LongCut logo

My Favorite AI Workflow

By Tina Huang

Summary

Topics Covered

  • Hyperspecific Apps Beat Commercial Software
  • Target Workflow Blockers Precisely
  • Cloud Scales, Local Owns Your Data

Full Transcript

Stop before you pay for another subscription. This is my favorite AI workflow right now. It is so powerful and you don't even need to know how to code, although it does help and it is essentially free or at least much much cheaper than paying for most software subscriptions. Let me explain by just

showing you guys first some use cases and then I will break it down step by step with a five-step framework on how to do this AI workflow. Sounds good? A

portion of this video is sponsored by Balt.

Use case number one, which was the bane of my existence, accounting. If you

yourself are a small business owner or know somebody that is a small business owner, you may understand what I mean when I say that accounting is the bane of my existence. Stuff like bookkeeping, filing taxes, payroll, filling out a million forms about a million things. It causes me physical pain just thinking

about it, especially because I do a lot of crossber work, meaning I hire people from all over the world and we also have clients from all over the world. So I have this very specific kind of setup that is not optimal for most commercial accounting software except now because

we actually built our own hypersp specific AI app to do accounting for us.

It is so custom this company. It's able to deal with specific regulations from specific countries. It can interpret different kinds of currencies and it's

specific countries. It can interpret different kinds of currencies and it's able to do bookkeeping and forms and things like that that abide by the laws of multiple different countries as well. We're also using open source AI and hosted locally, so it's not leaking things to the internet and it's free no

matter how many documents, receipts, and invoices, whatever that I want to feed through it. Commercial software, by the way, usually costs around like $15 to

through it. Commercial software, by the way, usually costs around like $15 to $30 per month. Use case number two, making custom slide decks. If you've

been to any of my live streams or have done any of our courses or workshops, you would have seen this very specific style of slides that we use. While

previously it would have taken ages to make slide decks for every single presentation, now we built this super custom slide deck maker where we can feed in the information and have it just come out with these very beautiful slides. So streamlined and custom to exactly what we want and also free. Use

slides. So streamlined and custom to exactly what we want and also free. Use

case number three. Another example, a couple months back for whatever reason I decided that I wanted to make a manga. But the problem is that I don't know how to draw. Like this is my drawing level. seeing as I did not have the patience,

to draw. Like this is my drawing level. seeing as I did not have the patience, unfortunately, to spend months and months drawing, which by the way, totally cool if that is what you want to do. But for me, it was kind of like a bucket list thing where I just wanted to make a manga. So, I made, you guessed

it, a hyperspecific custom AI app that was able to make an entire manga just based upon text prompts and my terrible drawings. I have so many more examples.

Let's see, what else have I made? Here's another custom email AI agent that's able to screen my emails and pre-draft them for me. And about a year ago, I also made another one that is able to handle like my seven calendars and tell me what I'm supposed to do every day. Yes, I have so many other examples, but

I think you get my point. So, these are examples of my favorite AI workflow right now, which I don't really know what the name is for this, so I shall call it hypersp specific apps. Definition of hyperspecific app is when

you use AI assisted coding to build software for very specific personal or business use cases that would never exist as commercial software. Here are

the pros. I think the biggest pro of building a hypersp specific app is that you're able to build custom features that are so specific to your use case that it's impossible for you to actually find from commercial software. Another

very attractive pro is that it is essentially free minus the building and setup cost especially if you use open source models. Like on average these apps cost me around $10 to $20 to build. And depending on where I'm hosting these apps, like where these apps live, you can run them again and again without

incurring additional costs, which I will actually talk about like hosting and stuff like that a little bit later in the video. But yes, for now, just realize that at the very minimum, it's going to be a lot cheaper than buying like a monthly subscription for most software. With building your own apps, you also do have full control over your software. So, you don't need to be

worried about like where it's sending its data, what's happening to it, things like that. And finally, you don't need to write code yourself to build

like that. And finally, you don't need to write code yourself to build something like this. Don't get me wrong, I think if you're interested in learning how to code, you absolutely should still do it because there is a ceiling to what you can actually build without knowing how to code. And overall, it would also save you a lot more money if you knew how to code, but you don't need to know

how to code when you're starting out building your hyperspecific apps, which is pretty cool. I'll explain the process in a bit. But first, I do want to mention some cons as well, some caveats. The first one is that it does take time.

