'm just gonna type something and say, use voice dictation here.
08:05Actually, we do need to display the price on each property and also a way to set the price or edit it if we didn't put that in already. Alright.
08:16We're gonna go ahead and lock it in, which basically just means that we're we've made the decision and we could always change it later but it's good that we're making as much of these critical design decisions upfront. Okay.
08:27So I finished going through all those big features and then deciding what's in scope and out of scope for each of those. Now we are up to breaking this up into milestones.
08:37And so Claude is recommending that we break this up into four buildable milestones and it did a pretty good job of like deciding what the order of these should be based on their dependencies. Right?
08:50And so I think this looks pretty good. If I wanted to be a little bit more ambitious and get through it faster or we can stretch it out and be a little bit more granular with like six different milestones, but I think the four is gonna work for us. So we're gonna go with that one.
09:03Okay. So it has everything that it needs. Remember, I only started with that basic overview of what I wanna build, but then we went through a step by step process to get in to get all these granular details confirmed or corrected.
09:18And now Claude has a ton of information that it can just pack into a highly detailed scoped out PRD document, product requirements document. And, um, it's gonna build that out as a as an HTML artifact.
09:33And then it's also going to plan out the milestones that we'll be able to follow to build this thing out. Okay. So our PRD, our product requirements document is all set and we have our milestones all created.
09:46So what it just did, and this is a result of the PRD creator skill, what it did was it created a folder in my code base called build plan.
09:56And inside that, we have our actual PRD. So let's open that up. And I can just open that in a browser since it's in the HTML file.
10:04And here we go. So we're gonna be calling this real estate manager. We're gonna call it deal flow.
10:09Right? And this PRD document sort of just pulls together all the details that we've just hashed out. Right?
10:16So we've got everything from like what it does, uh, what we are not building in v one. We've got the data model and how all these models sort of like relate to one another. And then we go through the process of our roadmap of the milestones that we're gonna build out and, you know, details of each one.
10:33So this gives us sort of like the source of truth or the road map, if you will, to go from idea into a built out version one. It turned that into this.
10:44This is a professional product requirements document. And this isn't just performative.
10:50This is actually going to guide every step of the process. And I'm gonna show you how that works here now. So so we have that PRD and then we also have these milestones.
10:59So we already planned out, like, we're gonna build this out in a series of four milestones. And inside each of these, we have a prompt dot m d.
11:08So the prompt dot m d was generated again by the PRD creator skill. And I put this together. I designed this whole process to follow the same workflow that I use for basically every app that I build.
11:21And this makes it really predictable so that I can spin up any app that I need in my business. Right?
11:26So here we go. So this is the prompt. And the way that this is gonna work is I'm gonna go into Claude code.
11:33And first, I'm gonna clear the context so that we clear that clear out that context window. And I can simply drag that prompt MD into Claude code. Just like that.
11:44But one more thing before I send that off, I wanna switch into plan mode. So just do shift tab until you see plan mode at the bottom. And now I'm going to have it read the prompt and it's gonna take and it's gonna use plan mode to create an implementation plan plan.
12:03So why run plan mode if we've already gone through the whole PRD creating process and that was a extensive planning session.
12:12Right? Well, the difference between what we did before in the PRD was we created more of like a product plan of like what's going in, what's going out, our roadmap.
12:22But now Claude is gonna take that. It's gonna see our requirements at least for milestone one and it's going to make its, like, technical coding implementation plan, which is quite different.
12:34It's gonna it's create a very technical plan, which you might not even know how to read that, and that's okay. And that's what's gonna make this whole thing work much more predictably. So Claude is churning on that.
12:45It actually fired off a couple of its own sub agents to sort of like look through these and reanalyze the plan a little bit. But just to show you what's going into this prompt, it's going to have Claude focus only on milestone one.
13:00We don't want it to go ahead and build everything else out. And something I like to do is have Claude write up this milestone log dot m d.
13:09And it's gonna create that and save it inside of the folder for each milestone while it builds the the work in that milestone.
