I built a $30k app with AI. Here's what I learned

Also: Join me on a live workshop in March.

TL;DR: Anyone can learn how to ship these projects without coding. To prove it I’m launching a paid live workshop in March over Zoom. Pre-sale is open now for the first 30 people, details at the end of the email. If you don’t want to wait for that, click here to sign up for the paid workshop.

---

The first time I met Pia was in mid-August. I wrote extensively about her problems and story which I shared in this post:

Storytelling at scale

Pia is an online coach who helps brand agency owners grow. Her clients’ stories are her stories, and businesses like hers are very human.

Marketing a coaching or consulting like Pia’s business is very straightforward:

Tell the stories of your best clients to find more of them.

But the execution is complicated. You need to manually sift through all interactions with your clients throughout your work together and craft compelling stories, proof of their results, and then tell those stories in a compelling way.

So Pia hired me to build an AI solution that could do just that and High Signal was born.

Take a bunch of high-noise data (Circle posts, comments, call transcripts, CRM notes, lead forms, onboarding memos) and turn it into high-signal content in three steps:

High Noise → High Signal → Content

First, I used carefully structured cognitive workflows to transform the raw data into high-signal content, which is meaningful from a business perspective.

For example, a list of every single comment and post made by John Doe is high-noise data.

But a one-paragraph description like this is different. It’s high-signal:

John Doe started working with Pia on August 17th because he felt like his branding agency was stuck at a level but demanding long hours burning him out. He felt pressure to grow but at his wits end from managing his existing business. John joined the No BS Agencies program and by September 2nd he scored his first $25k package. He completely restructured his client work and by November 21st he was already planning his vacation: "For the first time in five years, I'm going on a holiday this year!"

In one paragraph I’m telling the whole story of John Doe and all I needed to do was meticulously read through every single comment he made on Circle in the last six months.

This would’ve been exhausting, soul-crunching work before, but using LLMs helped me succeed with hundreds of students across tens of thousands of posts — without hallucination or manual work.

The first half of this project was spent on making sure this works.

By the time the system was up, every client of Pia’s got their own personalized story written specifically about their journey. The story is structured following the “Hero’s Journey”.

But they didn’t just get a story. We also created a structured timeline of their major milestones and a bunch of usable quotes. All High Signal content that will make marketing content creation a breeze.

High Signal then uses these story assets to extract content ideas from it.

You see we’re not using AI to generate new ideas or new content. The content is already in the data. The ideas, the transformation journey, and the AHA moments are all in the data already. It’s just noisy.

We’re using AI to extract information not to create it.

Because of this, the architecture was built so that hallucination is rarely a problem.

Now we can feed the high signal, low noise data into a pre-defined content format, like:

• Testimonials

• Case studies

• Reel videos

• Carousels

All we need is a bunch of templates so we ended up creating a few of those. Focusing on the personal branding vibe, we ended up creating templates that generate simple, personalized, high-impact posts via an intuitive interface.

AI generates a first draft and if you don’t like it, you can just regenerate the content piece on demand.

But we didn’t stop there. We also plugged it into Metricool, which is their social media management app of choice. Now you can also schedule and post your content with a single click on selected platforms:

The project was demoed yesterday and now we’re moving on to a 12-month Maintenance Period where we’ll be developing more templates and more functionality, fine-tuning the prompts to make sure they generate the kind of content we need, and also exploring options to commercialize the app, and make it available for others too.

Unit Economics

Alright, let’s answer the most important part.

Is the content it generates good? For the final content, I’d say it’s 80% there. For a first draft that you can edit and finalize in under a minute, it’s better than anything you’ve ever seen.

How much time will it save? The time required to generate high-quality content is reduced by over 90%. All the data collecting, extracting, reformatting tasks are done by AI and the final edit is done by a human.

How much does it cost to generate one piece of content? About $0.07. I haven’t yet plugged in o3-mini or DeepSeek R1 which will lower the costs even more (I was using o1-preview before which is significantly more expensive).

Technical Details

This app has five main components:

Front End: Hosted on Vercel. (Fully built with Lovable, I didn’t code one bit.)

Back End: Supabase handles all Edge Functions and cron jobs. (Fully built with Lovable)

Database: Supabase runs the entire database. (Fully built with Lovable, I didn’t code one bit.)

Cognitive Workflows: Built on a self-hosted n8n instance.

Content Generation: I generated the code with Cursor’s built-in assistant using o1-preview and then hired some help to finish it on Upwork because the deadline was near and I was in a rush. This is a standalone microservice currently running on Railway. I started working on it before using Lovable but I if I started with Lovable, I could’ve easily built this and avoided paying the Upwork freelancer.

Here’s the important bit: 99.99% of the code was written by AI.

All I had to do was plan it and then prompt Lovable accordingly. I don’t know how to code properly, I can even barely write functional code let alone nice.

Numbers:

The project took a very long time because I had a lot going on last year and I also had a bunch of dead ends. I was originally using JetAdmin and Make and I made the mistake of not thinking through beforehand if those platforms could handle this.

So 10 days ago I decided to go all in on Lovable and rebuild it from scratch there.

Here’s the end result after 10 days:

• 24,381 lines of code written by AI

• 0 lines of code written by me

• 647 GitHub commits

• 3 functional n8n workflows

• 561 Lovable messages used

Now let’s see a summary of all the costs of building this app:

• Lovable messages: $112

• OpenAI tokens burned in testing: ~$100

• Digital Ocean for n8n hosting: $6

• Railway: $5

• Vercel: free

• Help from Upwork: $650

Looking at all the git commit logs I spent ~48.6 hours building this, producing ~500 lines of functional working code per hour. That’s impressive, considering it’s 5-15x faster than what a developer could do and considering that I’m not even a coder.