On average, it takes me about 2 to 3 hours up to a day to build the first version of these hyperspecific apps. But there is a little bit of a learning curve. So when I was just starting out or if you're someone who has no idea how

curve. So when I was just starting out or if you're someone who has no idea how to build anything, zero engineering background, no coding skills, it might take you a little bit longer, like maybe a few days to a week to get started properly. So some time and effort, yes. Also, it could require a little bit of

properly. So some time and effort, yes. Also, it could require a little bit of setup and building cost. Specifically, if you are going to be using AI assisted coding, you're probably going to need to pay for a subscription to use a coding agent of your choosing. And if you want your app to be able to run continuously,

you might have to pay for some type of cloud service or a local machine that will continuously run. Although, if you're okay using like open source coding models that are not like as optimized for coding, um, or you're okay with things like maybe not running all the time, you can technically also do

this for free. All right, great. Let me now explain the five-step process for how to do this AI workflow to build a hyperspecific app. Step number one is to identify which workflow you want to build a hyperspecific app for. There's

like three categories of things that my hyperspecific apps usually fall under.

So, hopefully this is also helpful for you to come up with what you want to build. The first one are things that are the bane of your existence but

build. The first one are things that are the bane of your existence but unfortunately to exist you must actually do it. Biggest example for me was accounting. I will put on screen now some other examples in this category.

accounting. I will put on screen now some other examples in this category.

The second is things that you've been procrastinating on because it's like you know it's important but you just don't want to do it cuz it's a lot of work and you just don't feel like doing it but you know if you did it it would actually be much better for you. An example for me is tracking my exercise, my food, and

my sleep. things that I know that I should do but it's just like effort you

my sleep. things that I know that I should do but it's just like effort you know I will put on screen some other examples in this category and the third category is something that you want to do but you can't or haven't because you lack the skill or you don't have the time and you just like maybe don't have

the resources or energy to to dedicate to doing it. This could be like a fun hobby project. For example, for me it was making a manga except it was

hobby project. For example, for me it was making a manga except it was impossible because I'm so bad at drawing. And I know for others like many people want to build like a software or a game, but they just like don't have the skills and time to do that. I'm going to put on screen now some other

examples in this category. And I do think this is like a really interesting category because it literally enables you to do something that you simply could not do before.

That's pretty cool. All right, I hope this helps you decide what you may want to build a hyperspecific app on. And do put in the comments what you want to build. I genuinely really want to know what you guys are interested in

build. I genuinely really want to know what you guys are interested in building. Step number two is to map out the process workflow as in like map out

building. Step number two is to map out the process workflow as in like map out the current existing workflow. This step can take a lot of time or very little time depending on how well you understand the process. Let me explain.

For example, for the accounting stuff that I wanted to build this hyperspecific app for. A core feature would be bookkeeping. And this is something that I already understand the full workflow of because I've been doing it for like 5 years. Every month I needed to one collect all the receipts

and invoices. two, match them to all of my corporate credit cards and also my

and invoices. two, match them to all of my corporate credit cards and also my bank statements. And then three, have a final deliverable as a spreadsheet with

bank statements. And then three, have a final deliverable as a spreadsheet with all of the entries as transactions and the supporting documents for each transaction. And that would be the map that process didn't take me a lot of

transaction. And that would be the map that process didn't take me a lot of time to figure this out because I had already done it so many times. But for

mapping out the workflow for generating manga, for example, I literally had no idea the process of making manga. So, in order to map out this workflow, I actually have to go take an entire course from an expert to map out the process of what it takes to make a manga, which in case you're curious, is

this. So, after you map out your workflow, we can then move on to step

this. So, after you map out your workflow, we can then move on to step three, which is identifying where your hyperspecific app would fit in in that workflow. As amazing as technology and software and AI is these days, it's

workflow. As amazing as technology and software and AI is these days, it's still not possible to just like take your entire workflow that you just mapped out and stick it into AI and just tell it to like build the entire thing, unfortunately. And also, maybe you don't actually want that, right? like part of

unfortunately. And also, maybe you don't actually want that, right? like part of the process, you might want it to remain manual or there's no need for you to do that. So that's why we do want to explicitly go in there and identify

that. So that's why we do want to explicitly go in there and identify which parts of the workflow we want to be part of the hypersp specific app.