13:19You know, because when it goes through the work of coding it up, it'll probably run into all these little mini decisions and learnings and things. And when we move from milestone one into milestone two, and then clear the context again, we don't want Claw to just forget about everything that already happened in the previous milestones.
13:38So each of these subsequent prompts are going to tell Claw like, hey, we've already built a few previous milestones.
13:46Read those milestone logs to understand what was built. So now you can start fresh with that knowledge intact. I like that pattern to be able to pass along the learnings from one milestone into the next.
13:59Right? Okay. So it's been churning on that plan and before it actually finalizes its coding plan, it's gonna ask me a few more final detailed UX decisions or questions for me to decide before it we really move ahead here.
14:15So, uh, again, it's really helpful for me to be able to answer these questions upfront. Well, at first I thought that a snappier modal might work better, but I think in this case, let's go ahead with like full dedicated pages for each contact. So technical question about how the search should work and even if I'm not like fully knowledgeable of how this stuff technically works, Claude has given me a pretty good description of what it means and like sort of like the trade offs involved.
14:41So this one is like more instant, it's great for a single agent contact list, Simplest. Yeah. That sounds good to me.
14:46And some of these fields, do we want them to be plain text or rich text with markdown? Definitely want the markdown. And that's actually built into my design system already, so that'll be nice.
14:56Okay. Looks like it has everything it needs to write the plan. Alright.
14:59So Claude finished up its implementation plan for building out milestone one. We're gonna be building out the contacts feature here.
15:06Now, here is that technical implementation plan. And as you'll see, it's quite technical. There's a lot of jargon in here.
15:12You might not know what a lot of this stuff means. And to be honest, that's okay because I feel pretty confident that it's good to go.
15:21Now, I wouldn't normally feel that confident in it even using today's latest models. I happen to be using Claude Opus today.
15:29If we had not gone through the detailed extensive PRD process and having it interview me and extract all of my key decisions, if we had not gone through that whole exercise, I would not feel very confident in this technical plan.
15:45At least, I wouldn't trust that it actually is planning on building the thing that I wanted to build. Since we did create the PRD and since we did have it analyzed that in detail, I know is the implementation plan for those requirements.
16:00Right? So I'm good to go on that. I'm gonna go ahead and say yes and bypass permissions.
16:05So that just switched us over from plan mode over to, you know, building mode or like, execution mode, if you will. So now you can see Claude is actually getting to work. Now we are finally starting to break ground and build some code.
16:20And and here we go. So while that's working, I can jump over to one of my other projects if I want to, or I can go, you know, take a walk, get another coffee.
16:30I've probably had too many of those today. So, yeah. Let's see how we go.
16:35Alright. So our milestone one is all set, and I can see that it worked on that for about twenty three minutes. That a that was a meaty one.
16:42Let's see what happened in the milestone log that Claude wrote up here. So here is the detailed recap.
16:49These are some like technical notes that like I don't really need to spend too much time reading into these, but it's good to know that Claude is kind of passing along these technical notes and decisions and things that it ran into during implementation. It's gonna pass those along to the next agent that is gonna be working on milestone two.
17:06But I do like up here that I have this simple human readable bullet list of everything that was actually built in this milestone. So now that I'm about to open up the browser and check it out, I know what I should be able to test and be able to look for in terms of like what's new.
17:24Let's go ahead and run the server and see how we're doing. Alright. So my application template has this like little hello world page built into it.
17:31I'm just gonna log in. And here we go. Like, this is my standard layout.
17:35If you've seen some of my other build videos, you'll notice that my applications tend to pretty much look the same to one another. I often like change up the colors and fonts and stuff like that, but I've got my standard design system already pre installed for me to be able to go in and build any app for my business, which has been nice.
17:53Anyway, let's go back to the application itself and see what we've built so far. And the main thing that we've built but here we go, looking at the contacts page and a nice empty state.
18:04I like the two tone here. Okay. Yeah.