Disclaimer: This approach wouldn’t work if this was a larger project. If we were collaborating on the code or working in the enterprise, you couldn’t raw dog it like this. But for a sub $100k project, it’s doable.

If I wanted to count all the time I was working on the architecture and planning stage, designing all the cognitive workflows and data cleaning, that’d probably add up to 200+ hours easily. Some of that was legitimate work, some was just trying to figure out how to access data and producing documentation. But I would need to do that anyway even if I just hired developers to build the solution.

Key Learnings

#1 Make sure you have the right data

One of the biggest rabbit holes was the quality of the available data. I needed to scour through dozens of database tables, and thousands of rows, check data formatting issues, and clean up.

Just like this client, most clients who need no-code services don’t have a full-time developer. This means that a non-technical person is managing the database, which usually means that it’s fragmented.

I severely underestimated how much time I’d need to finish with data cleaning. Even though the client is supposed to provide the data in the format I need, they can only do so much if there’s tech debt in their database.

I spent anywhere between 50-100 hours just cleaning data. Ideally, this is either done by the client or you just price it in as an additional cost layer.

#2 Don’t rush into execution

This echoes the problems I see surfacing on the Lovable Discord.

If you plan out your software’s architecture, database, and dependencies beforehand, you’ll have a good time building things with Lovable. But if you just rush into the building, you’ll loop yourself into problems because neither you nor the AI will have any priorities or any actual idea what a good result would look like.

Here are my recommendations for that:

• Always start with User Stories and go backward. Write things down like “As a user, I want to…”.

• Think about how you’ll handle data, draw on the whiteboard, or experiment with Excel if you need to.

• Follow SOLID principles, even if you’re not writing code yourself.

• Build your app as a series of microservices. Your app is a bunch of small apps talking to each other.

• Follow TDD even if you’re not writing the code yourself. Any microservice you build is going to be a set of small actions. If you know how to test whether those actions work one by one, you can also use Lovable to write tests for that. That means you first create the test, and then you try to build code with Lovable that will pass the test. This way when something fails, you’ll know where to look.

These are a lot of ideas, concepts, and frameworks to digest all at once.

But as you can see if you want to learn how to build apps from scratch — thanks to apps like Lovable — you don’t need to finish that Python course anymore that you signed up for in 2022 and is on 17% completion.

You haven’t finished it in 3 years anyway and now you don’t have to.

All you need to do is think better, in a more structured, more efficient way about the building blocks of software, and the rest is carried by AI.

#3 You can make a living out of this

If you’re planning to build your app or thinking of a career change but becoming a software engineer seems daunting, you now have another option.

You can build small internal tools and MVPs for others if you learn how to think about software in a more structured way and use tools like Lovable.

You’ll enter a new, exciting world and you’ll discover tools engineers have been using for years. But you won’t code at all.

To see is that’s profitable, let’s take a look at the High Signal project:

The going rate for no-code developers is about $75 per hour on Upwork. You’ll find solutions for significantly cheaper or more expensive, but this is a good rate to go for.

If I don’t count the architecture work, but add my ~50 hours of work to this High Signal cost me $4518 to build.

If I add the additional 200 hours of my architecture work at the same hourly rate, that goes up to $19,518, which gives me a comfortable 35% profit. Not to mention that 96% of the cost is my time.

Join the Weekend Developer Live Sesh

I know what you’re thinking. That this is really great but it’s really complicated and you couldn’t do it yourself. Well, that might’ve been true a year ago, but not anymore.

Anyone can build SaaS apps if you know just a few basic principles and know how to use these tools.

The code of tomorrow is not written by humans.

I ran two experiments for free courses and over 600 people signed up for them in a week which is mind-blowing. After talking to many of you I decided to offer a third option:

A 2 x 120 minute LIVE Building Session, where we take you from ZERO TECH SKILLS to having a prototype of your app.

I call this the Weekend Developer Live Sesh and it’s a two-part paid Zoom workshop, pre-sale opens at $97 per person (only 30 seats available, then the price goes up to $147).

Here’s what we’ll cover. Not too deeply, just enough so you can get from 0 to 1:

• Fundamentals: Creating user stories, database structures, software architecture, integrations on paper.

• No-Code Tools: Get familiar with Zapier, Make, n8n and learn when to use what.

• AI-assisted App building: Learning how to use Lovable, Supabase, Vercel, Github and other tools as a non-coder.

• Commercial basics: Using third party APIs, taking payments from your users, handling multiple teams, etc.

Once you join, I will send you an invite to the Lumberjack Discord Server and you will be added to the #weekend channel. This is a private channel only available for live participants.

As I always say, knowledge is free, time is money. This 4 hour bootcamp will have a full cost of $150 at launch and opening pre-sale at $97. I will run this pre-sale for 2 weeks or until we have the 30 buyers, then the price goes up to $147.

The workshop will take place in March. If you buy in the presale, you’ll have the opportunity to ask for specific dates.

The sessions are fully recorded and you’ll get access to the recordings including a handbook afterwards.

---

If you don’t want to pay for this, it’s okay.

Knowledge is free, time is money.

I’ll still release the free email course later this year, you can sign up on this link.