Let's use the manga example here. So here's the workflow for making manga which I learned from taking that course from a professional manga. I learned

that there were nine steps to making a manga. One was to pick a topic including central values and central themes that I wanted to address. Two was to design the characters. Three was to have a major plot synopsis like an explanation for

characters. Three was to have a major plot synopsis like an explanation for what the manga should be about. Four is to write out the entire story and split them into chapters and scenes. Five is then to transform these scenes and

chapters into panel descriptions like what each little panel drawing should be portraying. Six is to actually sketch out these scenes and draw them and then

portraying. Six is to actually sketch out these scenes and draw them and then design the panels as well like how the panels are all supposed to fit together and how the pages are supposed to be divided. I was really surprised by how much like thought and detail and process went into just constructing how the

panels fit together. And seven is to fill in the details of these sketches like actually draw them nicely. Eight is to then finally bind them all together into one whole manga entity. And nine is to go publish it. So that is the entire workflow, but I don't actually want to include this entire workflow in the

hypersp specific app itself because steps one to five actually involve a lot of writing and I actually really like writing and I think it's really fun.

It's also where a lot of creativity comes in. So I don't actually want to just be like autogenerated for me. I might like use AI to help me, you know, flush out some stuff or like bounce ideas and stuff like that, but I do want to keep this process manual. Now, what I do need a lot of help with and what I do

want this hyperspecific app to be able to do is steps six, seven, and eight, cuz that's like the visual parts, like sketching out the scenes, designing the panels, filling the details, and then binding them all together. This is like the blocker for me. It is impossible for me to do this because I don't know how

to draw. So, that's why I want my hyperspecific app to be able to take

to draw. So, that's why I want my hyperspecific app to be able to take care of these steps for me. And finally, step number nine, which is to publish. I

don't actually care that much about that because I'm just doing it for my own like bucket list enjoyment. Amazing. Great. You now know what you want your hyperspecific app to do. We can now move on to the next step, which is actually building your hyperspecific app using good AI coding practices. I've actually

made an entire video on how to use AI coding in order to build apps both using code and no code tools. So, I do recommend checking that out. I'll link

it over here. I also made videos where I show you guys me building these apps and I'll link them in the description as well. So do check those out when you get serious on building your app. But I do want to give you a very fast quick start guide in this portion of the video too. So basically three steps. You want to take what you want your hyperspecific app to do and then change that into what

is called a product requirement document. This is a standard document for defining the purpose, features, functionalities, and the behaviors of a product. This is what you're going to be feeding into your AI coding tool. So it

product. This is what you're going to be feeding into your AI coding tool. So it

knows what you're supposed to be building. I'm going to actually link a prompted description which you can just copy paste into like any chatbot of your choosing whether that be like Chachi BT Claude Grock Gemini whatever you want and it would actually like ask you questions to help you come up with this

PRD. See I got you. All right. So after you have your PD you then want to pick

PRD. See I got you. All right. So after you have your PD you then want to pick your AI coding tool to build your hyperspecific app. In the AI coding fundamentals video I literally like have an entire section dedicated to explaining tools. So please go to that video for a lot more details. But

explaining tools. So please go to that video for a lot more details. But

basically there are a range of different AI coding tools that range from completely no code to for pro developers and those are good for specific types of use cases as well. For me personally because this is literally my job I have

tested pretty much the full spectrum like all of the different AI coding tools out there and I would use different AI coding tools for different purposes. Say for example, if I know that this hypersp specific app I'm

purposes. Say for example, if I know that this hypersp specific app I'm trying to build is going to be like a big complex project with lots of technical details, maybe even like pretty obscure technical things included in it and at some point I might even want to bring in other developers to help me with it. Then I would go directly with a codebased prodeveloper

tool like the accounting software for example, I primarily use cloud code for that. And when the project became bigger and I wanted to do more parallelization

that. And when the project became bigger and I wanted to do more parallelization like having multiple coding agents working at the same time and also like working with other developers, I switched over to warp which is really good for that kind of stuff. but for the hypersp specific apps on making slides

and also for the manga generator. These are web- based well scope projects that have pretty standard integration needs. So I actually built these using bulk who I'm actually really excited to say is the sponsor of this portion of the video. I really like their product as a web-based option. I've been using them

video. I really like their product as a web-based option. I've been using them for over a year and a half now. So thank you so much for reaching out to sponsor this portion of the video. Vault Vault is my go-to for creating really fast AI powered software and fully functional web apps. I built this generator

and the slide generator using Vault. It lets you go from an idea to a fully deployed productionready app just by describing what you want in plain English. It's crazy cuz it's not just the prototype. It's like an actual