18:07This this is looking good. Let's go ahead and create a new contact. Now this one is sending us to a dedicated page to create a new contact.
18:16One thing I don't love about this is that it did not auto focus to the name of a contact. Okay. So what type of contact is this?
18:24So is it a lead? Is it a buyer or a seller? Again, we're creating a CRM that is specific to the real estate industry here.
18:31So, you know, we can build in some like niche specific features. But you know, I noticed something here.
18:38Oh, there's a little bug. I just noticed that. When I type, it defocuses the field.
18:46Alright. So that's another thing that we're gonna need to fix. Let's just make sure that we can create this contact and we can.
18:53Looks good. We've got two contacts in the system here.
18:57We've got these nice labels. Again, these pull from my design system. Let's just make sure that the search is actually working.
19:03So yeah, I could just start typing the name and okay.
19:08So it doesn't search by email address, but will search by name. Let's also fix that.
19:15So those are the three things that I want to fix. I want to Yeah. And we can filter by types, so that's pretty cool as well.
19:20I'm in this like rapid fire mode now. So we've already done the milestone build out. Now I'm in this like rapid fire refinements mode.
19:27This is something that I do near the end of every milestone and I'm gonna just kind of throw a couple of fixes at it. I wanna keep the contacts the context intact since these are sort of some like follow on tasks.
19:38Let's see. The first one was to when we create a new contact alright. When we create a new contact and I enter the new contact page, I want the name field to be auto focused.
19:51Alright. So that's the first thing. I could just have it do that, but I'm gonna give it a few tasks to do all at once.
19:57There's a bug on the notes field on a contact.
20:01After I finished typing, it defocuses the field. It should not defocus.
20:06I should be able to continue typing. In the contacts search, I should be able to search by name or by email address.
20:15Alright. So I'm gonna have it turn on those fixes and we'll see how it does.
20:19Alright. So it's making those changes. It's almost done and I can see here that it's actually verifying its own work in the browser.
20:27It's using a headless browser. It's have it using a skill called agent browser, which I think is a free skill released by Vercel that allows it to check things in the browser.
20:37It even takes a screenshot of its work. So it'll just like create these like fake records and screenshot them and self verify.
20:46And that's really helpful to have it catch its own errors before it even delivers work to me. In addition to that, it is also writing its own tests and making sure that the tests pass a 100%.
20:59Again, it's just another round of self verification built in. And I have that stuff defined in my claud dot m d file.
21:09I search for verify. So another really helpful way to ensure that we're, you know, we're just not wasting our time.
21:17We're not . It just makes it kind of a smarter agent. Alright.
21:21So those fixes are done. Let's take a look. I'm gonna refresh.
21:24I gotta restart the server. Alright. First, let's see if we can search.
21:29Yes. So I can search by the person's email address or by their name. And if I edit a contact, test again, test.
21:41Yeah. So that bug is fixed. Very nice.
21:43And let's go ahead and create a new contact and good. We have auto focus the name, so I can just start, you know, typing away right now. Know how to spell his name.
21:52Alright. We're looking good. Alright.
21:53So my next move here would be to come in here and I'm going to clear this context and now we can move into the next milestone in our in our building here.
22:05So milestone two, I'm gonna take that prompt and I'm gonna drag that right in here. Just like I did before and I'll switch into plan mode with shift tab until it says plan mode.
22:17And then I'm just gonna run this over again. I'm gonna get that going. It's gonna check out milestone two.
22:21We're gonna make a technical plan, and I might answer a few questions, and then I'm gonna have it build again. Now I'm gonna have it I'm gonna push forward with a few more milestones here, and I'll check back here in just a few.
22:33Okay. So our real estate manager app is now actually done with all of the four milestones. So I'm gonna check it out and give it a once over to see if we need to do any final refinements.
22:44So I am I'm running the server here and yeah. Here we go. Alright.
22:48So I see that we haven't yet updated our title in the top left, so let's just get that taken care of real quick. Change the title in the top left of the application to RE Manager.