English. It's crazy cuz it's not just the prototype. It's like an actual working app with a real backend and running entirely on your browser. They

also dropped two new features that I think will be really relevant for you if you want to build hypersp specific apps. The first one is image generation right inside Vault. So if you need a thumbnail, a mockup, or any visual

inside Vault. So if you need a thumbnail, a mockup, or any visual assets for whatever it is that you're building, you don't have to jump between tools anymore. You can just generate it directly within the platform itself. The

tools anymore. You can just generate it directly within the platform itself. The

second is connectors. And this is what I'm most excited about. Vault now

directly plugs into platforms like GitHub, Notion, Linear, Jira, and many more. So your app is able to pull from real data, not just placeholder content.

more. So your app is able to pull from real data, not just placeholder content.

Now that is what it takes from a prototype to something that you actually can ship. Over 7 million people are using Bault now, including teams from

can ship. Over 7 million people are using Bault now, including teams from Fortune 500 companies. It is free to start and no credit card required. Link

is in the description. Thank you so much B for sponsoring this portion of the video. Now back to the video. Finally, to round out my brief guide to AI coding

video. Now back to the video. Finally, to round out my brief guide to AI coding tools, I want to say that there are free options as well if cost is a really limiting factor to you. You can use open source coding models. Just know that the coding experience itself is not going to be as optimized and you need to be more

careful about things like version control and and privacy and security things. So, I would only recommend this option to people who do have a little

things. So, I would only recommend this option to people who do have a little bit of a coding experience already and a lot more patience. I'll put on screen some tools and models to check out as well here. Okay, great. So, now you've picked your AI coding tool. You want to take your PRD, remember your PR from

earlier, and stick that into your AI coding tool to get started on the process of building your hyperspecific app. Sorry for sounding like a broken record here, but in the AI coding fundamentals video, I go into a lot more detail about how to go through this building process. So please do check out

that video. The AI coding principles I presented in that video I have dubbed

that video. The AI coding principles I presented in that video I have dubbed the tiny ferrets carry dangerous code framework which stands for thinking as in always be thinking and rethinking deeply about the features that you're

building. You must not throw your PRD at your AI coding tool and just run away.

building. You must not throw your PRD at your AI coding tool and just run away.

Second is to focus on existing frameworks. You don't need to reinvent the wheel as you're building your app. There are most likely existing frameworks that other developers have already built. So instead of trying to do something from scratch, help your AI agent provide relevant documentation and

frameworks to your AI tool so it can draw on the knowledge of past work. C

stands for checkpoints. Create checkpoints as you're building your app.

Thinking about version control using Git, GitHub. D stands for debugging.

Debug with care. Debug as you build. And C stands for context. Provide as much context as possible to your AI coding tool in the form of screenshots, examples, more detailed descriptions. The more context and details that you

can provide your AI, the better able it would be able to build what you have in mind. As a bonus tip, if the hypersp specific AI app that you're trying to

mind. As a bonus tip, if the hypersp specific AI app that you're trying to build, you also want to include an AI component to it. I know, right? Very

meta here. You're using AI to build an app that has AI in it. Think carefully

about what kind of model you want to use. I'm going to link in the description a couple of videos that I've done that may be able to help you decide which models to use. Wow, great. If you follow these steps by now, you should have the first version of your hypersp specific app. The first time that you do

this, probably going to take you a little bit longer, especially if you've never done AI coding or build apps before. But as you build more and more of these, I promise you, it's going to become so much easier and faster. These

days, depending on how complex the app I'm trying to build is, I can usually get the first version out in maybe like 2 to 3 hours up to like a day. Just

saying. Prior to AI coding tools, the development of really any type of application would have taken you like months to do and there's no way you could do this if you didn't know how to code obviously. But now you can and you can do it in like a few hours. That is pretty magical. All right, time for the

last step of this framework in developing your hyperspecific app and that is hosting. Hosting is providing the infrastructure of where your app lives and runs so you and other people can access it. Roughly speaking, there

are three different categories of hosting options. The first one is cloud options. This is when you can rent compute and you can dump this code like

options. This is when you can rent compute and you can dump this code like the application that you build and have it run on these cloud services and like get other people to run it for you. You pay for how much compute like how much running of this application that you use, it can scale better. So you can

have one user using it with 10,000 people using it. And there's also other tons of managed services that you can use as well like databases or queries.