23:01Cool. So it's gonna do that and okay.
23:05So I see we have like a home dashboard with some metrics. That's looking it's simple but it's it does the job. Right?
23:10We have the contacts which we already looked at before but now we have deals and we have follow ups. So very nice.
23:18Let's start making a deal to track in our real estate manager here, shall we? Alright. We can associate this with a contact.
23:25We can put it on a stage and let's try associating a photo. Here's a picture of a house and I'm gonna create the deal.
23:36Uh-oh. Alright. We have an error.
23:38So let's see what's going on here. What I like to do when there's an error like this is I'll just take a screenshot of it and we're gonna have Claude fix that for us.
23:50So I'm going to paste the screenshot and say, this happened when creating a new deal. I did attach a photo to the new deal on the new deal form before I submitted the form.
24:05So I have a hunch, you know, based on my experience with this type of thing that the issue might have been that since the deal was not actually created yet, and I was trying to attach a photo to it, that might have caused sort of that might have caused some sort of error.
24:21So Claude is gonna figure that out and smooth it out. You know, would that result in the same issue? So if I create the deal okay.
24:30So I was able to create it there and of course it might have already fixed it by now. And I can let's try to attach two photos to it.
24:38Save changes. Okay. So I see the two photos there now.
24:41It might have already fixed it. Of course, was just adding the photos when I was in edit mode. Let's go ahead.
24:48See, I'm sort of just like sprinting ahead here. I know that it's actually still working on it, but I just like to kind of move fast through these kind of final fixes and refinements. I'm gonna try to go ahead and create another deal.
24:59We'll call this deal two. This one's going to cat. And under contract and so again, I'm on a new deal that I have not actually created yet and I'm going to try to attach a photo to it.
25:14I'll do a different photo this time. I'll create that deal. Okay.
25:17So, yeah. While I was testing it, it must have finished its work here, although it's actually still self testing its own or self verifying its own work there.
25:26But I it looks like it has actually fixed it. Very nice. Let's go back to the deals view.
25:31I like how I could see the actual photo on these cards. So that's really cool. Oops.
25:36I kind of slipped over there. Yeah. Alright.
25:38So let's test out the checklist. So I'll do follow-up and I'm gonna add that as an item.
25:44The server's not running anymore. Probably because the agent was testing its own work, which interrupted my server.
25:51Okay. That tends to happen sometimes when I'm kind of multitasking. I'm gonna do a follow-up task.
25:58I'm gonna click add. Okay. So now we have a task here.
26:02It's sort of like to do list. Okay. I see.
26:05So yeah. This was by design. The checklist are sort of just to do list, but then I separately I have follow ups.
26:13So I can do, you know, sign contract and that would be the follow-up which would be due let's say tomorrow. Sign contract that is upcoming.
26:22I can click this button to complete it and I would see that listed here. And now I'm all caught up. So that's looking good.
26:29I wanna do a couple of final things to this app before I consider it like done or ready to go ahead and deploy. I wanna change the colors.
26:38You know, lot of my apps, if you watch my other videos on the channel, you'll notice that I'm using the same design system which has these like default like blue colors play and my app and design system already support dark mode. Yeah.
26:52I'm going to so it did finish up that verification. That's good. Can you suggest a different color palette to swap in?
27:00I'm gonna want you to update the colors as we define them in our design system. Give me a few options to choose from. Okay.
27:08So it looks like Claude actually hooked up some like palette options for me and it created an image and screenshot. So let me just take a look at that so I can see what I like.
27:20I'm gonna go to options and I'm gonna open that in the browser. Look at this.
27:24I didn't even like ask for this but it's just gonna make the whole process of recoloring the site a lot easier if I could just actually see what I like. Do I like a, b, c or d? Of course, I could, you know, tweak it even more if I want to, but on a real estate evergreen, green, that's that feels like the most obvious choice here.
27:44So I'm gonna go with yeah. Let's go with the a option there. Go back here, it used it's ask user question interface here, so I can just go with option a.
27:54Alright. So Claude is gonna make that update and the nice thing is that it's just gonna make the update in the design system here which should then apply everywhere else in the applications.
28:06So let's see how it does with that. Okay. So it did update the design system and I can see here right in the interface that it is all applied.
28:14Now all the accent colors are using the green, our secondary colors are using that to like amber color and if I switch into dark mode, everything's all good.
28:25Very nice. Yeah. This feels more like it would be like a real estate kind of management app.
28:30Now one more thing that I forgot to test earlier which is or it should be done is let's go into our settings, and we have this feature of checklist templates because like real estate deals tend to follow the same checklist or the same process most of the time. So, uh, let's go ahead and create a checklist template and we'll test that out.
28:47Let's say showing and let's say offer contract inspection closing.
28:55I'm sure there there are usually much more so many more steps to typical real estate deals, but that's a good that's a good start. And let's see if we can like reorder these. Yeah.
29:06Very nice. So we can. Alright.
29:07So let's go ahead and create a deal, another deal. We'll say deal three. Very nice.
29:13So we have our closing checklist automatically applied and that template the checklist template worked perfectly.
29:20And we have this nice progress bar that works really well also.
29:27I'm liking it. Okay. So I think I think we've got like a pretty good application for the real estate industry.
29:33Now, I were a real estate agent, I would probably use this and I might even make this a SaaS by, you know, adding features like, you know, the ability to invite other agents and I would add some Stripe Billing to, like, charge by the month for access to this. I would figure out how I wanna price it, you know, maybe price it by the number of properties that I'm managing in here or maybe just a flat monthly fee, something like that.
29:58Plenty of options that you can go with there, but, you know, now we have functioning application. The next step, of course, would be to, you know, get it all deployed and those are things that I cover in more depth in my in my training courses in Builder Methods Pro. But yeah, this this is really good and of course, we can continue to build this out and add more features to it as needed.
30:19So that's a look at how I would go about building a client management tool that's specific to the real estate agents workflow. And I bet that if you're in real estate, you probably know exactly how to make this even more useful and maybe even sellable as a SaaS. Now, you want the exact method to go from blank page to a working app and the thinking behind turning it into a business, that's in my free workshop, VibeCoder to Builder.
30:42You can join me for free by going to buildermethods.com/workshop. And if you've got something out of this, hit subscribe so that you don't miss my next build on the channel. See you next time.
The Hook
The bait, then the rug-pull.
A solo builder spends thirty minutes proving a claim: the most valuable software isn't a clever idea, it's the fix for whatever workflow your own industry duct-tapes together every day. He builds it live — a real estate CRM, from a blank product requirements document to a working, tested app — narrating the exact planning-to-milestone process that makes an AI coding agent trustworthy enough to hand real decisions to.
Named ideas worth stealing.
01:32model
PRD-first milestone build process
- Interview to build the PRD
- Define in-scope vs out-of-scope per feature
- Map the data model
- Break into dependency-ordered milestones
- Plan mode: technical plan per milestone
- Build and self-verify
- Rapid-fire bug-fix pass
- Clear context, repeat with the next milestone's log
The eight-step planning-to-build loop demonstrated across the video, from a one-idea brain dump to a finished, tested feature set.
Steal forany AI-assisted build that needs predictable phases from idea to shipped feature
13:00concept
Milestone log hand-off
Each milestone writes a short milestone-log.md summarizing what was built and any technical decisions made along the way. The next milestone's prompt tells the agent to read the prior logs before starting, so context can be cleared between sessions without losing continuity.
Steal forany multi-session AI build where the context window gets cleared between phases
How they asked for the click.
VERBAL ASK
30:30link
“you want the exact method to go from blank page to a working app and the thinking behind turning it into a business, that's in my free workshop... hit subscribe so that you don't miss my next build”
Soft plug for the free workshop right after the intro, then a fuller close-out CTA (workshop + subscribe) in the final seconds — both tied directly to what was just demonstrated rather than a generic ask.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