as an analogy, sort of like renting a hotel room. You're paying per night or hourly sometimes. There's like managed services like a concierge. There's like

hourly sometimes. There's like managed services like a concierge. There's like

a mini fridge and you can like invite lots of other people over and you can like leave anytime you want as well. You don't have to like clean up after yourself. This might be the best option for you. If you're hypersp specific app

yourself. This might be the best option for you. If you're hypersp specific app that you're building, you can foresee that there's going to be unpredictable traffic. There might be like a lot of people using it at the same time and

traffic. There might be like a lot of people using it at the same time and then suddenly like nobody using it for a while or if it involves like other team members coming in and also like tweaking the app and doing stuff to it as well.

and anything that involves having to scale and you don't really want to deal with the complexities of that. The downside of this is that it can get very expensive because it's like a hotel room, right? You get like all these like fancy things as a premium option here. You may be locked in to like a specific

vendor and their services and it's not private. What happens in the hotel doesn't stay in the hotel potentially. Some examples of cloud options would be like AWS, Google Cloud, Azour. I'm going to put on screen now some other examples of the cloud option as well. Now, the second option you can consider when

you're hosting your hypersp specific app is a VPS, a virtual private server. This

is a fixed virtual machine that you're renting out 24/7 for a flat fee. It's

kind of like renting out an apartment. You have a fixed monthly amount that you're renting it for, and during that time, it's your responsibility to keep it tidy and not disturb your neighbors. And you can also like decorate it and do what you want to it within reason. This is good if you're going to be building a

bunch of different hypersp specific apps that are all like relatively low traffic as in you're going to be the only person using the app uh or maybe just like a couple other people. It's customizable. You manage it yourself and you have a little bit more privacy as well cuz what you store inside like data for example

is a little bit more private. Do want to make a caveat here that's not completely private though because technically your virtual machine is still physically sitting on somebody else's data server, right? like the hardware still belongs to somebody else. So, if they like really really really wanted to go take a

look like your landlord if they really really really want to come in and look at your apartment, they can't. They still have the keys. Like they probably won't. Uh but they could. So, yes, that is one of the cons. And you do need to

won't. Uh but they could. So, yes, that is one of the cons. And you do need to manage things yourself. You got to clean up after yourself. And even if you don't have the app running, you still have to pay for that rental fee. So, I think this is more worth it if you do have multiple apps that are going to be running or you are actually going to be like constantly using the apps that you

build. Some examples of VPS providers include Hostinger, Hezner, Digital

build. Some examples of VPS providers include Hostinger, Hezner, Digital Ocean. Put some other examples on screen now. Now, finally, the third option is

Ocean. Put some other examples on screen now. Now, finally, the third option is hosting your hyperspecific app on your own hardware. This is a physical machine that you would just buy one time and it would just like run on your desk.

Continuing with our property analogy, this is like truly owning a home. The

pro is like you buy it once, it's truly yours. You really have like complete privacy. Nothing is going to leave your local machine and you exactly know what

privacy. Nothing is going to leave your local machine and you exactly know what you want to do with it. Fully customizable to whatever you want it to do. And after you buy it one time, the only thing that you really need to pay

do. And after you buy it one time, the only thing that you really need to pay for is like just the electricity for running your hardware. But just like owning a home, there are also downsides because you own everything. You also are responsible for everything. If something breaks, you got to fix it yourself

completely. Something needs updates, got to do it yourself. Your home internet

completely. Something needs updates, got to do it yourself. Your home internet could also be a bottleneck if it's not fast enough. physical failures,

downtime, things like that. This technically could also be the cheapest option for you if your laptop is relatively modern because you can actually use your laptop to be running your app. But the issue is that if you're like doing other stuff simultaneously, it might make your app

really slow or just like completely shuts down. Uh or if you close your laptop, then it would stop working. So that's why people often choose to buy separate hardware to run their apps. Like that machine is going to be dedicated to running your hyperspecific app and it's going to be running 24/7.

The very popular option these days is to buy a Mac Mini for this, but I'm going to put on screen now some other options that are available as well. Cool. Now

that you have hosting done, you officially have a properly functioning hyperspecific app. Isn't that cool? I'm genuinely so excited to have been able

hyperspecific app. Isn't that cool? I'm genuinely so excited to have been able to share this this workflow with you guys. I use it so much and it really is my favorite AI workflow right now. I hope you're going to be building your own hyperspecific app soon, too, if you haven't already. Let me know in the

comments what you want to build. Now, thank you so much for watching until the end of this video and I will see you guys in the next video or live stream.

Loading...

Loading video analysis...